豆包Seed 2.1多模态文本

Every once in a while a ComfyUI node makes you say "wait, it can do that?" This one takes an image, a video, or audio - or all three - and returns text analysis from ByteDance's doubao-seed-2.1-turbo multimodal model. It's the closest thing in this pack to a general-purpose vision-and-media interpreter: caption an image, transcribe or summarize audio, describe a video clip, answer questions about any of them. And no, the name isn't clickbait - this really is the Seed 2.1 model that ByteDance keeps as a closed API product (the same family as Seedream/Seedance, which the KB's ByteDance panel covers: closed flagship models, API-only).

The mechanism is a thin HTTP client over Volcengine's ARK platform. The node builds a request to the ARK "Coding Plan" endpoint (https://ark.cn-beijing.volces.com/api/coding/v3/responses), packages your text plus any attached image/video/audio as base64 content parts, sends it with your API key as a Bearer token, and extracts the text from the response. Notably it uses the OpenAI-compatible /responses API (that's why the class is called "Responses"), not the raw chat-completions endpoint. It's marked as always-changed (IS_CHANGED returns nan), so every run hits the API fresh instead of being cached.

The inputs

• api_key - a password field. The tooltip gives the key hint: leave it empty and it falls back to the ARK_API_KEY environment variable. So you can keep the key out of the workflow JSON entirely - which is the right call, because a workflow JSON is shareable and a key in it is a leak.
• model - default doubao-seed-2.1-turbo, editable string.
• 自定义提示词预设 - a system prompt (multiline). This is where you'd say "you are a fashion critic" or "describe only the background."
• 文本 - your question or instruction, default "请分析输入内容。" ("analyze the input content").
• Optional 图像 (IMAGE), 视频 (VIDEO), 音频 (AUDIO) inputs - the media to analyze.

Output: one STRING, text.

The catch you have to know

This needs a real ARK API key from Volcengine (the ByteDance cloud), and the "Coding Plan" endpoint implies a specific paid plan with that API. The KB's external-API-nodes doc is blunt about this whole category: an API-wrapper node that carries a credential is exactly the shape of thing that's been weaponized before in ComfyUI. This node is open source and only sends your key to Volcengine's official endpoint, so it's not malicious - but don't paste your key into shared workflows, and check the model name before you trust that a random workflow is pointing at the endpoint you think it is.

Install

From comfy_Pond_Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes && pip install -r requirements.txt

Needs requests (already in requirements). Honest take: this is a genuinely useful node for captioning pipelines and batch analysis, but it's the least "just works" node in the pack because it depends on your ARK account, region, and plan. If you already have ByteDance ARK access, it's a gem - a multimodal text engine inside ComfyUI that needs no local model. If you don't, the setup friction is on Volcengine's side, not this node's.