Seedance 2.0 · Text to Video

Text-to-video on the full Seedance 2.0 model: no image socket, no references, one multiline prompt in and a rendered clip with sound out, 4 to 15 seconds, in 480p through 4K. From $0.07 per second.

The distinguishing feature versus the Mini and Fast nodes is the resolution ladder - 4K is only on the full 2.0 and the Spicy variant - plus the web-search switch, which on this model has a price attached that you should see before you flip it.

The cost table, because this is the whole decision

At 480p the rate is $0.07/s. It's $0.152/s at 720p, $0.374/s at 1080p and $0.778/s at 4K. Audio doesn't change any of those numbers - generate_audio is priced in, so leave it on.

Now enable_web_search:

| Resolution | Standard | With web search | | --- | --- | --- | | 480p | $0.070/s | $0.114/s | | 720p | $0.152/s | $0.246/s | | 1080p | $0.374/s | $0.612/s | | 4k | $0.778/s | $1.248/s |

That's roughly a 60% premium and the author says so on the field itself: requests with it on are billed at the web-search rate. It earns its keep when a shot has to depict something real and specific - a particular vehicle, a landmark, a product - and it's money burnt on anything mood-driven. Off by default, and it should stay that way until you need it.

The rest of the controls

prompt is a multiline box and the author's checklist is the good one: subject, action, camera, light, sound, constraints. Sound is included because this model renders native audio in the same pass - a described effect is an instruction.

aspect_ratio covers 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 and adaptive (the default). With no source image, adaptive means the model's judgement; choose a real ratio when the output has a destination.

duration_seconds is the Fast-line style dropdown: -1 for smart duration, where the model picks the length, then 4 through 15. The length you get is what you pay for.

return_last_frame also emits the clip's final frame as an IMAGE - the honest way to continue a sequence into a second node. bitrate_mode picks standard or high delivery (bigger, crisper file, same generation). watermark adds the model's visible "AI generated" mark.

seed is fixed after generate, so a repeated queue of the same graph reuses the cached result instead of buying a second render.

Outputs

video → Save Video. last frame → an IMAGE socket, only populated with return_last_frame on. task id → the string that identifies the run in your console and in any support conversation.

Install

ComfyUI 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 (or set SPICY_API_KEY if this is a remote box). Restart once more so the node list loads for your account. Zero dependencies, zero model downloads, no GPU: the render happens on SpicyAPI.

Getting it to run without surprises

Set Settings > SpicyAPI > Max cost per run first. Every run is quoted before submission, and anything worse than your ceiling stops cleanly - with a text-to-video node at 4K that guardrail is the difference between a hobby and a bill. Interrupting doesn't cancel: the task keeps running server-side, is charged if it succeeds, and its result stays in your console for up to 14 days. 40004 means the combination you chose isn't served, and the message names the field. If a model you saw on the site isn't in your menu, restart ComfyUI - the catalogue is fetched at startup.