Wan 2.6 Flash · Image to Video

Five, ten or fifteen seconds, 720p or 1080p, off one opening frame, for $0.0225/second - about a quarter of what the full Wan 2.6 image-to-video node charges for the same length. The catch is in the feature list, not the quality: Flash doesn't generate a soundtrack. The clip comes back silent.

That makes it the iteration node. Draft motion at 5 seconds, get the shot right, then decide whether the finished piece needs real audio and move up a tier or leave it silent for an edit you're scoring anyway.

Where it fits

Wan 2.6 is one of the API-only Wans - 2.2 is the last release with downloadable weights - so this is the same deal as every 2.6 node in the pack: nothing loads locally, the frame and the job go over the network, and an MP4 comes back as a ComfyUI VIDEO. No VRAM, no checkpoint, works on a laptop. The cost of that convenience is per-second billing.

The one judgement call worth making up front: if the shot needs generated dialogue or ambience, this node can't give it to you and the full 2.6 node can. If it needs your own music or voiceover - which is most commercial work - silence is fine, and you're paying less for it.

Inputs that matter

image_url is the opening frame, so the look is settled before the prompt does anything. Write prompt as motion: what happens next, how the camera moves, the mood. Restating the still is the classic I2V mistake.

duration_seconds is 5, 10 or 15 - a real range, and the reason this node is more useful than it first looks. Fifteen seconds at 720p is about 34 cents; the same at 1080p costs more and takes longer.

shot_type is the sleeper control: single is one continuous take, the multi-shot option turns the clip into an edited sequence with cuts. If you're generating a 15-second thing that reads like a trailer, that's the setting, not a longer prompt.

resolution is 720p or 1080p. negative_prompt is available and worth using. enable_prompt_expansion is off by default - turn it on to have your brief rewritten into something fuller, off again when the wording is deliberate.

audio_url is the odd one, and the tooltip is explicit about it: an optional voice or music AUDIO input the clip follows for mouth shapes and pacing, while the clip itself stays silent. So you can hand it the dialogue track to get lip sync right and still deliver a clean, soundless file. Nice design, easy to misread as "this node has audio."

seed repeats a draw exactly. Outputs are video (→ Save Video) and task id.

Install

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

Restart, then Settings > SpicyAPI with a key from spicyapi.ai/console/keys. Manager users: search "SpicyAPI". The pack installs nothing - no Python dependencies, no model files - and the key lives in the ComfyUI user folder rather than in node widgets, which is why sharing a workflow doesn't share your credentials. On a server, set SPICY_API_KEY instead.

Troubleshooting that actually applies

No SpicyAPI key is set - go add the key. 40004, "no deployment serves this combination" - usually a resolution and duration pairing that isn't offered; change the parameter the message names. 50301 means the model has no usable deployment right now, which is worth retrying later rather than debugging. 50302 means it failed upstream and was refunded; re-queue. An error ending in request_id ... should be pasted straight into a support ticket, because that id names the exact request.

The self-inflicted ones are the same as everywhere in this pack: seeds default to fixed so a re-queue costs one, not two; randomize re-bills; cancelling a run does not cancel the server-side job or the charge; and the node badge is a per-second rate, so set Max cost per run in the SpicyAPI settings and let the quote police you.