Zhenzhen Doubao SeedEdit2.0

SeedEdit 2.0 is Doubao's instruction-based image editor - the "change the chicken leg into a burger" school of editing where you describe the change in a sentence and the model figures out the pixels. ComflySeededit is the Comfly pack's paid client for it, and it belongs to the same family as Qwen-Image-Edit and gpt-image: you hand it a picture, tell it what to change, and get an edited image back. No masks, no ControlNet, no local weights - just a prompt, a picture, and a per-call fee. For quick "make this car red" or "remove the watermark" style jobs, it's hard to beat the convenience.

The README is genuinely helpful here - it includes a recipe list for writing edit prompts that still holds up:

• Add/remove entities: "add a rainbow / delete the girl in the picture"
• Change an object: "change the chicken leg to a burger"
• Change style: "make it comic style"
• Change color: "make the clothes pink"
• Change expression/action: "make him cry / laugh"
• Change background: "change the background to the seaside / under the stars"

Inputs that matter

• image - required. This is the thing being edited.
• prompt - required, multiline. Use the recipes above; vague prompts give vague edits.
• scale - 0–1, default 0.5. How strongly the edit adheres to your instruction versus the source image. 0.5 is a reasonable middle; crank it for bigger changes.
• seed - -1 for random.
• add_logo - a Comfly-side watermark stamp: logo_position, logo_language (中文/英文), logo_text, logo_opacity. Off by default.

Outputs: edited_image (IMAGE), response, and image_url.

Install

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Or ComfyUI Manager → "Comfyui_Comfly" → restart. Nothing runs locally - credits and key from https://ai.comfly.chat (Comflyapi.json or the node's apikey).

Gotchas

The README's image constraints are real and they bite: JPEG/PNG/BMP, under 4.7 MB, under 4096×4096, and - the one that gets people - the long side can't be more than ~3× the short side, or the API errors out. Extreme panoramas or ultra-tall strips will fail; crop them first. And like every instruction-editor, SeedEdit re-emits the whole frame, so pixels you never asked about will shift - that's the model's nature, not a node bug. This is the 2.0 node; if you want the 3.0 model with a size field and guidance scale, that's Comfly_Doubao_Seededit in the same pack.