Zhenzhen Doubao Seedream4.0
Seedream 4.0 with multi-image reference and 4K on tap
- image1
- image2
- image3
- image4
- image5
- image
- response
- image_url
This is the flagship image node in the Comfly pack's Doubao lineup: Seedream 4.0, with up to 4K resolution, up to five reference images, and a batch mode that can hand you fifteen images from one call. Like everything in this pack it's a paid API client - the model runs on ByteDance's hardware, you pay per image, and nothing touches your GPU. If you want Seedream's aesthetic at high resolution with character/style references and you're okay with a metered bill, this is the node.
Inputs that matter
- prompt - required, multiline.
- model - string, default
doubao-seedream-4-0-250828. - resolution - 1K (default), 2K, or 4K. Higher resolution means a bigger bill and longer waits - 1K to iterate, 4K for finals.
- aspect_ratio - 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, 21:9, 9:21, or
Custom. ChooseCustomand thewidth/heightfields (64–8192) take over. - image1 … image5 - reference images. Wire one and the prompt becomes a reimagination/reference job; several lets you blend.
- max_images - 1–15 images per run. The README is blunt: the API bills per image, so more images = more money. Keep it at 1 until you know what you want.
- sequential_image_generation -
disabled(default) orauto; auto paces the batch so the model isn't stampeded. - stream - off by default; on, it streams partial output. Nice for long 4K batches, but it can complicate a plain graph.
- seed - -1 for random.
- watermark - default on. Flip it off for clean output.
Outputs: image (IMAGE), image_url, and response.
Install
The usual:
cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly
Or ComfyUI Manager → "Comfyui_Comfly" → restart. No models download; credits + key at https://ai.comfly.chat (Comflyapi.json or the node's apikey).
Gotchas
The watermark default-on strikes again - check it before queueing a 4K batch, because re-running fifteen 4K images because of a watermark is an expensive mistake. max_images is where people accidentally spend real money; it defaults to 1, keep it there while tuning. Custom sizing via aspect_ratio: Custom + width/height is a README-documented late addition, so older example workflows won't have it - set it yourself when you need exact canvases. And this node is image generation with references, not an instruction editor - for "change X to Y" jobs, use the SeedEdit nodes instead.
Inputs (19)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| model | STRING | doubao-seedream-4-0-250828 | — |
| response_format | COMBO | url | 2 options: url, b64_json |
| resolution | COMBO | 1K | 3 options: 1K, 2K, 4K |
| aspect_ratioopt | COMBO | 1:1 | 10 options: 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, +4 |
| widthopt | INT | 102464–8192 | — |
| heightopt | INT | 102464–8192 | — |
| apikeyopt | STRING | — | |
| image1opt | IMAGE | — | |
| image2opt | IMAGE | — | |
| image3opt | IMAGE | — | |
| image4opt | IMAGE | — | |
| image5opt | IMAGE | — | |
| sequential_image_generationopt | COMBO | disabled | 2 options: disabled, auto |
| max_imagesopt | INT | 11–15 | — |
| seedopt | INT | -1-1–2147483647 | — |
| watermarkopt | BOOLEAN | true | — |
| streamopt | BOOLEAN | false | — |
| skip_erroropt | BOOLEAN | false | 开启后,节点失败时不报错、按旧行为返回默认空结果;关闭时(默认)失败直接抛出错误。 |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| response | STRING | — |
| image_url | STRING | — |