WebUI Bridge Image Input

Img2img in ComfyUI is fiddly in a way that's easy to forget: it's not one thing, it's load an image, encode it to latent, pick a denoise, and remember to wire the mask if you're inpainting. WebUI Bridge Image Input packages that whole chore into a single node with a WebUI-style upload, and it hands you the three things you actually need: the image, the mask, and the denoise value as separate outputs.

The Bridge main node has img2img built into its sidebar too, but this standalone version is the one you reach for when you want the image flow visible on the canvas. It's the pack's "图生图/局部重绘" (img2img/local redraw) node.

How it works

You upload an image (and optionally a separate mask) inside the node - they're stored as string references to files. The node loads the pixels, and does something clever with masks: if you don't provide a mask but your image has an alpha channel, it uses the alpha as the mask. If you're in img2img mode with no mask at all, it emits an empty mask - the image still flows, the mask just does nothing. In inpaint mode you're expected to bring a mask.

The outputs mirror the main node's: image, mask, denoise (a float bounded to 0–1, default 0.55), mode (the string "img2img" or "inpaint") and status telling you what loaded. Those outputs wire straight into the rest of the img2img chain: image → VAE Encode (or into WebUI Bridge Inpaint Conditioning), denoise → your KSampler's denoise input so you don't have to duplicate the number in two places.

Inputs and outputs

• image - required, the upload.
• mask - optional; falls back to the image's alpha channel, or empty in img2img mode.
• mode - img2img or inpaint.
• denoise - 0.55 default. The single most important knob: 0.45-ish keeps composition, higher drifts further from the source. The pack's minimal img2img tutorial uses 0.45.
• Outputs: image, mask, denoise, mode, status.

How to install

Part of the ComfyUI-WebUI-Prompt-Bridge pack. ComfyUI-Manager → ComfyUI WebUI Prompt Bridge, or:

cd ComfyUI/custom_nodes
git clone https://github.com/dianfangsihuo/ComfyUI-WebUI-Prompt-Bridge.git

Restart ComfyUI. No models needed for this node itself.

Common issues

The classic beginner failure: you upload an image, hit generate, and get a brand-new unrelated image. That means your sampler's latent_image isn't coming from this node - upload alone doesn't touch the pipeline, you have to actually encode it. The pack's docs say it plainly: check that the KSampler's latent is driven by a VAE Encode or the Bridge's Inpaint Conditioning node. Second gotcha: don't confuse this with the main node's "read prompt from image" feature. That reads prompt metadata from a PNG/JPEG/WebP and needs no VAE or sampler at all; this node feeds image pixels into generation. They're two different flows that happen to both involve picking a file.