Zhenzhen Doubao Seedream3.0
Doubao Seedream 3.0 text-to-image, rented by the call
- image
- response
Doubao Seedream 3.0 is ByteDance's flagship text-to-image model, and this node is the Comfly pack's thin, paid client for it. There's no local model to download, no VRAM involved - you type a prompt, the job goes to Doubao's API through the Comfly relay, and an image comes back. If you've been seeing Seedream results everywhere and want them inside your ComfyUI graph without the GPU cost, this is the on-ramp. The catch, as with the whole pack: every image is a transaction, and you need credits at https://ai.comfly.chat.
Inputs that matter
- prompt - required, multiline.
- model - a string, default
doubao-seedream-3-0-t2i-250415. You can type a newer dated model id when Doubao bumps the line; the README keeps a changelog of these. - size - presets from 1024x1024 up to 1512x648, plus
Custom. Pick a preset for speed; pickCustomand fill inCustom_size(format matters:1280x720, width×height) when you need an exact canvas. The node validates the format and complains if you skip thex. - guidance_scale - 1–10, default 2.5. This is your prompt-adherence dial: higher clamps harder to the prompt, lower leaves room for interpretation.
- response_format -
url(default) orb64_json. URL is lighter; b64 is self-contained if you're scripting around it. - seed - -1 for random, or set for reproducibility.
- watermark - default on. Flip it off if you don't want the Doubao watermark baked in. (Worth checking on every run - the default is the surprising direction.)
Outputs: image (IMAGE) and response (the raw API payload).
Install
The usual Comfly drill:
cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly
Or ComfyUI Manager → "Comfyui_Comfly" → restart. No models download - it's a thin HTTP client. Key goes in Comflyapi.json or the node's apikey override.
Gotchas
The watermark default-on is the one that trips people - your first few "perfect" generations come back stamped and you have to re-run. Check it before you queue. Also note this node is the 3.0 line; the pack ships a separate Comfly_Doubao_Seedream_4 node with resolution tiers and multi-image reference, so if you came here wanting 4K output or image-to-image, you're on the wrong node. And since it's a relay, keep an eye on the response output when something looks off - the task log at ai.comfly.chat will tell you what the API actually said, which beats guessing.
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| model | STRING | doubao-seedream-3-0-t2i-250415 | — |
| response_format | COMBO | url | 2 options: url, b64_json |
| size | COMBO | 1024x1024 | 9 options: 1024x1024, 864x1152, 1152x864, 1280x720, 720x1280, 832x1248, +3 |
| Custom_size | STRING | 1536x1024 | — |
| guidance_scale | FLOAT | 2.51–10 | — |
| apikeyopt | STRING | — | |
| seedopt | INT | -1-1–2147483647 | — |
| watermarkopt | BOOLEAN | true | — |
| 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 (2)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| response | STRING | — |