Seed Volcengine I2I Config

If Seedream's T2I variant is the "generate from nothing" door, SeedVolcengineI2IConfig is the "change an image you already have" door to the same ByteDance API. It's the I2I sibling of the T2I node: same three doubao-seedream models, same 1K–4K size range, same watermark and seed controls - plus an image input that gets JPEG-encoded and sent to the API as the source.

The whole thing is still a config node: it outputs a MEDIAHUB_IMAGE_REQUEST for MediaHubGenerateImage to execute. Nothing renders until the execution node runs with your ARK key.

Inputs that matter

• image - the source IMAGE. Converted to base64 JPEG (quality 95) and shipped with the request.
• prompt - what to change. With I2I, the prompt describes the edit or restyle rather than a whole new scene.
• model - doubao-seedream-5-0-260128 (flagship), 4-5-251128, or 4-0-250828.
• size - 1K–4K output. The 4K ceiling applies here too, and it's the main reason to pick this over other providers' I2I nodes for high-res edit jobs.
• watermark - off by default.
• seed - -1 = random; set a value for reproducible edits.
• response_format - url / base64.

Output: request → MediaHubGenerateImage.

How the edit works

This is img2img in the classic sense - the source image goes in as pixels, not as a "character reference" like MiniMax's subject_image. Seedream recomposes from the prompt with the input as context. That means it's the right tool for restyling, background replacement, and "make it X" edits, and the wrong tool if you need a specific identity preserved - that's the MiniMax subject-reference job. Knowing which provider model treats your image as context versus as an anchor is half of using this pack well.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; under MediaHub/Provider/Seed Volcengine/Image. Dependencies: requests, Pillow, numpy. You need a Volcengine ARK key (ByteDance's console - it's the same key the Seedance video nodes use).

Common gotchas

The JPEG round-trip matters more on I2I than T2I because the source is your content: very clean source images in, cleaner edits out. And if you're editing at 4K, remember both input encoding and output rendering happen at full size - give the request a moment. Same pipeline rule as always: config node alone produces nothing, and if the edit drifts too far from the source, reduce the strength of the change in your prompt rather than hunting for a node setting - this node deliberately keeps that dial simple.