CatsAPI Seedream 5 Lite
ByteDance's image model without the GPU
- reference_image
- images
- file_paths
- cost_coins
- task_id
- metadata
The Seedream 5 Lite node puts ByteDance's Seedream 5 image model inside your ComfyUI graph for the price of a CatsAPI key and a few coins. Seedream is ByteDance's flagship image line and it has never shipped weights - this node is how you run it next to your local nodes without owning a data center. Lite is the entry tier of the Seedream 5 pair; its bigger sibling is the Pro node in this same pack. Same inputs, same outputs, different bill.
Why you'd reach for it
ByteDance runs the sharpest two-track posture in this space: components open, products closed. Everything it charges for - the whole Seedream line included - stays on the API, so the only way a Seedream 5 image enters a ComfyUI workflow is through a wrapper like this one. You'd grab it when you want that closed-model look mid-graph: a hero shot to upscale, a draft to composite, a style reference for a character pass - and you don't mind per-run coins or your prompt leaving the machine. If you've sworn off paid APIs, this isn't for you; if you already use the official ComfyUI Partner Nodes, think of this as the same idea billed through a different storefront.
How it works
Standard pack wrapper: cost preview → submit → poll → download → tensor. Your prompt goes up, the finished PNGs come down to ComfyUI/output/catsapi/, and the node stacks them into an IMAGE tensor you can wire straight into Preview Image. Reference images ride along as base64 PNGs, up to four of them.
The inputs that matter
prompt- multiline; the whole job.image_size-square(default), plus portrait/landscape 4:3 and 16:9. That's your only size control: Seedream 5 takesimageSizeand nothing else, so there's deliberately noresolutionorqualityfield to go hunting for.num_images- 1 to 4 results per run.seed- the one input this node has that most of its packmates don't.-1(default) means "let the API pick"; any non-negative value is forwarded for reproducibility. If you're iterating on a prompt, set a seed so you can actually compare outputs.max_coins- spending cap; 0 disables it.reference_image- up to 4 images. Over that is a hard error, not a silent drop.api_key_override- leave empty locally.
Outputs
images (the IMAGE tensor), file_paths (JSON list of the downloaded PNG paths), cost_coins, task_id, metadata.
Installing
cd ComfyUI/custom_nodes
git clone https://github.com/maodeyu180/ComfyUI_Catsapi.git
Restart ComfyUI and set the key:
export CATSAPI_API_KEY=cats-your-key
python main.py
~/.catsapi.env also works. Nothing to pip install.
Common issues
- Different results every run - you left
seedat -1. Set one. - "Too many reference images" - a batch of more than 4 into
reference_imageerrors out on purpose. - Can't find a quality/resolution knob - there isn't one;
image_sizeis the whole menu. - Key rejected as truncated - any key containing
...is refused before it's sent; paste the complete one.
And the usual wrapper honesty: this pack is the service's own integration with essentially no Reddit footprint, so don't paste your key into a shared workflow (api_key_override persists into the JSON), and remember every run is a paid call to a server you don't control.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| image_size | COMBO | square | 5 options: square, portrait_4_3, landscape_4_3, portrait_16_9, landscape_16_9 |
| num_images | INT | 11–4 | — |
| max_coins | INT | 00–100000 | — |
| reference_imageopt | IMAGE | — | |
| api_key_overrideopt | STRING | — | |
| seedopt | INT | -1-1–2147483647 | — |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| file_paths | STRING | — |
| cost_coins | INT | — |
| task_id | STRING | — |
| metadata | STRING | — |