Seedance 2.0 Mini · Image to Video

Mini is where the Seedance 2.0 line gets affordable. $0.0246078 per second - call it two and a half cents - for image-to-video at 480p or 720p, 4 to 15 seconds, sound included. The full 2.0 node charges $0.07/s for the same job with 1080p and 4K on top; Mini skips the top resolutions and takes roughly a third of the money.

For a lot of work that's the right trade. Drafts, social cuts, anything that ends up on a phone screen: 720p is plenty, and three takes at Mini prices cost less than one at full. Where you notice the difference is fine detail - faces at distance, small text, fast motion - and that's a resolution-and-model-quality question, so test on your own material rather than trusting either of our opinions.

What's on the node

image_url is an IMAGE socket, so Load Image connects straight in, and the pack uploads the picture for you. prompt is the direction - the author's framing is "subject, action, camera, light, sound, and constraints", which is a good checklist when output comes back generic.

aspect_ratio defaults to adaptive and can be forced to 16:9, 9:16, 4:3, 3:4, 1:1 or 21:9. resolution is 480p or 720p, full stop. duration_seconds is the Fast-line dropdown: -1 at the top for smart duration (the model picks the length), then 4 to 15 - and the billed quantity either way.

generate_audio is on by default and, on this tier, doesn't change the price. generate_audio off saves nothing, so leave the sound in.

enable_web_search is the interesting extra on the Mini nodes and it's off by default. Switch it on and the model searches the web for prompt-relevant information before rendering - genuinely useful when a shot has to depict something specific and real, and a surcharge when it doesn't. The published numbers: 480p goes from $0.0246078/s to $0.036/s, 720p from $0.05292/s to $0.072/s. Worth it for a landmark or a product; not worth it for a mood piece.

Then the three presentation switches that the 2.0 line all share: return_last_frame (also outputs the final frame as an IMAGE, the clean way to hand a shot to the next one), bitrate_mode (standard or high, a bigger, crisper delivered file), and watermark. Optional last_image_url closes the shot on an image you supply.

seed starts on fixed control-after-generate: queue the same graph twice and ComfyUI reuses the first result instead of buying a second one. Randomize on purpose, not by accident, because every run is billed.

Outputs

video → Save Video or a preview node. last frame → any IMAGE input, populated only with return_last_frame on. task id → a string; it's the task's id in the SpicyAPI console and it's what a support request needs.

Install

Manager → search SpicyAPI → Install. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

Restart, then Settings > SpicyAPI → paste a key from spicyapi.ai/console/keys (on a server, set SPICY_API_KEY instead), then restart once more so the node list matches your account. No pip installs, no weights, no GPU needed - the work happens on SpicyAPI's servers.

The realistic failure list

No key → "No SpicyAPI key is set". No headroom → the run stops with a message quoting the worst-case cost above your cap; raise the cap in Settings if you meant it. A 40004 means no deployment serves that combination and names the offending field. A request_id in the error is support currency. And a cancelled run is a failed wait, not a cancelled job - it finishes server-side, gets charged if it succeeds, and its result stays in your console for up to 14 days.