Skip to content

Seed

Template
Seed, in the ComfyUI template browser under BytePlus
Workflow file
Seed.json
BytePlus nodes
BytePlus LLM
Keys needed
ModelArk API key
Also uses
Load Image, Preview as Text (built into ComfyUI)

Text. The template sends an image (the ComfyUI logo) and a prompt to Seed 2.0 Pro, which describes the subject, the setting, the colors and the mood and suggests a short title. Change the prompt to ask anything else about the picture.

The Seed template: Load Image connected to image_1 of the BytePlus LLM node, whose text goes to Preview as Text
Three nodes: Load Image, BytePlus LLM and Preview as Text.
  1. Load Image supplies the picture.
  2. BytePlus LLM receives it as image_1 together with the prompt “Describe this image in detail: the subject, the setting, the colors and the mood. Then suggest a short title for it.” The model is Seed 2.0 Pro.
  3. Preview as Text shows the STRING output. The node also has a raw_json output that is not connected.
  1. Add your ModelArk API key in Settings → BytePlus if you have not yet (API keys).
  2. Open the template browser and choose Seed under BytePlus.
  3. Press Run. The answer appears in Preview as Text.
  • prompt: ask for what you need, for example alt text, a list of objects or a caption in a given tone.
  • model: the list also has Seed 2.0 Lite and Mini, Seed 2.1 Turbo, DeepSeek V4.1 Flash and GLM 5.3 Flash. They take the same image input.
  • system_prompt: an advanced input. The template sets a short assistant role; change it to set a style or a language for the answer.
  • reasoning_mode and reasoning_effort: the template uses auto and medium. Raise the effort for harder questions; it makes the model think longer.
  • detail: the image detail level sent with each image; the template uses high, and low is the other choice.
The text answer in Preview as Text: a description of the ComfyUI logo with subject, layout, colors and mood
Example result.

ModelArk bills each request by the tokens it uses. The seed control is set to randomize, so every run asks again and bills again. Set it to fixed and ComfyUI reuses the cached result of an unchanged workflow without billing.