Seedream 4 Multi-Image Blending

Composition by text is hard. "Combine the lighting of image A with the subject of image B" is the kind of instruction models fumble - but Seedream 4 Multi-Image Blending is built for exactly that. You feed it two to four images, describe what you want in the resulting single image, and the model merges them: style from one, subject from another, color from a third. It's the dedicated sibling of the unified Seedream 4 node's blend mode, with the inputs pinned open so you don't need the auto-adding slots. Since Seedream 4 is API-only (see the KB's ByteDance panel - the whole Seedream line stays on the paid API), this node is the ComfyUI front end for that cloud call.

This is a genuinely useful tool for character and style work. Want your OC in another image's aesthetic? Reference a character shot plus a style frame. Designing something on a consistent palette? Blend two reference images and describe the scene. Where local IP-Adapter-and-ControlNet stacks would take six nodes and careful tuning, this is one node and one prompt - at the cost of paying per blend and having less granular control.

How it works

Each connected image is base64-encoded, and the node sends them all as an image array with your prompt to the Seedream 4 endpoint. The sequential_image_generation flag is forced off - this mode deliberately produces a single merged output rather than a sequence. Two images are required; three and four are optional. The result is downloaded and returned as a normal IMAGE tensor.

Inputs and outputs that matter

• image1 and image2 - required references. Who you want and the look you want, typically.
• image3 / image4 - optional extra references, up to four total.
• prompt - how the merge should resolve: what's foreground, what's background, what should dominate.
• width / height - output size, default 2048×2048.
• seed - reproducibility, -1 for random.

Output: one IMAGE tensor, the blended result. Save it or pipe 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, set your BytePlus API key in config.ini or BYTEPLUS_API_KEY. Manager: search "ComfyUI Seed API Integration".

Where people get burned

The usual cloud failure mode: a blank black image means the pack swallowed an error - read the console for the real cause (key, quota, model permission). On the craft side, remember the model can't read your mind about which trait comes from which image; if the prompt doesn't assign roles, it picks. "Style from the first image, keep the subject from the second" beats "merge these." And don't expect a clean cut-out paste - this is a generative blend, so anatomy and geometry can drift. Treat the seed like your best friend: the same four inputs and a fixed seed will give you reproducible blends to iterate from. Cheap enough to explore, at two impressions this node is the pack's quiet sleeper.