Seedance 2.0 Mini · Image to Video

Mini is the draft tier of the Seedance 2.0 family: 480p or 720p, with sound, at the lowest per-second rate of the three. If you're the sort of person who renders nine versions of a five-second shot before committing, this node saves you real money over Fast or the full 2.0.

It's not a toy, either. Resolution and price are the things Mini gives up, not the feature set - it keeps the smart-duration option, the closing-frame input, the watermark switch, and the last-frame output.

How it works

Standard anatomy for this pack: the node turns your widgets into a request, uploads the source image (PNG, re-encoded to JPEG only if it can't fit the 10 MiB ceiling), quotes the job so you see the worst case before anything is held, submits it with an idempotency key, polls while showing the estimate and elapsed time, then downloads and decodes the MP4 into a ComfyUI VIDEO. Uploads are cached against your key for 12 hours, so changing the prompt doesn't re-send the picture.

The inputs that matter

image_url - the source image; the tooltip phrases it as one you're authorised to process, which is the vendor being careful rather than you being restricted. A batch sends its first image.

prompt - subject, action, camera, light, sound, constraints. That last word earns its place: negative phrasing works poorly on these models, so state constraints as things that are present ("clean studio backdrop, sharp focus") rather than things that aren't.

duration_seconds - 4 to 15, or -1 for smart duration where the model picks the length itself. Smart duration is great for a shot that shouldn't be padded and terrible for predicting a bill; the quote covers you.

resolution - 480p or 720p. Start at 480p for concept passes; 720p is the ceiling here, no matter what you type.

aspect_ratio - adaptive by default so it follows your image, plus 16:9, 9:16, 4:3, 3:4, 1:1, 21:9.

enable_web_search - lets the model search the web for context before rendering, and requests with it on are billed at the web-search rate. Even on Mini that's a real mark-up. Off unless the prompt genuinely needs a fact.

generate_audio - on by default, and at this tier the per-second rate doesn't change with it in the shipped table. Keep the sound.

return_last_frame - off by default; the last frame output is blocked until you turn it on.

bitrate_mode - standard or high. watermark - off. seed - fixed, so re-queueing won't bill you twice.

last_image_url - optional closing frame. Useful for landing a shot exactly where the next one starts.

Outputs

video, last frame (an IMAGE, only with return_last_frame on), and task id. Chaining two shots through last frame → image_url is the cleanest workflow in this pack, and Mini is a cheap place to rehearse it.

Installing it

Manager → PopNow → Install → restart. Or:

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

Key from popnow.ai/api/console/keys into Settings > PopNow or POPNOW_API_KEY; restart again so the catalog loads. Nothing else to install - no weights, no extra Python.

Where people get burned

The commonest self-inflicted wound is leaving Mini on after the concept works. It's the cheap tier; it looks like the cheap tier, especially on faces and fast motion. Promote the winning take to Fast or 2.0 for the final.

Web search is the quiet cost multiplier, and it's on all four Mini nodes, so it's easy to set once and forget.

And the hosted-model ground rules: the content filter is ByteDance's, so a refusal isn't something a different pack fixes; cancelling stops the wait but not the render; and Settings > PopNow > Max credits per run is the only real guard against a session that costs more than you intended.