Skip to content

greenzorro/comfyui-workflow-upscaler

Repository files navigation

comfyui-workflow-upscaler

🇬🇧 EN | 🇨🇳 中文

ComfyUI workflows for upscaling.

Updates

  • 9/21/2024 Use lora to add details
  • 4/24/2024 Created
  • 5/9/2024 Changed model to Dreamshaper 8
  • 5/20/2024 Use Ultimate Upscaler instead of Ksampler
  • 5/21/2024 Changed model to Dreamshaper XL Lightning; Use Anyline as ControlNet instead of ControlNet sd1.5 tile
  • 5/22/2024 Multi-step upscaling for better result

🔍 Sharpening upscaler

⏬ Sharpening upscaler

Add details to an image to boost its resolution. Only one upscaler model is used in the workflow.

🎨 Creative upscaler

⏬ Creative upscaler

Add more details with AI imagination. The output looks better, elements in the image may vary.

Low denoise value for latent image and ControlNet to keep the composition.

When it does not work for you, try adjusting ControlNet weight and denoise.

Examples

Source image:

Upscaled image:

Source image:

Upscaled image:

Prerequisite

Custom nodes

Node Link
ComfyUI_UltimateSDUpscale https://github.com/ssitu/ComfyUI_UltimateSDUpscale.git
ComfyUI-Anyline https://github.com/TheMistoAI/ComfyUI-Anyline.git

Models

Dir Model Link
checkpoints DreamShaperXL_Lightning.safetensors https://huggingface.co/Lykon/dreamshaper-xl-lightning/resolve/main/DreamShaperXL_Lightning.safetensors?download=true
controlnet mistoLine_rank256.safetensors https://huggingface.co/TheMistoAI/MistoLine/resolve/main/mistoLine_rank256.safetensors?download=true
upscale_models 4x-UltraSharp.pth https://huggingface.co/philz1337x/upscaler/resolve/main/4x-UltraSharp.pth?download=true
embeddings negativeXL_D.safetensors https://civitai.com/api/download/models/134583
loras add-detail-xl.safetensors https://civitai.com/api/download/models/135867?type=Model&format=SafeTensor

Where to run?

Besides running locally, you may run them on the cloud without extra configuration, using my notebook code:

Thanks

Special thanks to clarity-upscaler which inspired my workflow! Clarity-upscaler is based on A1111 and has not offered ComfyUI workflow yet(update:now it has). I played with the models it uses and somehow created my own simplified alternative. I have to say that mine is not as good as the original clarity-upscaler, but it works for me.