SD.Next

SD. Next is app based on AUTOMATIC1111 stable diffusion webui

See Installation example and Test 1 - 46 on https://wiki.liutyi.info/display/DEVOPS/SD.Next

Test 26 - Create Ai book logo
SD.Next supported models
SD.Next on Linux

SD.Next openvino docker compose
SD.Next btrfs deduplication example
Edit SD.Next task que using sqlite3
SD.Next by liutyi docker compose

Test 47 - Allo theatre play cover

Test 26 - Create Ai book logo

Artificial intelligence

Try different Ai models to generate book cover for bookstack site

Model/ Prompt (Size)	shuttle-3.1-aesthetic [6e1883042a]	FLUX.1-dev [3de623fc3c]	HiDream-I1-Full [fe6156b63d]
Artificial intelligence (440x248)	Prompt: Artificial intelegence Parameters: Pipeline: FluxPipeline\| Steps: 20\| Size: 440x248\| Seed: 77718796\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: shuttle-3.1-aesthetic Time: 2m 27.73s \| total 231.52 pipeline 125.49 preview 49.76 move 12.63 prompt 12.62 te 12.59 decode 9.61 offload 8.82 \| GPU 32220 MB 25% \| RAM 25.51 GB 20%	Prompt: Artificial intelligence Parameters: Pipeline: FluxPipeline\| Steps: 20\| Size: 440x248\| Seed: 1998877164\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: FLUX.1-dev Time: 2m 35.03s \| total 253.05 pipeline 126.53 preview 58.23 decode 16.36 offload 15.57 move 12.13 prompt 12.12 te 12.08 \| GPU 32290 MB 25% \| RAM 25.02 GB 20%	Prompt: Artificial intelligence Parameters: Pipeline: HiDreamImagePipeline\| Steps: 20\| Size: 440x248\| Seed: 3499019024\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: HiDream-I1-Full Time: 17m 25.07s \| total 1170.85 pipeline 980.22 move 48.86 prompt 47.79 te 45.14 preview 20.11 decode 15.95 offload 12.73 \| GPU 39228 MB 31% \| RAM 58.95 GB 47% (Result image is 1360x768)
	OmniGen2 [453419871d]	juggernautXL [33e58e8668]	juggernaut_reborn [338b85bc4f]
Artificial intelligence (440x248)	Prompt: Artificial intelligence Parameters: Pipeline: OmniGen2Pipeline\| Steps: 20\| Size: 440x248\| Seed: 2523245077\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: OmniGen2 Time: 1m 6.42s \| total 66.44 pipeline 66.39 \| GPU 15824 MB 12% \| RAM 3.5 GB 3%	Prompt: Artificial intelligence Parameters: Pipeline: StableDiffusionXLPipeline\| Steps: 20\| Size: 440x248\| Seed: 177179858\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: juggernautXL_juggXIByRundiffusion\| Model hash: 33e58e8668 Time: 18.20s \| total 28.38 pipeline 14.98 preview 9.86 move 1.71 prompt 1.10 decode 0.58 \| GPU 7336 MB 6% \| RAM 3.66 GB 3%	Prompt: Artificial intelligence Parameters: Pipeline: StableDiffusionPipeline\| Steps: 60\| Size: 440x248\| Seed: 2954274510\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: juggernaut_reborn\| Model hash: 338b85bc4f Time: 29.25s \| total 31.76 pipeline 28.67 preview 2.40 decode 0.47 \| GPU 2448 MB 2% \| RAM 3.65 GB 3%
	Chroma [c98e7c057c]	Cosmos-Predict2-2B-Text2Image [015332720f]	stable-diffusion-3.5-large [ceddf0a7fd]
Artificial intelligence (440x248) (880x496) (880x496)	Prompt: Artificial intelligence Parameters: Pipeline: ChromaPipeline\| Steps: 40\| Size: 440x248\| Seed: 3631163972\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: Chroma Time: 6m 29.28s \| total 568.84 pipeline 380.49 preview 167.66 decode 5.11 offload 4.59 move 3.67 prompt 3.66 te 3.64 \| GPU 26474 MB 21% \| RAM 19.4 GB 15%	Prompt: Artificial intelligence Parameters: Pipeline: Cosmos2TextToImagePipeline\| Steps: 24\| Size: 880x496\| Seed: 2752054978\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: Cosmos-Predict2-2B-Text2Image Time: 2m 38.78s \| total 170.21 pipeline 158.76 te 11.43 \| GPU 9802 MB 8% \| RAM 15.96 GB 13%	Prompt: Artificial intelligence Parameters: Pipeline: StableDiffusion3Pipeline\| Steps: 24\| Size: 880x496\| Seed: 2224330216\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: stable-diffusion-3.5-large Time: 6m 18.68s \| total 385.22 pipeline 370.44 move 6.55 prompt 5.93 decode 2.29 \| GPU 16764 MB 13% \| RAM 28.94 GB 23%
	stable-cascade [a89f66d459]	Cosmos-Predict2-14B-Text2Image [015332720f]	F-Lite [2dae1306bd]
Artificial intelligence (880x496)	Prompt: Artificial intelligence Parameters: Pipeline: StableCascadeCombinedPipeline\| Steps: 20\| Size: 880x496\| Seed: 2709937372\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: stable-cascade Time: 1m 8.60s \| total 68.76 pipeline 68.52 \| GPU 12382 MB 10% \| RAM 3.04 GB 2% (Result image is 896x512)	Prompt: Artificial intelligence Parameters: Pipeline: Cosmos2TextToImagePipeline\| Steps: 24\| Size: 880x496\| Seed: 2413833806\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: Cosmos-Predict2-14B-Text2Image Time: 13m 40.24s \| total 832.43 pipeline 820.22 te 12.19 \| GPU 27678 MB 22% \| RAM 38.84 GB 31%	Prompt: Artificial intelligence Parameters: Pipeline: FLitePipeline\| Steps: 40\| Size: 880x496\| Seed: 1332699729\| CFG scale: 6\| App: SD.Next\| Version: 4e449ca\| Operations: txt2img\| Model: F-Lite Time: 10m 2.56s \| total 616.40 pipeline 597.66 offload 6.64 te 5.45 decode 4.88 gc 1.75 \| GPU 29710 MB 23% \| RAM 21.04 GB 17%

System Info

app: sdnext.git updated: 2025-07-12 hash: 4e449ca9 url: https://github.com/vladmandic/sdnext.git/tree/dev
arch: x86_64 cpu: x86_64 system: Linux release: 6.11.0-29-generic
python: 3.12.3 Torch 2.7.1+xpu 
device: Intel(R) Arc(TM) Graphics (1) ipex: 
ram: free:113.3 used:12.03 total:125.33
xformers: diffusers: 0.35.0.dev0 transformers: 4.53.2
active: xpu dtype: torch.bfloat16 vae: torch.bfloat16 unet: torch.bfloat16
base: Diffusers/nvidia/Cosmos-Predict2-14B-Text2Image [015332720f] refiner: none vae: none te: none unet: none

SD.Next supported models

https://vladmandic.github.io/sdnext-docs/Models/

List of Models by Size

Created on 2025-07-15. Last update 2025-08-08

Publisher	Model	Version	Size	Diffusion Architecture	Model Params	Text Encoder(s)	TE Params	Auto Encoder
Wan-AI	WAN	2.2 A14B	118G	MMDiT	14.28B	UMT5-XXL	5.6B	48ch VAE
Wan-AI	WAN	2.1 14B	78.52GB	MMDiT	14.28B	UMT5-XXL	5.6B	16ch VAE
HiDream-AI	HiDream	I2 Fast/Dev/Full	42.71 GB + 15.69	MMDiT	17.10B	CLiP ViT-L + ViT+G + T5-XXL + LLama-3.1-8B	0.12B + 0.69B + 2.95B + 4.54B	16ch VAE
Qwen	Qwen-Image	1	56.10GB	MMDiT	20.43	QWen-2.5	8.29B
nVidia	Cosmos-Predict2 T2I	14B	37.36GB	MMDiT	14.26B	T5-XXL	4.86	WAN-VAE
Black Forest Labs	Flux	1 Dev/Schnell Kontext Krea	32.93GB	MMDiT	11.9B	CLiP ViT-L + T5-XXL	0.12B + 4.76B	16ch VAE
ShuttleAI	shuttle	3.1	32.93GB	MMDiT	11.9B	CLiP ViT-L + T5-XXL	0.12B + 4.76B	16ch VAE
WAN-AI	WAN	2.2 5B	32G	tbd	5B	tbd	tbd	tbd
FAL	AuraFlow	0.3	31.90GB	MMDiT	6.8B	UMT5	12.1B	VAE
VectorSpaceLab	OmniGen	v2	30.50GB	Transformer	3.97B	Qwen-VL-2.5	3.75B	VAE
Thudm	CogView	4	30.39GB	DiT	6.37B	GLM-4	9.40B	VAE
Kandinsky	Kandinsky	3	27.72GB	Unet	3.05B	T5-XXXL	8.72B	VQ
Wan-AI	WAN	2.1 1.3B	27.72GB	MMDiT	1.42xB	UMT5-XXL	5.68B	16ch VAE
StabilityAI	Stable Diffusion	3.5 Large	26.98GB	MMDiT	8.05B	CLiP ViT-L + ViT+G + T5-XXL	0.12B + 0.69B + 4.76B	16ch VAE
lodestones	Chroma	36	26.84GB	MMDiT	8.9B	CLiP ViT-L + T5-XXL	0.12B + 4.76B	16ch VAE
Ostris	Flex	1 Alpha	25.65GB	MMDiT	4.0B	CLiP ViT-L + T5-XXL	0.12B + 2.95B	16ch VAE
Thudm	CogView	3 Plus	24.96GB	DiT	2.85B	T5-XXL	4.76B	VAE
PixArt	Α, Σ	XL 2	21.3GB	DiT	0.61B	T5-XXL	4.76B	VAE
AlphaVLLM	Lumina	2	20.75GB	DiT	2.61B	Gemma-2	2.61B	16ch VAE
FreePix	F-Lite		19.81GB	MMDiT	9.8B	T5-XXL	2.95B	16ch VAE
Bria AI	Bria	3.2	18.66GB	MMDiT	3.78B	T5-XXL	4.76B	16ch VAE
Kwai	Kolors	N/A	17.40GB	UNnet	2.58B	ChatGLM	6.24B	VAE
StabilityAI	Stable Diffusion	3.5 Medium	15.89GB	MMDiT	2.25B	CLiP ViT-L + ViT+G + T5-XXL	0.12B + 0.69B + 4.76B	16ch VAE
NVLabs	Sana	1.5 4.8B	15.58GB	MMDiT	4.72B	Gemma2	2.61B	DC-AE
DeepFloyd	IF	L	15.48GB	Multi-stage UNet	0.61B + 0.93B	T5-XXL	4.76B	Pixel
VectorSpaceLab	OmniGen	v1	15.47GB	Transformer	3.76B	Phi-3	0	VAE
StabilityAI	Stable Diffusion	3.0 Medium	15.14GB	MMDiT	2.0B	CLiP ViT-L + ViT+G + T5-XXL	0.12B + 0.69B + 4.76B	16ch VAE
Tencent	HunyuanDiT	1.2	14.09GB	DiT	1.5B	BERT + T5-XL	3.52B + 1.67B	VAE
PlaygroundAI	Playground	2.x	13.35GB	UNet	2.56B	CLiP ViT-L + ViT+G	0.12B + 0.69B	VAE
nVidia	Cosmos-Predict2 T2I	2B	13.32GB	MMDiT	1.96B	T5-XXL	4.86	WAN-VAE
DeepFloyd	IF	M	12.79GB	Multi-stage UNet	0.37B + 0.46B	T5-XXL	4.76B	Pixel
NVLabs	Sana	1.0 1600M	12.63GB	MMDiT	1.60B	Gemma2	2.61B	DC-AE
Warp AI	Wuerstchen	N/A	12.16GB	Multi-stage UNet	1.0B + 1.05B	CLiP ViT-L + ViT+G	0.12B + 0.69B	42x VQE
StabilityAI	Stable Cascade	Medium	11.82GB	Multi-stage UNet	1.56B + 3.6B	CLiP ViT-G	0.69B	42x VQE
NVLabs	Sana	1.5 1.6B	9.49GB	MMDiT	1.60B	Gemma2	2.61B	DC-AE
Segmind	SSD-1B	N/A	8.72GB	UNet	1.33B	CLiP ViT-L + ViT+G	0.12B + 0.69B	VAE
AlphaVLLM	Lumina	Next SFT	8.67GB	DiT	1.7B	Gemma	2.5B	VAE
NVLabs	Sana	1.0 600M	7.51GB	MMDiT	0.59B	Gemma2	2.61B	DC-AE
Salesforce	BLIP-Diffusion	N/A	7.23GB	UNet	0.86B	CLiP ViT-L + BLiP-2	0.12B + 0.49B	VAE
StabilityAI	Stable Diffusion	XL	6.94GB	UNet	2.56B	CLiP ViT-L + ViT+G	0.12B + 0.69B	VAE
Koala	Koala	700M	6.58GB	UNet	0.78B	CLiP ViT-L + ViT+G	0.12B + 0.69B	VAE
Segmind	Vega	N/A	6.43GB	UNet	0.75B	CLiP ViT-L + ViT+G	0.12B + 0.69B	VAE
Thu-ML	UniDiffuser	v1	5.37GB	U-ViT	0.95B	CLiP ViT-L + CLiP ViT-B	0.12B + 0.16B	VAE
Kandinsky	Kandinsky	2.2	5.15GB	Unet	1.25B	CLiP ViT-G	0.69B	VQ
StabilityAI	Stable Cascade	Lite	4.97GB	Multi-stage UNet	0.7B + 1.0B	CLiP ViT-G	0.69B	42x VQE
PlaygroundAI	Playground	1	4.95GB	UNet	0.86B	CLiP ViT-L	0.12B	VAE
MeissonFlow	Meissonic	N/A	3.64GB	DiT	1.18B	CLiP ViT-H	0.35B	VQ
Open-MUSE	aMUSEd	256	3.41GB	ViT	0.60B	CLiP ViT-L	0.12B	VQ
StabilityAI	Stable Diffusion	2.1	2.58GB	UNet	0.86B	CLiP ViT-H	0.34B	VAE
StabilityAI	Stable Diffusion	1.5	2.28GB	UNet	0.86B	CLiP ViT-L	0.12B	VAE
IDKiro	SDXS	N/A	2.05GB	UNet	0.32B	CLiP ViT-L	0.12B	VAE
Segmind	Tiny	N/A	1.03GB	UNet	0.32B	CLiP ViT-L	0.12B	VAE

Generation Examples

Searching for highest resolution available on SD.Next ipex (dev 2025-07-15) on ultra 9 185H + 128GB RAM

Test Prompt: VibrantlySharp style, 3pic_vist4, animeniji, in a soft-focus painterly concept art style Upper body profile view, autumn dryad elf adorned in translucent red foliage and bark-textured silk, wind animating her hair and dress into motion, background: ancient twilight forest, distant amber sunlight diffused through fog and drifting leaves.

Stable Diffusion 3.5 Large+ class models

Text to Image generation examples for models with size larger than 26.8GB. It is range from lodestone Chroma and SD3.5L which both are about 27GB in size. And up to Wan 2.1 14B which is about 75GB in size. Decent quality is generated in about 10 minutes on this Model size class on Intel Core Ultra 9 185H with 128GB of DDR5 (ipex, iGPU)

🪪- Models require identification, access request, agree on terms

Model

Low steps

High steps

Wan2.2-T2V-A14B

256x256 - 2m 7.55s @ 24 STEPS

256x256 - 4m 21.77s @ 50 STEPS

512x512 -5m 55.53s @ 20STEPS

768x768 - hangs

768x1024 - hangs

1024x1024- 27m 6.16s @ 20STEPS
1024x1024- 66m 59.70s @ 50STEPS

UniPC Multistep Scheduler

generation progress + preview

⭐⭐

2025-07-30

2025-07-31 - OOM for all

Time: 27m 6.16s | total 1721.93 pipeline 1625.83 preview 64.05 vae 10.38 offload 10.13 te 9.62 gc 2.15 post 0.31 | GPU 39894 MB 31% | RAM 29.77 GB 24%

Time: 66m 59.70s | total 4090.56 pipeline 4019.37 preview 46.47 vae 10.70 offload 10.06 gc 2.17 te 2.04 post 0.32 | GPU 40040 MB 31% | RAM 29.63 GB 24%

Wan2.1-T2V-14B

256x256 - 1m 41.11s @ 16 STEPS

512x512 - 5m 48.96s @ 20STEPS

512x512 - 11m 42.14s @ 40STEPS

512x768 - hangs

UniPC Multistep Scheduler

generation progress + preview

⭐⭐⭐⭐

512x512- 5m 48.96s @ 20 STEPS

Time: 5m 48.96s | total 511.83 pipeline 348.65 preview 140.76 offload 9.90 vae 8.66 gc 2.12 te 1.99 post 0.29 | GPU 39054 MB 30% | RAM 29.72 GB 24%

512x512- 11m 42.14s @ 40 STEPS

Time: 11m 42.14s | total 1079.70 pipeline 701.83 preview 345.45 vae 12.42 offload 11.25 te 6.87 gc 2.12 post 0.28 | GPU 38830 MB 30% | RAM 29.44 GB 23%

HiDream I1 Full🪪

(Meta-Llama-3.1-8B-Instruct 🪪)

~~512x512 generates 1024x1024~~

~~768x768 generates 1024x1024~~

1024x1024 -10m 29.57s @ 12 STEPS

1024x1024 -16m 42.62s @ 20 STEPS

1024x1024 -41m 42.55s @ 50 STEPS

~~1280x1280 generates 1024x1024~~

~~2048x2048 generates 1024x1024~~

UniPC Multistep Scheduler

generation progress + preview

⭐⭐⭐⭐⭐

1024x1024 - 10m 29.57s @ 12 STEPS

Time: 10m 29.57s | total 685.95 pipeline 593.18 move 21.17 prompt 21.13 te 18.92 decode 14.95 offload 11.79 preview 4.42 gc 0.61 post 0.27 | GPU 4508 MB 4% | RAM 58.59 GB 47%

1024x1024 - 41m 42.55s @ 50 STEPS

Time: 41m 42.55s | total 2576.47 pipeline 2466.40 preview 22.89 move 20.63 prompt 20.52 te 18.32 decode 15.24 offload 12.13 gc 0.55 post 0.27 | GPU 4572 MB 4% | RAM 58.59 GB 47%

HiDream I1 Dev🪪

1024x1024 17m 47.83s @ 20 STEP

1024x1024 23m 59.26s @ 28 STEP

FlowMatch LCM Scheduler

generation progress + preview

⭐⭐⭐⭐⭐

1024x1024 - 17m 47.83s @ 20 STEPS

Time: 17m 47.83s | total 1151.56 pipeline 1019.03 move 30.23 prompt 30.11 te 27.23 decode 18.25 offload 15.44 preview 10.76 gc 0.74 post 0.32 | GPU 4378 MB 3% | RAM 58.68 GB 47%

1024x1024 - 23m 59.26s @ 28 STEPS

Time: 23m 59.26s | total 1529.78 pipeline 1389.13 move 34.36 prompt 34.23 offload 22.57 te 21.92 decode 15.51 preview 11.74 gc 0.53 post 0.26 | GPU 4510 MB 4% | RAM 58.7 GB 47%

HiDream I1 Fast🪪

CFG (0-2)

1024x1024 3m 39.03s @ 4 STEP

1024x1024 20m 16.60s @ 24 STEP

FlowMatch LCM Scheduler

generation progress + preview

⭐⭐⭐⭐⭐

1024x1024 - 3m 39.03s @ 4 STEPS

Time: 3m 39.03s | total 249.11 pipeline 194.59 offload 18.99 decode 14.74 move 9.39 prompt 9.24 preview 1.85 gc 0.57 post 0.32 | GPU 4440 MB 3% | RAM 58.61 GB 47%

1024x1024 - 20m 16.60s @ 24 STEPS

Time: 20m 16.60s | total 1283.51 pipeline 1178.55 move 22.60 prompt 22.47 te 20.27 decode 15.17 offload 12.25 preview 11.87 gc 0.53 post 0.27 | GPU 4502 MB 4% | RAM 58.63 GB 47%

HiDream E1 Full 🪪

~~768x768 generates 1024x1024~~

1024x1024 7m 33.98s @ 8 STEP

1024x1024 10m 25.10s @ 12 STEP

1024x1024 13m 40.41s @ 16 STEP

1024x1024 26m 55.45s @ 32 STEP

UniPC Multistep Scheduler

generation progress + preview

⭐⭐⭐⭐

1024x1024 - 13m 40.41s @ 16 STEPS

Time: 13m 40.41s | total 860.48 pipeline 782.43 move 21.89 prompt 19.66 decode 15.80 offload 12.73 preview 7.64 gc 0.52 post 0.27 | GPU 4500 MB 4% | RAM 58.7 GB 47%

1024x1024 - 26m 55.45s @ 32 STEPS

Time: 26m 55.45s | total 1665.14 pipeline 1578.31 move 20.86 prompt 18.67 preview 17.96 decode 16.01 offload 13.01 gc 0.54 post 0.27 | GPU 4564 MB 4% | RAM 58.64 GB 47%

Qwen-Image

With true_cfg=0

512x512 - 2m 0.22s @ 20 steps

768x768 - 5m 23.78s @ 20 steps

1024x1024 - 9m 18.46s @ 16 steps

1024x1024 - 11m 34.66s @ 20 steps

1024x1024 - 18m 46.91s @ 32 steps

1024x1536 - 21m 59.50s @ 20 steps

1600x1600 - 49m 24.79s @ 20 steps

With true_cfg=4

1024x1024 - 18m 46.05s @ 16 steps

1024x1024 - 37m 42.68s @ 32 steps

FlowMatch Euler DiscreteScheduler

progress without preview

⭐⭐⭐⭐

created 2025-08-05

update 2025-08-08

1024x1024 - 9m 18.46s @ 16 STEPS

Time: 9m 18.46s | total 559.47 pipeline 558.44 te 0.98 | GPU 61110 MB 48% | RAM 3.1 GB 2%

1024x1024 - 18m 46.05s @ 16 STEPS

Time: 18m 46.05s | total 1132.26 pipeline 1125.98 te 4.41 vae 1.79 | GPU 60262 MB 47% | RAM 3.13 GB 2%

1024x1024 - 18m 46.91s @ 32 STEPS

Time: 18m 46.91s | total 1127.93 pipeline 1126.88 te 0.98 | GPU 61110 MB 48% | RAM 3.1 GB 2%

1024x1024 - 37m 42.68s @ 16 STEPS

Time: 37m 42.68s | total 2264.61 pipeline 2262.65 te 1.89 | GPU 60264 MB 47% | RAM 3.12 GB 2%

Cosmos-Predict2 T2I 14B 🪪

256x256 - 2m 31.30s @ 20 STEPS

512x512 - 5m 54.52s @ 20 STEPS

768x768 - 10m 23.35s @ 16 STEPS

768x768 - 12m 41.39s @ 20 STEPS

768x768 - 20m 9.11s @ 32 STEPS

768x1024 - hangs

FlowMatch Euler DiscreteScheduler

progress without preview

"diffusers_offload_mode": "model"

⭐⭐⭐⭐

768x768- 10m 23.35s @ 16 STEPS

Time: 10m 23.35s | total 643.47 pipeline 623.02 te 11.87 vae 8.56 post 0.32 gc 0.31 | GPU 27938 MB 22% | RAM 38.86 GB 31%

768x768- 20m 9.11s @ 32 STEPS

Time: 20m 9.11s | total 1224.37 pipeline 1208.85 vae 7.85 te 7.65 post 0.26 | GPU 27938 MB 22% | RAM 38.88 GB 31%

shuttle-3.1-aesthetic

512x512 - 6m 11.47s @ 50 STEPS

768x768 - 4m 35.14s @ 20 STEPS

768x1024 - 6m 3.73s @ 20 STEPS

1024x1024 -8m 5.95s @20 STEPS

1024x1280 - 4m 31.69s @ 8 STEPS

1024x1536 - 3m 58.29s @ 6 STEPS

1024x1536 - 20m 29.93s@ 32 STEPS

1280x1600 - 13m 44.74s @ 8 STEPS

1536x2048 - 12m 45.33s @ 6 STEPS

FlowMatch Euler DiscreteScheduler

generation progress + preview

⭐⭐⭐⭐⭐

2025-07-30

1024x1536 - 3m 58.29s @ 6 STEPS

Time: 3m 58.29s | total 244.61 pipeline 229.67 decode 8.30 preview 6.31 gc 0.55 post 0.30 | GPU 37402 MB 29% | RAM 3.21 GB 3%

1024x1536 - 20m 29.93s@ 32 STEPS

Time: 20m 29.93s | total 1270.02 pipeline 1221.19 preview 36.27 decode 8.43 prompt 1.92 te 1.89 gc 0.55 post 0.31 | GPU 37066 MB 29% | RAM 3.22 GB 3%

FLUX.1-dev 🪪

64x64 - 58.97s @ 16 STEPS

128x128 - 1m 0.43s @ 16 STEPS

256x256 - 1m 7.51s @ 16 STEPS

512x512 - 2m 0.41s @ 16 STEPS

768x768 - 3m 41.69s @ 16 STEPS

768x1024 - 4m 54.68s @ 16 STEPS

768x1024 - 14m 56.94s@ 40 STEPS

1024x1536 - 5m 9.01s @ 8 STEPS

1024x1536 - 20m 13.60s @ 32 steps

FlowMatch Euler DiscreteScheduler

generation progress + preview

⭐⭐⭐⭐

Updated 2025-07-30

1024x1536 - 5m 9.01s @ 8 STEPS

Time: 5m 9.01s | total 321.00 pipeline 300.51 decode 8.20 preview 8.10 prompt 1.96 te 1.94 gc 0.52 post 0.30 | GPU 37048 MB 29% | RAM 3.24 GB 3%

1024x1536 - 20m 13.60s @ 32 STEPS

Time: 20m 13.60s | total 1217.57 pipeline 1204.89 decode 8.40 prompt 1.91 te 1.89 gc 0.54 post 0.31 | GPU 37048 MB 29% | RAM 3.23 GB 3%

FLUX.1-Krea-dev 🪪

512x512 - 7m 27.95s @ 20 STEPS

768x768 - 2m 24.02s @ 8 STEPS

768x1024 - 2m 23.60s @ 8 STEPS

1024x1024 - 6m 22.68s @ 16 STEPS

1024x1024 - 12m 35.18s @ 32steps

1024x1536 - hangs

FlowMatch Euler DiscreteScheduler

⭐⭐⭐⭐

Added 2025-08-01

Time: 6m 22.68s | total 471.62 pipeline 376.23 preview 84.90 decode 6.43 prompt 1.90 te 1.88 gc 0.25 | GPU 34688 MB 27% | RAM 3.39 GB 3%

Time: 12m 35.18s | total 1494.92 pipeline 748.48 preview 731.88 decode 6.66 prompt 3.78 te 3.76 gc 0.31 | GPU 35284 MB 27% | RAM 2.97 GB 2%

FLUX.1-schnell 🪪

768x768 - ok

768x1024 - 1m 24.86s @ 4 STEPS

768x1024 - 7m 45.92s @ 24 STEPS

1024x1024 - hangs

FlowMatch Euler DiscreteScheduler

⭐⭐⭐⭐

768x1024 - 1m 24.86s @ 4 STEPS

Time: 1m 24.86s | total 127.77 pipeline 74.92 preview 36.64 decode 9.63 offload 6.53 post 0.29 gc 0.27 | GPU 33784 MB 26% | RAM 25.14 GB 20%

768x1024 - 7m 45.92s @ 24 STEPS

Time: 7m 45.92s | total 903.61 pipeline 438.91 preview 400.19 move 14.96 prompt 14.96 te 14.91 decode 11.74 offload 7.80 post 0.30 gc 0.28 | GPU 33614 MB 26% | RAM 25.17 GB 20%

AuraFlow v0.3

~~512x512 - 2m 4.61s @ STEPS 16~~

768x768 - 4m 37.98s @ STEPS 16

768x768 - 14m 17.12s @ STEPS 50

768x768 - 29m 46.26s @ STEPS 100

1024x1024 - hangs

FlowMatch Euler DiscreteScheduler

generation progress + preview

⭐

768x768- 14m 17.12s @ 50 STEPS

Time: 14m 17.12s | total 1163.36 pipeline 853.70 preview 306.23 decode 3.15 gc 0.47 post 0.25 | GPU 17638 MB 14% | RAM 2.75 GB 2%

768x768- 29m 46.26s @ 100 STEPS

Time: 29m 46.26s | total 3499.63 pipeline 1782.79 preview 1713.40 decode 3.13 gc 0.56 post 0.32 | GPU 17638 MB 14% | RAM 2.76 GB 2%

Wan2.2-TI2V-5B

~~256x256 - 1m 26.51s @ STEPS 50~~

~~512x512 - 47.81s @ STEPS 20~~

~~512x512 - 1m 48.58s @ STEPS 50~~

768x768 - 3m 50.42s @ STEPS 50

704x1280 - 2m 17.40s @ STEPS 20

704x1280 - 5m 29.53s @ STEPS 50

1024x1024 - 2m 28.14s @ STEPS 20

UniPC Multistep Scheduler

progress without preview

⭐⭐

Time: 2m 17.40s | total 140.87 pipeline 137.10 te 2.03 vae 1.68 post 0.29 gc 0.27 | GPU 25558 MB 20% | RAM 3.09 GB 2%

$00028-2025-07-29-Wan2.2-TI2V-5B-Diffusers-1280x704-Seed1460879190-CFG6-STEP50.jpg$

Time: 5m 29.53s | total 332.80 pipeline 329.25 te 2.02 vae 1.47 post 0.28 gc 0.25 | GPU 25558 MB 20% | RAM 3.09 GB 2%

OmniGen2

512x512 - 2m 12.11s @ 20STEPS

768x768 - 5m 39.35s @ 20STEPS

1024x1024 - 9m 33.83s @ 16 STEPS

1024x1024 - 12m 35.39s @ 20STEPS

1024x1024 - 30m 29.41s @ 50STEPS

1280x1280 - 9m 30.35s @ 8STEPS

1280x1280 - 14m 26.83s @ 12STEPS

1280x1280 - 23m 49.50s @ 20STEPS

1280x1280 - 60m 14.48s @ 50STEPS

1600x1600 - 50m 50.81s @ 20STEPS

2048x2048 - hangs

FlowMatch Euler DiscreteScheduler

generation progress not shown

⭐⭐⭐⭐

1024x1024 - 9m 33.83s @ 16 STEPS

Time: 9m 33.83s | total 573.61 pipeline 573.55 post 0.26 | GPU 18054 MB 14% | RAM 2.71 GB 2%

1024x1024 - 30m 29.41s @ 50 STEPS

Time: 30m 29.41s | total 1829.18 pipeline 1829.12 post 0.27 | GPU 18054 MB 14% | RAM 2.72 GB 2%

CogView 4

256x256 - 1m 11.07s @ 16 STEPS

512x512 - 2m 44.92s @ 16 STEPS

768x768 - 6m 44.86s @ 16 STEPS

1024x1024 - 14m 9.97s @ 16 STEPS

1024x1024 - 43m 56.72s@ 50 STEPS

1280x1280 - hangs

FlowMatch Euler DiscreteScheduler

progress without preview

⭐⭐⭐⭐

1024x1024 - 14m 9.97s @ 16 STEPS

Time: 14m 9.97s | total 857.86 pipeline 825.42 decode 16.59 move 15.57 gc 0.53 post 0.29 | GPU 7926 MB 6% | RAM 31.98 GB 26%

1024x1024 - 43m 56.72s @ 50 STEPS

Time: 43m 56.72s | total 2644.35 pipeline 2612.41 decode 16.72 move 14.90 gc 0.53 post 0.26 | GPU 7926 MB 6% | RAM 31.99 GB 26%

Kandinsky 3

768x768 - 1m 5.40s @ 20STEPS

768x1024 - 45.45s @ 8 STEP

768x1024 - 1m 9.35s @ 20 STEP

768x1024 - 2m 35.93s @ 64 STEP

1024x1024 - hangs

DDPM Scheduler

progress without preview

⭐⭐

768x1024 - 1m 9.35s @ 20 STEPS

Time: 1m 9.35s | total 73.34 pipeline 64.79 offload 4.26 decode 4.24 post 0.29 gc 0.27 | GPU 29500 MB 23% | RAM 18.93 GB 15%

768x1024 - 2m 35.93s @ 64 STEPS

Time: 2m 35.93s| total 159.72 pipeline 151.59 offload 4.05 decode 4.03 post 0.29 gc 0.27 | GPU 29482 MB 23% | RAM 18.93 GB 15%

Wan2.1-T2V-1.3B

256x256 - 23.16s @ 16 STEPS

512x512 - 54.41s @ 20STEPS

768x768 - 3m 43.11s @ 40STEPS

1024x1024 - 7m 35.17s @ 40STEPS

1280x1280 - 12m 42.76s @ 40STEPS

(resolution to be divisible by 16)

UniPC Multistep Scheduler

generation progress + preview

⭐

512x512 - 54.41s @ 20 STEPS

Time: 54.41s | total 96.96 pipeline 54.15 preview 36.26 gc 2.09 te 1.98 offload 1.88 vae 0.80 | GPU 15416 MB 12% | RAM 3.16 GB 3%

1280x1280- 12m 42.76s @ 40 STEPS

Time: 12m 42.76s | total 1282.37 pipeline 762.43 preview 511.77 vae 2.40 gc 2.11 te 1.98 offload 1.88 post 0.30 | GPU 21686 MB 17% | RAM 3.14 GB 3%

Stable Diffusion 3.5 L 🪪

512x512 - 4m 18.84s @ 32 STEPS

512x768 - 3m 25.31s @ 16 STEPS

512x768 - 11m 9.34s @ 50 STEPS

768x768 - @

1024x1024 - 14m 15.60s @ 16 steps

1024x1024 - 28m 56.13s @ 32 steps

1024x1536 - 29m 24.32s @ 16 steps

1024x1536 - 58m 47.43s @ 32 steps

FlowMatch Euler DiscreteScheduler

generation progress + preview

⭐⭐⭐⭐

update 2025-07-30

Time: 14m 15.60s | total 921.29 pipeline 848.32 preview 61.93 decode 6.98 prompt 3.64 gc 0.65 post 0.30 | GPU 29320 MB 23% | RAM 3.11 GB 2%

Time: 28m 56.13s | total 1743.07 pipeline 1729.54 preview 6.68 decode 6.57 gc 0.26 | GPU 29320 MB 23% | RAM 3.0 GB 2%

Chroma 36

512x512 - 4m 59.14s @ 20 STEPS

768x768 - 7m 4.50s @ 16 STEPS

768x768 - 21m 56.62s @ 50 STEPS

768X1024 - 12m 0.43s @ 20 steps

1024x1024 - hangs

progress with preview

FlowMatch Euler DiscreteScheduler

⭐⭐⭐

only some CFG/resolutions is ok

768x768 - 7m 4.50s @ 16 STEPS

Time: 7m 4.50s | total 627.03 pipeline 416.87 preview 197.83 decode 7.35 offload 4.69 gc 0.49 post 0.26 | GPU 11132 MB 9% | RAM 19.33 GB 15%

768x768 - 21m 56.62s @ 50 STEPS

Time: 21m 56.62s | total 1945.92 pipeline 1308.96 preview 624.37 decode 7.38 offload 4.90 gc 0.50 post 0.27 | GPU 11154 MB 9% | RAM 19.62 GB 16%

chroma-unlocked-v48-detail-calibrated

512x768 - 4m 51.82s @ 16 STEPS

768x768 - 7m 6.56s @ 16 STEPS

768x768 - 14m 11.86s @ 32 STEPS

768x1024 - 9m 30.73s @ 16 STEPS

1024x1024- 13m 3.35s @ 16 STEPS

FlowMatch Euler DiscreteScheduler

⭐⭐⭐

Added 2025-08-01

Time: 7m 6.56s | total 563.91 pipeline 423.91 preview 137.06 decode 2.61 gc 0.28 | GPU 28122 MB 22% | RAM 2.94 GB 2%

Time: 14m 11.86s | total 1162.50 pipeline 849.22 preview 302.91 prompt 3.72 te 3.70 decode 2.62 gc 0.30 | GPU 28134 MB 22% | RAM 2.95 GB 2%

SD.Next Config

{
  "samples_filename_pattern": "[seq]-[date]-[model_name]-[height]x[width]-Seed[seed]-CFG[cfg]-STEP[steps]",
  "diffusers_offload_mode": "none",
  "diffusers_to_gpu": true,
  "device_map": "gpu"
}

Dockerfile variables

# intel specific environment variables
ENV IPEX_SDPA_SLICE_TRIGGER_RATE=1
ENV IPEX_ATTENTION_SLICE_RATE=0.5
ENV IPEX_FORCE_ATTENTION_SLICE=1
ENV IPEXRUN=true

SD.Next on Linux

HW and OS level articles

SD.Next on Linux

SD.Next openvino docker compose

Install

vi Dockerfile
vi docker-compose.yml
docker compose up -d
docker logs sdnext-openvino --follow
#docker compose down

Dockerfile

example without proxy on branch master

# SD.Next OpenVINO Dockerfile
# docs: <https://github.com/vladmandic/sdnext/wiki/Docker>

# base image
FROM ubuntu:noble

# metadata
LABEL org.opencontainers.image.vendor="SD.Next"
LABEL org.opencontainers.image.authors="disty0"
LABEL org.opencontainers.image.url="https://github.com/vladmandic/sdnext/"
LABEL org.opencontainers.image.documentation="https://github.com/vladmandic/sdnext/wiki/Docker"
LABEL org.opencontainers.image.source="https://github.com/vladmandic/sdnext/"
LABEL org.opencontainers.image.licenses="AGPL-3.0"
LABEL org.opencontainers.image.title="SD.Next OpenVINO"
LABEL org.opencontainers.image.description="SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models"
LABEL org.opencontainers.image.base.name="https://hub.docker.com/_/ubuntu:noble"
LABEL org.opencontainers.image.version="latest"

# essentials
RUN apt-get update && \
    apt-get install -y --no-install-recommends --fix-missing \
    software-properties-common \
    build-essential \
    ca-certificates \
    wget \
    gpg \
    git

# intel compute runtime
RUN wget -qO - https://repositories.intel.com/gpu/intel-graphics.key | gpg --yes --dearmor --output /usr/share/keyrings/intel-graphics.gpg
RUN echo "deb [arch=amd64,i386 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu noble client" | tee /etc/apt/sources.list.d/intel-gpu-noble.list
RUN apt-get update

RUN apt-get install -y --no-install-recommends --fix-missing \
    intel-opencl-icd \
    libze-intel-gpu1 \
    libze1

# required by pytorch / ipex
RUN apt-get install -y --no-install-recommends --fix-missing \
    libgl1 \
    libglib2.0-0 \
    libgomp1

# python3.12
RUN apt-get install -y --no-install-recommends --fix-missing \
    python3 \
    python3-dev \
    python3-venv \
    python3-pip

# cleanup
RUN /usr/sbin/ldconfig
RUN apt-get clean && rm -rf /var/lib/apt/lists/*

# stop pip and uv from caching
ENV PIP_NO_CACHE_DIR=true
ENV UV_NO_CACHE=true

# set paths to use with sdnext
ENV SD_DOCKER=true
ENV SD_DATADIR="/mnt/data"
ENV SD_MODELSDIR="/mnt/models"
ENV venv_dir="/mnt/python/venv"

# paths used by sdnext can be a volume if necessary
#VOLUME [ "/app" ]
#VOLUME [ "/mnt/data" ]
#VOLUME [ "/mnt/models" ]
#VOLUME [ "/mnt/python" ]
#VOLUME [ "/root/.cache/huggingface" ]

# git clone and run sdnext
RUN echo '#!/bin/bash\ngit status || git clone https://github.com/vladmandic/sdnext.git .\n/app/webui.sh "$@"' | tee /bin/startup.sh
RUN chmod 755 /bin/startup.sh

# actually run sdnext
WORKDIR /app
ENTRYPOINT [ "startup.sh", "-f", "--use-openvino", "--uv", "--listen",  "--insecure", "--update", "--debug", "--api-log", "--log", "sdnext.log" ]

# expose port
EXPOSE 7860

# healthcheck function
# HEALTHCHECK --interval=60s --timeout=10s --start-period=60s --retries=3 CMD curl --fail http://localhost:7860/sdapi/v1/status || exit 1

# stop signal
STOPSIGNAL SIGINT

example with proxy

(to be removed or replaced to one is working) and on Developer branch

# SD.Next OpenVINO Dockerfile
# docs: <https://github.com/vladmandic/sdnext/wiki/Docker>

# base image
FROM ubuntu:noble

# metadata
LABEL org.opencontainers.image.vendor="SD.Next"
LABEL org.opencontainers.image.authors="disty0"
LABEL org.opencontainers.image.url="https://github.com/vladmandic/sdnext/"
LABEL org.opencontainers.image.documentation="https://github.com/vladmandic/sdnext/wiki/Docker"
LABEL org.opencontainers.image.source="https://github.com/vladmandic/sdnext/"
LABEL org.opencontainers.image.licenses="AGPL-3.0"
LABEL org.opencontainers.image.title="SD.Next OpenVINO"
LABEL org.opencontainers.image.description="SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models"
LABEL org.opencontainers.image.base.name="https://hub.docker.com/_/ubuntu:noble"
LABEL org.opencontainers.image.version="latest"

# essentials
RUN apt-get update && \
    apt-get install -y --no-install-recommends --fix-missing \
    software-properties-common \
    build-essential \
    ca-certificates \
    wget \
    gpg \
    git

# intel compute runtime
RUN https_proxy=http://proxy.example.com:18080/  wget -qO - https://repositories.intel.com/gpu/intel-graphics.key | gpg --yes --dearmor --output /usr/share/keyrings/intel-graphics.gpg
RUN echo "deb [arch=amd64,i386 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu noble client" | tee /etc/apt/sources.list.d/intel-gpu-noble.list
RUN https_proxy=http://proxy.example.com:18080/  apt-get update

RUN https_proxy=http://proxy.example.com:18080/  apt-get install -y --no-install-recommends --fix-missing \
    intel-opencl-icd \
    libze-intel-gpu1 \
    libze1

# required by pytorch / ipex
RUN https_proxy=http://proxy.example.com:18080/  apt-get install -y --no-install-recommends --fix-missing \
    libgl1 \
    libglib2.0-0 \
    libgomp1

# python3.12
RUN https_proxy=http://proxy.example.com:18080/  apt-get install -y --no-install-recommends --fix-missing \
    python3 \
    python3-dev \
    python3-venv \
    python3-pip

# cleanup
RUN /usr/sbin/ldconfig
RUN https_proxy=http://proxy.example.com:18080/  apt-get clean && rm -rf /var/lib/apt/lists/*

# stop pip and uv from caching
ENV PIP_NO_CACHE_DIR=true
ENV UV_NO_CACHE=true

# set paths to use with sdnext
ENV SD_DOCKER=true
ENV SD_DATADIR="/mnt/data"
ENV SD_MODELSDIR="/mnt/models"
ENV venv_dir="/mnt/python/venv"

# paths used by sdnext can be a volume if necessary
#VOLUME [ "/app" ]
#VOLUME [ "/mnt/data" ]
#VOLUME [ "/mnt/models" ]
#VOLUME [ "/mnt/python" ]
#VOLUME [ "/root/.cache/huggingface" ]

# git clone and run sdnext
RUN echo '#!/bin/bash\ngit status || git clone https://github.com/vladmandic/sdnext.git -b dev .\n/app/webui.sh "$@"' | tee /bin/startup.sh
RUN chmod 755 /bin/startup.sh

# actually run sdnext
WORKDIR /app
ENTRYPOINT [ "startup.sh", "-f", "--use-openvino", "--uv", "--listen",  "--insecure", "--server-name sdnext.example.com", "--update", "--debug", "--api-log", "--log", "sdnext.log" ]

# expose port
EXPOSE 7860

# healthcheck function
# HEALTHCHECK --interval=60s --timeout=10s --start-period=60s --retries=3 CMD curl --fail http://localhost:7860/sdapi/v1/status || exit 1

# stop signal
STOPSIGNAL SIGINT

docker-compose.yml

services:
  sdnext-openvino:
    build:
      dockerfile: Dockerfile
    image: sdnext-openvino:latest
    container_name: sdnext-openvino
    restart: unless-stopped
    devices:
      - /dev/dri:/dev/dri
# if Intel CPU got NPU 
#      - /dev/accel:/dev/accel
# if need proxy for internet access
#    environment:
#      - https_proxy=http://proxy.example.com:18080/
    ports:
      - 7860:7860
    volumes:
      - /docker/sdnext/app-volume:/app
      - /docker/sdnext/mnt-volume:/mnt
      - /docker/sdnext/huggingface-volume:/root/.cache/huggingface
      - /dev/shm:/dev/shm

Application screenshots

$image.png$

Post-install fixes

add some python libraries

That is not something you beed to do, but in rare case of it kind of missing in requirements.txt

# In case of some dependencies is missing (like sentencepiece in Dev branch at 2025-07-14)
# docker exec -it sdnext-openvino bash
python3 -m venv /mnt/python/venv
/mnt/python/venv/bin/pip3 install sentencepiece

Fix access to gated repo

(in case access is granted and token available)

Access error to gated repo

sdnext-ipex  | 20:01:16-163840 ERROR    Load model: repo="black-forest-labs/FLUX.1-Kontext-dev"
sdnext-ipex  |                          login=True 401 Client Error. (Request ID:
sdnext-ipex  |                          Root=1-6875620c-4ddbac246caa80376513448a;bea8cc0a-fb48-
sdnext-ipex  |                          452c-a738-debd0038cb43)
sdnext-ipex  |
sdnext-ipex  |                          Cannot access gated repo for url
sdnext-ipex  |                          https://huggingface.co/api/models/black-forest-labs/FLU
sdnext-ipex  |                          X.1-Kontext-dev/auth-check.
sdnext-ipex  |                          Access to model black-forest-labs/FLUX.1-Kontext-dev is
sdnext-ipex  |                          restricted. You must have access to it and be
sdnext-ipex  |                          authenticated to access it. Please log in.

# docker exec -it sdnext-openvino bash
# export venv_dir=/mnt/python/venv/
source "${venv_dir}"/bin/activate
git config --global credential.helper store
/mnt/python/venv/bin/huggingface-cli login
# docker restart sdnext-openvino
# Download or use the model after a restart

System info

app: sdnext.git updated: 2025-07-14 hash: 79d0fb81 url: https://github.com/vladmandic/sdnext.git/tree/dev
arch: x86_64 cpu: x86_64 system: Linux release: 6.11.0-29-generic
python: 3.12.3 Torch: 2.7.1+cpu
device: Intel(R) Arc(TM) Graphics (iGPU) openvino: 2025.2.0
ram: free:95.64 used:29.69 total:125.33 gpu: total:125.33
xformers:  diffusers: 0.35.0.dev0 transformers: 4.53.2
active: cpu dtype: torch.float32 vae: torch.float32 unet: torch.float32
base: Diffusers/nvidia/Cosmos-Predict2-2B-Text2Image [acdb5fde99] refiner: none vae: none te: none unet: none

app: sdnext.git updated: 2025-08-02 hash: e90ac68d url: https://github.com/vladmandic/sdnext.git/tree/dev
arch: x86_64 cpu: x86_64 system: Linux release: 6.14.0-27-generic
python: 3.12.3 Torch: 2.7.1+cpu
device: Intel(R) Iris(R) Xe Graphics (iGPU) openvino: 2025.2.0
ram: free:59.83 used:34.14 total:93.97 gpu: total:93.97
xformers:  diffusers: 0.35.0.dev0 transformers: 4.54.1
active: cpu dtype: torch.float32 vae: torch.float32 unet: torch.float32
base: Diffusers/Kwai-Kolors/Kolors-diffusers [7e091c7519] refiner: none vae: none te: none unet: none

openvino (float32) note

Memory consumption in openvino (float32) significantly higher than ipex (bfloat16) so it may cause OOM on same HW with same models than ipex is OK with (Flux.1 dev kontext example, openvino/gpu iGPU Intel Arc)

Reinstall

Clean reinstall

Remove all sdnext related date and have clean install

docker compose down --rmi local
docker system prune --all --force
rm -Rf /docker/sdnext/*
docker compose up -d
docker compose logs --follow

Preserve downloaded models

save the models and output images

docker compose down --rmi local
docker system prune --all --force
rm -Rf /docker/sdnext/app-volume/ /docker/sdnext/python-volume/ /docker/sdnext/mnt-volume/python/
docker compose up -d
docker compose logs --follow

Config example for iGPU

cat /docker/sdnext/mnt-volume/data/config.json
{
  "no_half": false,
  "openvino_devices": [
    "GPU"
  ],
  "samples_filename_pattern": "[seq]-[date]-[model_name]-[height]x[width]-Seed[seed]-CFG[cfg]-STEP[steps]",
}

Example image generation on iGPU

SD.Next OpenVINO Intel Xe iGPU on Intel core i7-1355U + 96GB DDR5 (35/55W power limits) float32

Kolors, Cosmos, Flux, Chroma not work

Model	resolution	time (at 20 STEPS)
dreamshaper_8 (SD1.5)	512x512	55.17s
dreamshaper_8 (SD1.5)	768x768	2m 27.75s
cyberrealistic_v90 (SD1.5)	768x768	2m 20.86s
Stable Diffusion 2.1 (SD2.1)	512x512	55.17s
Stable Diffusion 2.1 (SD2.1)	768x768	2m 13.59s
Stable Diffusion 2.1 (SD2.1)	1024x1024	4m 40.88s
TempestV0.1-Artistic (SDXL)	512x512	1m 58.98s
TempestV0.1-Artistic (SDXL)	768x768	3m 23.79s
TempestV0.1-Artistic (SDXL)	1024x1024	5m 24.46s
TempestV0.1-Artistic (SDXL)	1280x1280	8m 25.92s
cyberillustrious_v60Alt (SDXL)	1024x1024	5m 13.36s
cyberrealisticXL_v60 (SDXL)	1024x1024	5m 28.79s
cyberrealisticPony_v127Alt	1024x1024	5m 31.72s
PixArt-Sigma-XL-2-512-MS	512x512	1m 30.74s
PixArt-Sigma-XL-2-1024-MS	1024x1024	6m 26.39s
Stable Diffusion 3.5 medium (SD3.5)	512x512	2m 47.35s
Stable Diffusion 3.5 medium (SD3.5)	768x768	4m 40.50s
Stable Diffusion 3.5 medium (SD3.5)	1024x1024	8m 52.75s
Stable Diffusion 3.5 large (SD3.5)	512x512	7m 7.58s

Stable Diffusion 3.5 medium

Prompt: car

Time: 4m 40.50s | total 313.18 pipeline 251.68 preview 32.68 decode 28.80 | RAM 33.87 GB 36%

PixArt-Sigma-XL-2-512-MS

Prompt: car

Time: 1m 30.74s | total 97.25 pipeline 76.88 decode 13.60 preview 6.51 | RAM 22.46 GB 24%

SD1.5

Prompt: car

Time: 2m 29.85s | total 184.46 pipeline 118.10 preview 34.60 decode 31.73 | RAM 6.66 GB 7%

Config example for CPU

{
  "no_half": false,
  "openvino_devices": [
    "CPU"
  ],
  "samples_filename_pattern": "[seq]-[date]-[model_name]-[height]x[width]-Seed[seed]-CFG[cfg]-STEP[steps]",
  "device_map": "cpu",
  "diffusers_generator_device": "CPU"
}

Example image generation on CPU

SD.Next openvino CPU on Intel core i7-1355U + 96GB DDR5 (25/55W power limits) float32

Model	resolution	time (at 20 STEPS)
Kolors	1280x1280	34m 51.83s
FLUX.1-Krea-dev	512x512	37m 28.47s (96GB RAM + 42GB swap)
~~FLUX.1-Krea-dev~~	~~768x768~~	96GB RAM + 64GB swap is not enough
dreamshaper_8 (SD1.5)	512x512	2m 58.45s
dreamshaper_8 (SD1.5)	768x768	7m 53.50s
dreamshaper_8 (SD1.5)	1024x1024	18m 10.01s
TempestV0.1-Artistic (SDXL)	512x512	5m 16.53s
TempestV0.1-Artistic (SDXL)	768x768	11m 0.86s
TempestV0.1-Artistic (SDXL)	1024x1024	20m 41.39s
TempestV0.1-Artistic (SDXL)	1280x1280	34m 5.18s

Kolors

Prompt: car

Execution: Time: 34m 51.83s | total 2091.82 pipeline 1960.52 decode 131.16 | RAM 90.33 GB 96%

Prompt: car

Flux.1 Krea

93GB of RAM + ~ 46GB of swap

Prompt: car

Execution: Time: 37m 28.47s | total 2292.37 pipeline 2224.84 decode 23.42 prompt 19.16 te 19.13 preview 5.59 | RAM 91.66 GB 98%

Use nginx for SD.Next

install nginx

# SSL certificate
#vi /etc/ssl/private/wildcard.pem
#chmod 600 /etc/ssl/private/wildcard.pem
#vi /etc/ssl/certs/wildcard.ca
# Nginx
sudo apt install -y nginx
vi /etc/nginx/conf.d/sdnext.conf
systemctl restart nginx
systemctl status nginx

nginx config example

server {
        listen       80;
        server_name sdnext.example.com;
        return 301 https://$server_name$request_uri;
    }
server {
        listen 443 ssl;
        server_name sdnext.example.com;
        # SSL Settings
        ssl_certificate /etc/ssl/private/wildcard.pem;
        ssl_certificate_key /etc/ssl/private/wildcard.pem;
        ssl_trusted_certificate /etc/ssl/certs/wildcard.ca;
        add_header Strict-Transport-Security 'max-age=15552000; includeSubDomains';
        ssl_protocols TLSv1.3 TLSv1.2;
        ssl_ciphers "EECDH+ECDSA+AESGCM EECDH+aRSA+AESGCM EECDH+ECDSA+SHA384 EECDH+ECDSA+SHA256 EECDH+aRSA+SHA384 EECDH+aRSA+SHA256 EECDH+aRSA+RC4 EECDH EDH+aRSA HIGH !RC4 !aNULL !eNULL !LOW !3DES !MD5 !EXP !PSK !SRP !DSS";
        ssl_session_cache shared:SSL:20m;
        ssl_session_timeout 1h;
        ssl_prefer_server_ciphers on;
        ssl_stapling on;
        ssl_stapling_verify on;
        # Timeouts
        client_max_body_size 50G;
        client_body_timeout 600s;
        proxy_read_timeout  600s;
        proxy_send_timeout  600s;
        send_timeout        600s;
        # Set headers
        proxy_set_header Host              $http_host;
        proxy_set_header X-Real-IP         $remote_addr;
        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
        # enable websockets
        proxy_http_version 1.1;
        proxy_set_header   Upgrade    $http_upgrade;
        proxy_set_header   Connection "upgrade";
        proxy_redirect     off;
        # DNS
        resolver 1.1.1.1 8.8.4.4 valid=300s;
        resolver_timeout 5s;

        # ==========  Alternative image download page BEGIN
        location ~* /(download-text)/.*.(gif|jpg|png)$ {
         add_header Content-disposition "attachment; filename=$1";
         types {}
         default_type application/octet-stream;
         expires -1;
        }

        location /download-text {
         alias /docker/sdnext/mnt-volume/data/outputs/text/;
         autoindex on;
         expires 1s;
         location ~* ^.+\.(jpeg|gif|png|jpg)
          {
           add_header Content-disposition "attachment";
          }
        }
        # ========== Alternative image download page END
        
        location / {
         proxy_pass http://127.0.0.1:7860;
        }
}

SD.Next on Linux

SD.Next btrfs deduplication example

in case SD.Next linux server got btrfs partition we might enjoy deduplication of the files and when we use a large number of models that deduplication may save some space

btrfs filesystem du --summarize --human-readable /docker/sdnext/
compsize -x /docker/sdnext/
jdupes --one-file-system --dedupe --recurse --size /docker/sdnext/

Execution example

root@server6:~ # btrfs filesystem du --summarize --human-readable /docker/sdnext/huggingface-volume
     Total   Exclusive  Set shared  Filename
 399.51GiB   399.51GiB       0.00B  /docker/sdnext/huggingface-volume

 root@server6:~ # compsize -x /docker/sdnext/huggingface-volume
Processed 526 files, 1860252 regular extents (1860252 refs), 48 inline.
Type       Perc     Disk Usage   Uncompressed Referenced
TOTAL       89%      357G         399G         399G
none       100%      175G         175G         175G
zlib        81%      181G         223G         223G


root@server6:~ # jdupes --one-file-system --dedupe --recurse --size /docker/sdnext/huggingface-volume

root@server6:~ # btrfs filesystem du --summarize --human-readable /docker/sdnext/huggingface-volume
     Total   Exclusive  Set shared  Filename
 399.51GiB   294.70GiB    36.56GiB  /docker/sdnext/huggingface-volume

 root@server6:~ # compsize -x /docker/sdnext/huggingface-volume
Processed 526 files, 1516941 regular extents (1861709 refs), 48 inline.
Type       Perc     Disk Usage   Uncompressed Referenced
TOTAL       89%      297G         333G         399G
none       100%      151G         151G         175G
zlib        80%      146G         182G         223G

Example run on only one folder. Result might be better on FS level. But just as an example

root@server6:~ # jdupes --one-file-system --dedupe --recurse --size /docker/sdnext/huggingface-volume
Scanning: 353 files, 527 items (in 2 specified)
  [SRC] /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-E1-Full/blobs/f1ecf4952c2076dc9ef09f54814a2d4b3f14bfa9
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/f1ecf4952c2076dc9ef09f54814a2d4b3f14bfa9
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/f1ecf4952c2076dc9ef09f54814a2d4b3f14bfa9
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/f1ecf4952c2076dc9ef09f54814a2d4b3f14bfa9

  [SRC] /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/cf6c4d78b3b4f575a8b296ee5fccf6ab4e14c686
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/cf6c4d78b3b4f575a8b296ee5fccf6ab4e14c686
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/cf6c4d78b3b4f575a8b296ee5fccf6ab4e14c686

  [SRC] /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/f99e44fb320546035ca153638f1f2276e0cd2e04
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/f99e44fb320546035ca153638f1f2276e0cd2e04
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/f99e44fb320546035ca153638f1f2276e0cd2e04

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/481c7b2b39771c44df6dd8d13ee12ed072d731b4a650bd092885d4d52db229ad
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/481c7b2b39771c44df6dd8d13ee12ed072d731b4a650bd092885d4d52db229ad

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/c0ef3a140898e228a3520c9adec60743d2e8e5b3d229651bb37f1a3921919f99
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/c0ef3a140898e228a3520c9adec60743d2e8e5b3d229651bb37f1a3921919f99

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/a451792c739c05bca4606190cc2dd16731411bac03b4cf6aacc5767321f857c9
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/a451792c739c05bca4606190cc2dd16731411bac03b4cf6aacc5767321f857c9

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/f93148bcc04052a169e1e49bfcf6125df6cf9bf243cb9c627da75266cf8e35c3
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/f93148bcc04052a169e1e49bfcf6125df6cf9bf243cb9c627da75266cf8e35c3

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/7e76e18d224531b8197a46231cb53daf7f2f6ca707130252becf933026ac4eea
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/7e76e18d224531b8197a46231cb53daf7f2f6ca707130252becf933026ac4eea

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/60ece61b46ecb3e6a5b705ea304bc97535317c2a
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/60ece61b46ecb3e6a5b705ea304bc97535317c2a

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-1.3B-Diffusers/blobs/2fd01c57dddd8ae386d518c69c087c8ba8c73804
  ====> /docker/sdnext/huggingface-volume/hub/models--Wan-AI--Wan2.1-T2V-14B-Diffusers/blobs/2fd01c57dddd8ae386d518c69c087c8ba8c73804

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Freepik--F-Lite/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--ostris--Flex.2-preview/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Freepik--F-Lite/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--ostris--Flex.2-preview/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Freepik--F-Lite/blobs/c8728bc3dca59d2616a2a594cbac3ddb9eb77d5b
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/c8728bc3dca59d2616a2a594cbac3ddb9eb77d5b
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/c8728bc3dca59d2616a2a594cbac3ddb9eb77d5b
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/c8728bc3dca59d2616a2a594cbac3ddb9eb77d5b
  ====> /docker/sdnext/huggingface-volume/hub/models--ostris--Flex.2-preview/blobs/c8728bc3dca59d2616a2a594cbac3ddb9eb77d5b

  [SRC] /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-2B-Text2Image/blobs/30dea2006604226b944972536b9fc1ae40e3ef39b844140641cc450af5490158
  ====> /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-14B-Text2Image/blobs/30dea2006604226b944972536b9fc1ae40e3ef39b844140641cc450af5490158

  [SRC] /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-2B-Text2Image/blobs/d90c625c136567ef62dc13d7fde0d54393a2b08e7400b3594d38c6a9ba12f897
  ====> /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-14B-Text2Image/blobs/d90c625c136567ef62dc13d7fde0d54393a2b08e7400b3594d38c6a9ba12f897

  [SRC] /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-2B-Text2Image/blobs/371ddccda7716967086fd98ef7e2f8e14f96f6ae
  ====> /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-14B-Text2Image/blobs/371ddccda7716967086fd98ef7e2f8e14f96f6ae

  [SRC] /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-2B-Text2Image/blobs/6c85dc465ff1a4392f21aeefb7ab6ea7b9b990a9
  ====> /docker/sdnext/huggingface-volume/hub/models--nvidia--Cosmos-Predict2-14B-Text2Image/blobs/6c85dc465ff1a4392f21aeefb7ab6ea7b9b990a9

on /docker/sdnext level it looks like

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Freepik--F-Lite/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/huggingface-volume/hub/models--ostris--Flex.2-preview/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-dev/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-Kontext-dev/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-schnell/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--lodestones--Chroma/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--ostris--Flex.2-preview/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--shuttleai--shuttle-3.1-aesthetic/blobs/ec87bffd1923e8b2774a6d240c922a41f6143081d52cf83b8fe39e9d838c893e

  [SRC] /docker/sdnext/huggingface-volume/hub/models--Freepik--F-Lite/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Dev/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Fast/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--HiDream-ai--HiDream-I1-Full/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/huggingface-volume/hub/models--ostris--Flex.2-preview/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-dev/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-Kontext-dev/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--black-forest-labs--FLUX.1-schnell/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--lodestones--Chroma/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--ostris--Flex.2-preview/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
  ====> /docker/sdnext/mnt-volume/models/Diffusers/models--shuttleai--shuttle-3.1-aesthetic/blobs/a5640855b301fcdbceddfa90ae8066cd9414aff020552a201a255ecf2059da00
...
Deduplication done (6565 files processed)

SD.Next on Linux

Edit SD.Next task que using sqlite3

Sometimes it is not possible to remove hanged task from SD.Next - it is instantly re-scheduled.

that is can be done by direct edit of sqlite3 database like shown below

# apt install sqlite3
sqlite3 /docker/sdnext/app-volume/extensions-builtin/sd-webui-agent-scheduler/task_scheduler.sqlite3

sqlite> .tables
app_state  task

sqlite> .schema task
CREATE TABLE task (
        id VARCHAR(64) NOT NULL,
        api_task_id VARCHAR(64),
        api_task_callback VARCHAR(255),
        name VARCHAR(255),
        type VARCHAR(20) NOT NULL,
        params TEXT NOT NULL,
        script_params BLOB NOT NULL,
        priority INTEGER NOT NULL,
        status VARCHAR(20) NOT NULL,
        result TEXT,
        bookmarked BOOLEAN,
        created_at DATETIME DEFAULT (datetime('now')) NOT NULL,
        updated_at DATETIME DEFAULT (datetime('now')) NOT NULL,
        PRIMARY KEY (id)
);


sqlite> .schema task
CREATE TABLE task (
        id VARCHAR(64) NOT NULL,
        api_task_id VARCHAR(64),
        api_task_callback VARCHAR(255),
        name VARCHAR(255),
        type VARCHAR(20) NOT NULL,
        params TEXT NOT NULL,
        script_params BLOB NOT NULL,
        priority INTEGER NOT NULL,
        status VARCHAR(20) NOT NULL,
        result TEXT,
        bookmarked BOOLEAN,
        created_at DATETIME DEFAULT (datetime('now')) NOT NULL,
        updated_at DATETIME DEFAULT (datetime('now')) NOT NULL,
        PRIMARY KEY (id)
);
sqlite> SELECT id, name, status  FROM task;
task(fhraih0ryzwm1iq)||done
task(93jw8bgegiaczkr)||pending
220fd698-05db-4224-8480-03ced279d906||pending
sqlite> DELETE FROM task WHERE id = '220fd698-05db-4224-8480-03ced279d906';

sqlite> .exit

SD.Next on Linux

SD.Next by liutyi docker compose

Personal preferences fork from Vlad SD.Next for Intel Arc iGPU system based on Intel Core Ultra products. Might be also compatible with All other HW that SD.Next is suppose to work on (but that is upstream project is better to be used). Fork is also

keeps standard theme as default.
keeps scheduler as built in plugin.
uses long filenames by default.
Have Dockerfile and containers that based on consumer Intel drivers not DC product one.
Keeps Intel Extension for Pytorch (ipex) compatibility (as long as it is relevant).

So there are 2 versions available

IPEX

Container: https://hub.docker.com/repository/docker/liutyi/sdnext-ipex/tags/25.11-ipex/
Source code: https://github.com/liutyi/sdnext/tree/ipex

PyTorch 2.7.1+xpu; IPEX 2.7.10+xpu. It was working better than Pytorch 2.7.1+xpu w/o ipex for iGPU. But might not be actually that better than 2.9.1+xpu. The only reason (at least it seems to) to continue use it is that it provides GPU memory usage verbose stats.

PyTorch

Docker container: https://hub.docker.com/repository/docker/liutyi/sdnext-ipex/tags/25.11-pytorch
Source code: https://github.com/liutyi/sdnext/tree/pytorch

Based on PyTorch 2.9.1+xpu

Docker-compose example

services:
  sdnext-ipex:
    container_name: sdnext-ipex
    # PyTorch 2.7.1+xpu and ipex 2.7.10+xpu for Intel
#    image: liutyi/sdnext-ipex:25.11-ipex
    # PyTorch 2.9.1+xpu for Intel
    image: liutyi/sdnext-ipex:25.11-pytorch
    restart: unless-stopped
    devices:
      # GPU
      - /dev/dri:/dev/dri 
      # NPU (optional)
      - /dev/accel:/dev/accel
    environment:
      - HF_DEBUG=True
      # To USE Nano Banana get key https://aistudio.google.com/api-keys
#     - GOOGLE_API_KEY=AIxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
    ports:
      - 7860:7860
    volumes:
      - /docker/sdnext/app-volume:/app
      - /docker/sdnext/mnt-volume:/mnt
      - /docker/sdnext/hf-volume:/root/.cache/huggingface
      - /dev/shm:/dev/shm

Test 47 - Allo theatre play cover

First drafts

ChatGPT

Starting with ChatGPT getting Posters like that for Play Allo

ChatGPT Image Oct 11, 2025, 01_10_36 PM.jpg

Also getting the prompt from ChatGPT for try this at home.

A dark, atmospheric theatrical poster for a Ukrainian play titled "АЛЛО". 
Two semi-transparent silhouettes, a man on the left and a woman on the right, face each other across the composition. 
Between them, a glowing vintage telephone handset floats vertically, connected by a wire, symbolizing distant communication. 
The background is textured, with subtle cracks and a moody bluish glow. 
The title "АЛЛО" is written in large, glowing Cyrillic letters at the top. 
Below, in simple thin font, three lines of text: "Бестіарій", "За п’єсою Сергія Марена", "Режисер Оленка Вахрамєєва". 
The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. 
Poster style, cinematic, minimalistic, modern graphic design.

Some at home generations are

Flux.2 Dev

Flux.2 Dev uint4

Qwen Image Pruning 13B

Flux.1 Dev SRPO

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "ALLO" is written in large, glowing letters at the top. Below, in simple thin font, three lines of text: "Bestiarii", "Serhiy Maren play", "Directed by Olenka Vahrameieva". The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man (30y.o) on the right and a woman (40y.o) on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "ALLO" is written in large, glowing letters at the top. Below, in simple thin font, three lines of text: "Bestiarii", "Play by Serhii Maren", "Directed by Olenka Vahrameieva". The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

HiDream I1 Full

HiDream + Chat GPT

WAN 2.2 T2I 14B

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Time: 57m 15.22s | total 3752.86 pipeline 3435.19 move 194.08 preview 68.90 vae 16.87 onload 15.91 offload 15.27 te 6.56 | GPU 38210 MB 32% | RAM 76.44 GB 61%

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a 30 y.o. man on the right and a 40 y.o. woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Enhance prompt

original

gemma-3-4b-it

qwen3-4b
Instruct-2507

Llama-3.2-3B
Instruct

A dark, atmospheric theatrical poster.
Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition.
Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication.
The background is textured, with subtle cracks and a moody bluish glow.
The title "АЛЛО" is written in large, glowing letters at the top.
The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere.
Poster style, cinematic, minimalistic, modern graphic design.

Dark, atmospheric theatrical poster featuring two semi-translucent humanoid figures – male silhouette positioned predominantly on the right side, exhibiting broad shoulders and sharply angled jawline; female silhouette situated on the left, characterized by flowing hair cascading down her back and delicate facial features – facing one another directly within the frame’s central axis. Both figures wear tailored charcoal grey Victorian frock coats extending past their knees, subtly revealing high collars adorned with intricate silver embroidery. A single, intensely saturated crimson vintage rotary telephone handset, rendered with visible scratches and aged brass detailing, levitates vertically between the subjects, illuminated from behind by a diffuse blue light source casting long shadows. The handsets coiled black cable extends seamlessly into the darkness beyond. Background consists of heavily distressed concrete texture overlaid with a pervasive, muted cyan luminescence emanating from hairline fractures spiderwebbing across the surface. At the apex of the composition, АЛЛО is emblazoned in colossal, dynamically lit neon lettering—a sharp, angular typeface radiating intense turquoise energy—dominating the upper portion of the poster. Overall impression evokes a sense of isolation, suspenseful anticipation, and melancholic beauty through minimalist execution and chiaroscuro lighting techniques reminiscent of film noir cinematography.

A dark, atmospheric theatrical poster depicts two semi-transparent humanoid figures centered along the vertical axis: a male silhouette on the right with broad shoulders and a sharply defined jawline, standing tall and composed; a female figure on the left, distinguished by long, flowing hair streaming downward over her back and refined, ethereal facial features. Both wear impeccably tailored charcoal gray Victorian frock coats that extend just below the knee, their high collars intricately embroidered with fine silver thread patterns that catch faint glimmers under ambient light. Between them, suspended mid-air like a spectral bridge, rests a single vintage rotary telephone handset — its once-polished brass now worn smooth with deep scratches and weathered patina — glowing with an eerie inner intensity due to illumination from behind. The handset emits a soft pulse of blood-red hue, contrasting starkly against the surrounding tones. From its base, a thick, coiled black cable snakes silently into the void beyond the frame, vanishing into shadowy depth. Behind the figures, the backdrop reveals cracked, uneven concrete surfaces textured with coarse grit and fissures, each crack pulsing gently with a dim, diffused cyan glow resembling bioluminescent veins spreading outward. This luminous network forms a subtle web across the wall-like expanse, creating a haunting sense of hidden connection beneath silence. At the very top center of the poster, the word АЛЛО dominates in massive, jagged neon letters — crafted in a sharp, futuristic angular font — emitting a radiant burst of electric turquoise light that seems to vibrate slightly, as though humming with latent power. The entire scene is bathed in dramatic chiaroscuro lighting: deep contrasts between bright highlights and profound blacks amplify tension, drawing focus to the interplay of form, space, and stillness. The mood is one of quiet solitude, unspoken dread, and bittersweet longing—an emotional resonance forged through minimalism, cinematic realism, and the timeless elegance of early 20th-century design fused with modern surrealism.

A dark, cinematic still life photograph depicting a dimly-lit, smoke-filled study, where a lone figure sits hunched over a cluttered wooden desk, surrounded by stacks of dusty leather-bound books and scattered papers.
The figure, a middle-aged man with a gaunt face and sunken eyes, wears a worn, brown tweed jacket with frayed cuffs and a faded waistcoat, his hands stained with ink and cigarette ash. His head is bowed forward, resting upon his upturned palms.
On the opposite side of the room, a large, ornate mirror hangs askew against the wall, its glassy surface reflecting the faint glow of a dying candle flame cast onto the adjacent bookshelf. Shelves overflow with ancient tomes bound in cracked leather, some lying open to reveal yellowed pages filled with handwritten notes and cryptic symbols.
In the foreground, a battered, metal typewriter sits atop a stack of crumpled paper, its keys smeared with dried bloodstains. A half-empty glass of whiskey rests beside the machine, next to a folded piece of paper bearing a scribbled message scrawled in red ink.
Soft, diffused light pours in through the grimy windowpane above, illuminating the space with an eerie, otherworldly ambiance. Shadows dance across the walls, accentuating the air of foreboding and desperation that pervades this forsaken sanctuary.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Prompt: Dark, atmospheric theatrical poster featuring two semi-translucent humanoid figures – male silhouette positioned predominantly on the right side, exhibiting broad shoulders and sharply angled jawline; female silhouette situated on the left, characterized by flowing hair cascading down her back and delicate facial features – facing one another directly within the frame’s central axis. Both figures wear tailored charcoal grey Victorian frock coats extending past their knees, subtly revealing high collars adorned with intricate silver embroidery. A single, intensely saturated crimson vintage rotary telephone handset, rendered with visible scratches and aged brass detailing, levitates vertically between the subjects, illuminated from behind by a diffuse blue light source casting long shadows. The handsets coiled black cable extends seamlessly into the darkness beyond. Background consists of heavily distressed concrete texture overlaid with a pervasive, muted cyan luminescence emanating from hairline fractures spiderwebbing across the surface. At the apex of the composition, АЛЛО is emblazoned in colossal, dynamically lit neon lettering—a sharp, angular typeface radiating intense turquoise energy—dominating the upper portion of the poster. Overall impression evokes a sense of isolation, suspenseful anticipation, and melancholic beauty through minimalist execution and chiaroscuro lighting techniques reminiscent of film noir cinematography.

Prompt: A dark, atmospheric theatrical poster depicts two semi-transparent humanoid figures centered along the vertical axis: a male silhouette on the right with broad shoulders and a sharply defined jawline, standing tall and composed; a female figure on the left, distinguished by long, flowing hair streaming downward over her back and refined, ethereal facial features. Both wear impeccably tailored charcoal gray Victorian frock coats that extend just below the knee, their high collars intricately embroidered with fine silver thread patterns that catch faint glimmers under ambient light. Between them, suspended mid-air like a spectral bridge, rests a single vintage rotary telephone handset — its once-polished brass now worn smooth with deep scratches and weathered patina — glowing with an eerie inner intensity due to illumination from behind. The handset emits a soft pulse of blood-red hue, contrasting starkly against the surrounding tones. From its base, a thick, coiled black cable snakes silently into the void beyond the frame, vanishing into shadowy depth. Behind the figures, the backdrop reveals cracked, uneven concrete surfaces textured with coarse grit and fissures, each crack pulsing gently with a dim, diffused cyan glow resembling bioluminescent veins spreading outward. This luminous network forms a subtle web across the wall-like expanse, creating a haunting sense of hidden connection beneath silence. At the very top center of the poster, the word АЛЛО dominates in massive, jagged neon letters — crafted in a sharp, futuristic angular font — emitting a radiant burst of electric turquoise light that seems to vibrate slightly, as though humming with latent power. The entire scene is bathed in dramatic chiaroscuro lighting: deep contrasts between bright highlights and profound blacks amplify tension, drawing focus to the interplay of form, space, and stillness. The mood is one of quiet solitude, unspoken dread, and bittersweet longing—an emotional resonance forged through minimalism, cinematic realism, and the timeless elegance of early 20th-century design fused with modern surrealism.

Prompt: A dark, cinematic still life photograph depicting a dimly-lit, smoke-filled study, where a lone figure sits hunched over a cluttered wooden desk, surrounded by stacks of dusty leather-bound books and scattered papers. The figure, a middle-aged man with a gaunt face and sunken eyes, wears a worn, brown tweed jacket with frayed cuffs and a faded waistcoat, his hands stained with ink and cigarette ash. His head is bowed forward, resting upon his upturned palms. On the opposite side of the room, a large, ornate mirror hangs askew against the wall, its glassy surface reflecting the faint glow of a dying candle flame cast onto the adjacent bookshelf. Shelves overflow with ancient tomes bound in cracked leather, some lying open to reveal yellowed pages filled with handwritten notes and cryptic symbols. In the foreground, a battered, metal typewriter sits atop a stack of crumpled paper, its keys smeared with dried bloodstains. A half-empty glass of whiskey rests beside the machine, next to a folded piece of paper bearing a scribbled message scrawled in red ink. Soft, diffused light pours in through the grimy windowpane above, illuminating the space with an eerie, otherworldly ambiance. Shadows dance across the walls, accentuating the air of foreboding and desperation that pervades this forsaken sanctuary.

Cosmos-Predict2 14B

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a 30yo man in t-shirt on the right and a 40yo woman in a black dress on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Other prompts

QWEN	HiDream	Cosmos	WAN
Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Felix) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 720x1280\| Seed: 42\| CFG scale: 5\| App: SD.Next\| Version: f71db69\| Pipeline: QwenImagePipeline\| Operations: txt2img\| Model: Qwen-Image	Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Felix) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 720x1280\| Seed: 42\| CFG scale: 5\| App: SD.Next\| Version: f71db69\| Pipeline: HiDreamImagePipeline\| Operations: txt2img\| Model: HiDream-I1-Full	Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Felix) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 720x1280\| Seed: 42\| CFG scale: 5\| App: SD.Next\| Version: f71db69\| Pipeline: Cosmos2TextToImagePipeline\| Operations: txt2img\| Model: Cosmos-Predict2-14B-Text2Image	Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Felix) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 720x1280\| Seed: 42\| CFG scale: 5\| App: SD.Next\| Version: f71db69\| Pipeline: WanPipeline\| Operations: txt2img\| Model: Wan2.2-T2V-A14B-Diffusers
Kolors		Flux.2 Dev
Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Tymofiy) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 1024x1024\| Seed: 2589760219\| CFG scale: 6\| App: SD.Next\| Version: 196988c\| Pipeline: KolorsPipeline\| Operations: txt2img\| Model: Kolors-diffusers Time: 4m 10.60s \| total 406.34 pipeline 244.02 callback 152.61 decode 6.52 te 2.85 gc 0.28 \| GPU 19966 MB 16% \| RAM 31.83 GB 25%	Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Tymofiy) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font. Parameters: Steps: 50\| Size: 720x1280\| Seed: 2589760219\| CFG scale: 6\| App: SD.Next\| Version: 196988c\| Pipeline: KolorsPipeline\| Operations: txt2img\| Model: Kolors-diffusers Time: 3m 52.89s \| total 459.23 pipeline 228.57 callback 138.35 preview 86.62 decode 4.28 te 1.09 gc 0.28 \| GPU 19490 MB 15% \| RAM 28.45 GB 23%

tempestByVlad_baseV01 [8bfad17222]

$2025-10-14-02036-tempestByVlad_baseV01-720x1280-Seed42-CFG3.5-AG0-STEP20.jpg$

Prompt: Dramatic, high-contrast theater poster for the play 'ALLO' (Hello). The composition features a stark, bisected visual: two separate, dimly lit, brutalist concrete bunkers, one containing a man (Tymofii) and the other a woman (Margo), both holding vintage rotary phones to their ears. The only connection is a single, glowing, neon red telephone cord that stretches across the dark space, symbolizing hope and love against isolation. The style should be gritty and cinematic, reminiscent of high-contrast noir photography with a post-apocalyptic, sci-fi atmosphere. Minimalist text placement for the title 'ALLO' in a clean, sans-serif font.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man (30y.o) on the right and a woman (40y.o) on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a 30 y.o. man on the right and a 40 y.o. woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "ALLO" is written in large, glowing letters at the top. The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Prompt: A dark, atmospheric theatrical poster. Two semi-transparent silhouettes, a man on the right and a woman on the left, face each other across the composition. Between them, a vibrant red vintage telephone handset floats vertically, slightly backlighted, connected by a black wire, symbolizing distant communication. The background is textured, with subtle cracks and a moody bluish glow. The title "АЛЛО" is written in large, glowing letters at the top. Below, in simple thin font, three lines of text: "Бестіарій", "Автор: Сергій Марен", "Режисер: Оленка Вахрамєєва". The overall color palette is dark, teal and black, with soft lighting and an enigmatic, dramatic atmosphere. Poster style, cinematic, minimalistic, modern graphic design.

Time: 4m 39.02s | total 520.22 pipeline 275.01 callback 141.31 preview 96.73 decode 3.97 prompt 2.87 gc 0.26 | GPU 10172 MB 8% | RAM 28.75 GB 23%

More prompts from ChatGPT

Surreal Connection	Minimalist Symbolism	*Duality & Silence*
A surreal theatrical poster for the Ukrainian play "АЛЛО". In the center, a cracked concrete wall with a glowing fissure shaped like a telephone handset, radiating soft light. On the left side of the crack, cold blue shadows with abstract human silhouettes leaning in, on the right side warm greenish light with other faint silhouettes. The title "АЛЛО" appears at the top in large glowing Cyrillic letters, distressed texture. At the bottom, thin minimalist typeface with three lines of text: 1) "Бестіарій" 2) "За п’єсою Сергія Марена" 3) "Режисер Оленка Вахрамєєва". Dark, symbolic, high-contrast poster design.	A minimalistic poster for the Ukrainian play "АЛЛО". A single old rotary telephone sits in the center on cracked concrete floor, illuminated by a spotlight in an otherwise dark underground room. The phone cord splits into two glowing threads, extending left and right into shadowy emptiness. The background fades into textured darkness with subtle mist. The word "АЛЛО" is written above in bold, glowing Cyrillic letters. At the bottom in thin modern font: 1) "Бестіарій" 2) "За п’єсою Сергія Марена" 3) "Режисер Оленка Вахрамєєва". Cinematic, moody, symbolic design.	A theatrical poster for the Ukrainian play "АЛЛО". The composition is split vertically into two halves: on the left side a faded silhouette of a man in cool blue tones, on the right a silhouette of a woman in warm green tones. Between them hangs an old telephone receiver glowing faintly, symbolizing fragile communication. Background: textured dark concrete wall with scratches and cracks. The title "АЛЛО" in large bold Cyrillic letters glows faintly at the top. At the bottom, three lines of text in very thin minimalist font: 1) "Бестіарій" 2) "За п’єсою Сергія Марена" 3) "Режисер Оленка Вахрамєєва". Minimal, moody, dramatic poster.
ChatGPT

Gemini

GROK

hunyuan-image

Tempest by Vlad

Kolors (4m 15.07s GPU 20442 MB 17% \| RAM 48.46 GB 39%)

Flux.1 Krea Dev 12B (Time: 18m 22.46s \| GPU 26322 MB 22% \| RAM 68.46 GB 55%)

Qwen Image 20B

HiDream I1 14B (Time: 40m 50.94s \| GPU 50506 MB 42% \| RAM 92.39 GB 74%)

Wan 2.2 14B

Cosmos Predict 2 14B

Flux.2 Dev uint4

Flux.2 Dev

SD.Next

Test 26 - Create Ai book logo

Artificial intelligence

System Info

SD.Next supported models

List of Models by Size

Generation Examples

Stable Diffusion 3.5 Large+ class models

Wan2.2-T2V-A14B

Wan2.1-T2V-14B

HiDream I1 Full🪪

HiDream I1 Dev🪪

HiDream I1 Fast🪪

HiDream E1 Full 🪪

Qwen-Image

Cosmos-Predict2 T2I 14B 🪪

shuttle-3.1-aesthetic

FLUX.1-dev 🪪

FLUX.1-Krea-dev 🪪

FLUX.1-schnell 🪪

AuraFlow v0.3

Wan2.2-TI2V-5B

OmniGen2

CogView 4

Kandinsky 3

Wan2.1-T2V-1.3B

Stable Diffusion 3.5 L 🪪

Chroma 36

chroma-unlocked-v48-detail-calibrated

SD.Next Config

Dockerfile variables

SD.Next on Linux

SD.Next openvino docker compose

Install

Dockerfile

example without proxy on branch master

example with proxy

docker-compose.yml

Application screenshots

Post-install fixes

add some python libraries

Fix access to gated repo

Access error to gated repo

login to huggingface manually

System info

openvino (float32) note

Reinstall

Clean reinstall

Preserve downloaded models

Config example for iGPU

Example image generation on iGPU

Stable Diffusion 3.5 medium

PixArt-Sigma-XL-2-512-MS

SD1.5

Config example for CPU

Example image generation on CPU

Kolors

Flux.1 Krea

Use nginx for SD.Next

install nginx

nginx config example

SD.Next btrfs deduplication example

Edit SD.Next task que using sqlite3

SD.Next by liutyi docker compose

IPEX

PyTorch

Docker-compose example

Test 47 - Allo theatre play cover

First drafts

ChatGPT

Flux.2 Dev

Flux.2 Dev uint4

Qwen Image Pruning 13B

Flux.1 Dev SRPO

HiDream I1 Full

HiDream + Chat GPT

WAN 2.2 T2I 14B

Enhance prompt

original

gemma-3-4b-it

qwen3-4b
Instruct-2507

Llama-3.2-3B
Instruct