ComfyUI Presets • Download Status Dashboard

This is newest FLUX 2 Turbo LoRA preset. It is working with amazing speed and quality. If you get out of RAM error, add --no-cache to your backend args. Native resolution is 2048x2048 and it takes like 40 seconds on RTX 5090. This preset has better art quality compared to realism version.

This is newest FLUX 2 Turbo LoRA preset. It is working with amazing speed and quality. If you get out of RAM error, add --no-cache to your backend args. Native resolution is 2048x2048 and it takes like 40 seconds on RTX 5090. This preset has better art quality compared to realism version. This preset has lower quality and uses lower RAM and VRAM due to GGUF.

This is newest FLUX 2 Turbo LoRA preset. It is working with amazing speed and quality. If you get out of RAM error, add --no-cache to your backend args. Native resolution is 2048x2048 and it takes like 40 seconds on RTX 5090. This preset has better realism compared to HQ version.

This is newest FLUX 2 Turbo LoRA preset. It is working with amazing speed and quality. If you get out of RAM error, add --no-cache to your backend args. Native resolution is 2048x2048 and it takes like 40 seconds on RTX 5090. This preset has better realism compared to HQ version. This preset has lower quality and uses lower RAM and VRAM due to GGUF.

This is a FLUX 2 Quality 2 tier. The difference vs Quality 1 is that, this one uses 2x faster sampler. That is why. The difference is minimal. Moreover you can increase step count to 40 for more quality.

This is a FLUX 2 Quality 2 tier. The difference vs Quality 1 is that, this one uses 2x faster sampler. That is why. The difference is minimal. Moreover you can increase step count to 40 for more quality.

FLUX Dev Models - Krea and SRPO as well: 20-40 Steps - Just change base model to FLUX Dev or FLUX Krea or FLUX SRPO. You can use GGUF models as well

Official FLUX Kontext Dev Edit Any Image

Use this preset to inpaint small areas as shown in tutorial video. Just directly apply when doing inpainting. It will only change inpainting Init Image related parameters. Set step count accordingly. For example when Init Image Creativity is set to 0.6, you need 7 steps do to actually 4 steps. E.g. Qwen Fast LoRA requires 4 steps. So set steps count to 7 in this case.

This is a preset to do outpainting on images with FLUX Fill Dev model as shown in tutorial. Currently FLUX Fill Dev model is the very best to outpaint images without changing rest of the image. Outpaint images with small portions if you get errors. Try multiple different seeds if you don't get perfect results.

This is Qwen Image 2512 Model at 2656x2656 pixel realism preset and it makes extremely high-quality realistic images on trained Qwen Image models like self-trained Qwen Image LoRA or Qwen Image Fine Tuned model. This is 4 steps base + 4 steps upscale config. This significantly increases the realism with a little bit loss of fantastic prompts capability. You can increase the used realism LoRA strength to 0.8 for even further realism. 0.6 is working really good and balanced. Last updated 2026 January 1.

This is Qwen Image 2512 Model at 2656x2656 pixel realism preset and it makes extremely high-quality realistic images on trained Qwen Image models like self-trained Qwen Image LoRA or Qwen Image Fine Tuned model. This is 4 steps base + 4 steps upscale config. This significantly increases the realism with no loss of complex prompting capability. Last updated 2026 January 1.

This is fast Qwen Image Realism generation with Qwen Image 2512 model. Only 8 steps and CFG 1. It uses newest res_2s Sampler and beta57 Scheduler for realism. res_2s is 2x slower than euler but much better realism. With 2026 January 1, this preset massively upgraded.

This is ultra fast Qwen Image generation with Qwen 2512 model. It uses newest LoRA. Last updated at 2026 January 1.

This is fast Qwen Image Edit newest model 25-11 (like FLUX Kontext) with 12 steps. It supports up to 3 input images and this is an amazing model. We are using Qwen-Image-FP8-Lightning-4steps-V1.0-fp32 and it is working perfect to speed up.

This is fast Qwen Image Edit newest model 25-11 (like FLUX Kontext) with 12 steps. It supports up to 3 input images and this is an amazing model. We are using Qwen-Image-FP8-Lightning-4steps-V1.0-fp32 and it is working perfect to speed up.

This is fast Qwen Image Edit newest model 25-11 (like FLUX Kontext) with 12 steps. It supports up to 3 input images and this is an amazing model. We are using Qwen-Image-FP8-Lightning-4steps-V1.0-fp32 and it is working perfect to speed up.

This is Qwen Image Edit 2511 - 2656x2656 pixel realism preset and it makes extremely high quality realistic images. This preset is made for training your subject on Qwen Image Edit 2511 model. It yields excellent realism and quality and prompt following. The very best so far. Last update 2026 January 1.

This is Qwen Image 2656x2656 pixel realism preset and it makes extremely high-quality realistic images on trained Qwen Image models like self-trained Qwen Image LoRA or Qwen Image Fine Tuned model. This is 4 steps base + 4 steps upscale config. This significantly increases the realism with a little bit loss of fantastic prompts capability. You can increase the used realism LoRA strength to 0.8 for even further realism. 0.6 is working really good and balanced.

This is a preset to generate stylized UHD images with your Qwen trained models whether it is Qwen Image base or 2512 or Qwen Image Edit Plus 2509 or 2511 model. This preset does 12 steps base and 10 steps upscale into 2656x2656 pixels. This is slower than Tier 2 but higher quality.

This is a preset to generate stylized UHD images with your Qwen trained models whether it is Qwen Image base or 2512 or Qwen Image Edit Plus 2509 or 2511 model. This preset does 12 steps base and 4 steps upscale into 2656x2656 pixels.

SDXL Models Best Settings With Realism Negative Prompt

Wan 2.1 FusionX & Self Forcing LoRA Text to Video
8-10 Steps

Wan 2.1 Image To Video Base Model: 20-50 Steps

Wan 2.1 LightX2V LoRA Image to Video: 4-10 Steps

Wan 2.1 Text To Video Base Model: 20-50 Steps

Wan 2.2 Image to Video No-LoRA Higher Quality Improved Workflow - 20 Steps or More. 16 FPS, 81 frames and first 4 frames trimmed. This is highest quality Image To Video. You can increase steps count if you wish for even better quality but it will be slow. RIFE 2x Frame interpolation enabled by default thus 5 seconds 32 FPS generated

Wan 2.2 Text to Video No-LoRA Higher Quality Improved Workflow - 20 Steps or More. 16 FPS, 81 frames and first 4 frames trimmed. This is highest quality Text To Video. You can increase steps count if you wish for even better quality but it will be slow. RIFE 2x Frame interpolation enabled by default thus 5 seconds 32 FPS generated

Generate amazing realistic images with Wan 2.2 - extremely fast for default 1280x720p (can do excellent 1920x1080p as well) resolution and realistic. Type prompt above <refiner>

Extremely efficient Wan2.2 Image To Video with A14B-Moe-Distill-Lightx2v base model. 4 steps only and fast. Increase to 8 steps if you want better quality. No LoRA needed. Default settings are 16 FPS and 81 frames with RIFE 2x frame interpolation enabled. Thus output is 5 seconds at 32 FPS.

Latest Wan 2.2 Text to Video lightning config. Uses latest speed up LoRAs with best samplers and schedulers setup. Only 4 steps total. You can increase step count to 8 if you wish. 16 FPS - 81 Frames default. RIFE 2x Frame interpolation also enabled by default so output is 32 FPS - 5 seconds.

This is top quality Z Image Turbo preset. It does 9 steps and 1.25 upscale with 9 more steps. Thus this is highest quality. It generates 1920x1920 pixel total resolution.

This is fast Z Image Turbo preset. It does 9 steps. It is a fast and high quality and generates 1536x1536 pixel total.