SeedEdit 3 Image-to-Image

SeedEdit is ByteDance's instruction-following editor: you hand it an image, tell it what to change, and it makes the edit - recolor, restyle, add, remove - without you touching a mask. This node (seededit-3-0-i2i-250628) is the ComfyUI wrapper for it, and in a world where local editing usually means mask pipelines and inpainting, this is refreshingly direct. It's the "describe the change" school of editing: "change the sky to dusk," "turn this photo into a watercolor." It keeps everything else about the image intact and returns a tensor you can keep working with.

There's a place for both approaches, and knowing when to use which is the real skill. SeedEdit is excellent at global and semi-global changes - style, palette, scene dressing - and it's one call, no mask to fumble. Local surgical edits with a precise mask are still sometimes better on a local inpainting model, where you control exactly which pixels change and don't pay per attempt. This is the paid path; ByteDance's whole SeedEdit line is API-only, per the KB's ByteDance panel.

How it works

Your image is base64-encoded and sent with your prompt and a guidance scale to the images endpoint. The size field has an adaptive option that keeps your original dimensions - the thing you'll want most of the time, since editing shouldn't change the canvas. Higher guidance_scale means stricter prompt adherence; lower gives the model more creative freedom. Result comes back as a URL, is downloaded, and becomes a normal IMAGE tensor.

Inputs and outputs that matter

• prompt - the edit instruction. Specific beats vague: "replace the red car with a blue vintage truck," not "change the car."
• image - the source. Required.
• size - adaptive (keeps source dimensions) or one of nine fixed aspect presets.
• guidance_scale - default 5.5, range 1–20. Push up when the model ignores you.
• seed - reproducibility, -1 for random.

Output: one IMAGE tensor with the edited image. Save it, upscale it, feed it onward.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/FloyoAI/ComfyUI-Seed-API
cd ComfyUI-Seed-API
pip install -r requirements.txt

Restart, put your BytePlus API key in config.ini or as BYTEPLUS_API_KEY. Manager: search "ComfyUI Seed API Integration".

Where people get burned

The blank-black-image fallback again - if the node returns black, the real error is in the console, and it's almost always key/quota/permission related. On the craft side, the classic mistake is a vague prompt: SeedEdit is an editor, not a mind reader, so say what stays and what changes. And because guidance_scale maxes at 20, cranking it won't turn a weak prompt into a strong edit - fix the prompt first. Finally, "adaptive" size is your friend for edits; the fixed presets will resize your image, which you usually don't want mid-edit. Paid per call, and on a 20% click-through, people clearly find this one worth the curiosity - it just needs the right expectations set.