CatsAPI Seedance 2.0 Mini

The Seedance 2.0 Mini node is the cheaper half of the pack's ByteDance video pair. Seedance is ByteDance's text-to-video line, and like every ByteDance flagship it's API-only - no weights, ever. This node gets you the Mini variant in ComfyUI: a 480p/720p ceiling, 4 to 15 seconds, and a smaller coin bill than the full Seedance 2.0 node sitting right next to it in the same menu.

Why you'd reach for it

When Seedance first showed up in ComfyUI as an API node, the loudest community reaction was about the price - "extremely expensive" was the top comment, and the reply was that you were just seeing the real cost of these renders. Mini is ByteDance's answer to that: the same family at a friendlier price point for prototyping, social clips, and storyboards. If you've got a local Wan or LTX setup, this isn't a replacement - it's the closed-model option for shots your open weights can't pull, paid in per-run coins instead of GPU time.

How it works

The same wrapper machinery as the rest of the pack: cost preview, submit, poll every three seconds, download the mp4 to ComfyUI/output/catsapi/, return the path. Two code-level details are worth knowing. First, the node always sends inputMode=reference, and Mini deliberately sends no mode parameter - the full Seedance 2.0 node sends mode=fast; Mini doesn't. Second, those three image inputs - start_image, end_image, reference_images - are merged in the order start → references → end and sent as reference material only. They steer the look; they do not give you strict first-frame or last-frame control.

The inputs that matter

• prompt - the action and the look.
• resolution - 480p (default) or 720p. That's the entire range; Mini never goes higher.
• duration - 4 to 15 seconds, default 8.
• aspect_ratio - 16:9, 9:16, 4:3, 3:4, 1:1, 21:9.
• max_coins - the spending gate; 0 = unlimited.
• start_image / end_image / reference_images - up to 4 images total across all three. Feed more and you get a hard error, not a truncation. The site's schema says 9, but its worker keeps only 4, so the node enforces the real limit for you.
• api_key_override - leave empty unless you're on a hosted platform like RunningHub that can't set env vars.

Outputs

video_path (the mp4 on disk), cost_coins, task_id, metadata. The mp4 lands in ComfyUI/output/catsapi/ with a seedance20Mini_... filename.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/maodeyu180/ComfyUI_Catsapi.git

Restart, then:

export CATSAPI_API_KEY=cats-your-key
python main.py

It'll also read ~/.catsapi.env. No pip dependencies to worry about.

Common issues

• "Too many images" - you combined start + refs + end beyond 4. Trim your inputs; that's the effective worker limit, not the node being pedantic.
• First frame not matching your start image - expected. These are references, not hard keyframes; if you need strict pinning, Seedance isn't the tool.
• Cost creep - 720p and longer durations multiply the bill. Keep the default 480p/8s while you're learning and check cost_coins after each run.
• Slow jobs - a video task can poll for minutes. It's not hung; the client gives up at 15 minutes and reports the last status.

This is a young, self-published pack with essentially zero Reddit footprint, so your due diligence is reading the small source and never sharing a workflow with api_key_override filled in - the value persists into the workflow JSON.