Ernie Inpainting

is an Inpaint application based on ERNIE-Image.

After a user uploads an image and draws a mask on the area to be modified, the workflow first synchronously scales the image and mask according to a set ratio, encodes them into latents via VAE, and writes the mask into the latent noise mask. Inpaint is performed exclusively within the selected local region, and the final result is decoded and output. Unlike standard image-to-image, this flow does not redraw the entire image; instead, it achieves localized region control through SetLatentNoiseMask + DifferentialDiffusion, making it ideal for specified area replacement, local restoration, and local detail optimization.

This workflow is suitable for:

• Changing hairstyles, accessories, or local clothing details
• Replacing a specific local object in the scene
• Local outpainting and local repair
• Modifying only a specific section without affecting the overall composition

Recommended usage:
First upload the original image, then paint a mask over the area you want to modify. The positive prompt should directly describe "what this area should become"; the negative prompt is used to suppress unwanted elements. This current flow defaults to steps=20, cfg=4, denoise=1, which is a relatively thorough Inpaint solution, ideal for clearly replacing the contents of the masked area.  

🎁Claim RH Coins first, then try out the workflow! Click your avatar in the top right corner → Invitation Code → Enter [rh-v1111] to instantly receive 1000 RH Coins, and log in daily to claim another 100 coins~ For more ComfyUI workflows, tutorials, and gameplay, welcome to follow our WeChat Official Account (AIKSK), with simultaneous updates on Douyin / Bilibili / Xiaohongshu / YouTube (AI-KSK)✨