Nodes/ComfyUI-Jimeng-API/Jimeng Seedream 3
ComfyUI Node

Jimeng Seedream 3

The old reliable image node, gently on its way out

By fkxianzhou·Created 12 months ago·Updated 7 days ago· 82
Jimeng Seedream 3
  • client
  • image
  • response
prompt
guidance_scale5.0
size
width1024
height1024
seed0
generation_count1
watermarkfalse

Seedream 3 is the pack's plainest image node: type a prompt, get an image. No reference inputs, no group generation, no web search - it's the 2025 workhorse of ByteDance's image line, and the pack ships it because plenty of people still have workflows built on it. You'd reach for it if you already have a prompt pipeline tuned for Seedream 3 and just want it to keep running.

The important thing to know is the caveat from the README: the underlying model, doubao-seedream-3-0, is in the official retirement pipeline. The platform is winding down its quota, and eventually the service goes away. The pack marks the node deprecated and points you to Seedream 5 as the replacement. Nothing about the node is broken today; it's just running on borrowed time. If you're starting fresh, skip it.

As with everything in this pack, there's no local model - you need a paid Volcano Ark API key and the image is generated on ByteDance's servers.

How it works

Straightforward request/response: your prompt goes to the fixed doubao-seedream-3-0-t2i model with your generation settings, the API renders, and the node returns the image. There's no model_version dropdown because there's only one model behind this node. It's the simplest image node in the pack for that reason.

The inputs that matter

  • prompt - the whole ballgame, since there's nothing else to condition on.
  • guidance_scale - 1.0 to 10.0, default 5. How hard the model sticks to your prompt. Push up for literal prompts, down if you want the model to breathe.
  • size - a preset combo (the recommended sizes), with width and height as manual overrides up to 8192px.
  • seed - from -1 (random) up to 2^31-1. Set a fixed seed for reproducible results.
  • generation_count - up to 2048; batch a whole grid of images in one go.
  • watermark - off by default; turn on if your use case needs the platform watermark.

Outputs

  • image - the generated image as a ComfyUI IMAGE.
  • response - the raw API response JSON, useful when you need the task details or want to debug.

Installing and the thing to watch

Standard pack install: ComfyUI Manager (search "ComfyUI Jimeng API") or git clone https://github.com/fkxianzhou/ComfyUI-Jimeng-API into custom_nodes, restart. The volcengine-python-sdk[ark] dependency installs itself. The usual pack quirks apply - legacy UI only, ByteDance content moderation on prompts.

The node-specific watch is the deprecation: expect quota to shrink and errors to creep in as the platform retires doubao-seedream-3-0. The README's advice is direct - move to Seedream 5, which gives you the same prompt-to-image flow plus reference images and group generation. Seedream 3 was great for its time; this is the part where you gently migrate.

CategoryJimengAI

Inputs (9)

NameTypeDefaultDescription
clientJIMENG_CLIENT
promptSTRING
guidance_scaleFLOAT5.01–10
sizeCOMBO9 options: 1024x1024 (1:1), 864x1152 (3:4), 1152x864 (4:3), 1280x720 (16:9), 720x1280 (9:16), 832x1248 (2:3), +3
widthINT10241–8192
heightINT10241–8192
seedINT0-1–2147483647
generation_countINT11–2048
watermarkBOOLEANfalse

Outputs (2)

NameTypeDescription
imageIMAGE
responseSTRING