ComfyUI YSC HighRes-Fix

A custom node for ComfyUI designed to simplify the "High-Resolution Fix" process. This node takes an input image, performs a direct upscale, and then re-encodes it into a latent image, preparing it for a final sampling pass to add details.

What It Does

This node replicates the functionality of a high-resolution fix pipeline. Instead of just resizing an image, it upscales it and then uses a diffusion sampling process (denoising) to add fine details that are consistent with the original image's content.

How It Works

The node's operation is straightforward:

1. Image Input: You provide a low-resolution image (e.g., the output of your initial generation).
2. Brute Upscale: The node applies a simple, direct upscale to the image to increase its dimensions.
3. Latent Encoding: This larger, upscaled image is then encoded back into the latent space.
4. Image sampling: This latent image will be denoised by a ksampler-like method, generating a greater and better image.
5. Latent Output: The node outputs this new, larger latent image.

How to Use (Example Workflow)

1. Generate your initial image (e.g., 512x512) as you normally would.
2. Connect the output image from your VAE Decode node to the image input of the YSC HighRes-Fix node.
3. Connect the model, VAE, positive and negative prompt and a upscale model to the respective input of YSC HighRes-Fix node.
4. Crucially, set the denoise on this second KSampler to a value like 0.3 to 0.55. This value controls how much detail is added. I recommend do not use high values (0.7+) because the image will be broken.
5. Adjust the upscale range (1 to 4x upscale). For low VRAM users, I recommend upscaling up to 2 times using this node.
6. Connect the LATENT output of this node to a final VAE Decode node to get your high-resolution image.
7. Optional: After generating your upscaled image, you can increase even more the resolution using Ultimate SD Upscale

Inputs

• image: The source image (from a VAE Decode or Load Image node) that you want to upscale and refine.
• checkpoint model: A checkpoint used to generate images. This one will be used by the embedded ksampler of the node.
• positive and negative prompts: For quality handling
• VAE: For the embedded VAE encoder of the node.
• Upscale model: Used to make the first raw upscale.

Outputs

• LATENT: The upscaled and encoded latent image, ready for a VAE decode.

Installation

1. Navigate to your ComfyUI/custom_nodes/ directory.
2. Clone this repository:

   git clone https://github.com/yannickcruz/ComfyUI-YSC-HighResFix.git
   

   (Replace the URL with your actual repository URL)
3. Alternatively, download the .py files from this repository and place them directly in the ComfyUI/custom_nodes/ directory.
4. Restart ComfyUI.

You can now install this node via ComfyUI Manager!

Credits

• This node was possible thanks to the code of the TheBill2001's comfyui-upscale-by-model, that inspired me to build the first raw upscale of this node.