Ernie Image-to-Image

This is an image-to-image redrawing application based on ERNIE-Image.

After the user uploads a reference image, the workflow first scales the image according to a set aspect ratio, encodes it into latents via VAE encode, combines it with positive and negative prompts into the sampler for generation, and finally decodes and outputs the image. Unlike regular text-to-image, this workflow follows a standard image-to-image redrawing pipeline with a very clear core structure: LoadImage → Resize → VAEEncode → KSampler → VAEDecode → SaveImage. It is better suited for style upgrades, detail enhancement, and image optimization while preserving the main subject and composition of the original image.

This workflow is suitable for:

• Portrait retouching
• Illustration redrawing
• Product image optimization
• Poster enhancement
• Image quality and style upgrades while preserving the original composition

Recommended usage:
Upload an image with a clear subject and definite composition; use positive prompts to describe your desired style, texture, and visual effect; use negative prompts to constrain unwanted elements. The default sampling parameters for this current workflow are steps=20, cfg=4, denoise=0.85, which belongs to a noticeably modified redrawing scheme, making it more suitable for "style upgrades" and "obvious optimization".

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