Image-to-video chains in ComfyUI — from a still to a clip on hosted models
Image-to-video is the workflow where ComfyUI's graph shape pays off: a still goes in, motion comes out, and everything in between is visible. Here's the chain that runs it on hosted models — no multi-gigabyte video checkpoints on your drive.
The chain
Two nodes: something that produces a still (LivePair Image, a Load Image node, or an edit chain you already have) wired into LivePair Video. The video node reads the attached image as the i2v source frame — the prompt then describes motion, not content: camera drift, hair movement, the subject turning.
Outputs are a VIDEO tensor plus the result URL. Preview it in ComfyUI, or chain it into frame extraction, interpolation, or an upscale pass.
Picking the model line
The dropdown covers the Wan i2v line and Seedance i2v models, including the private-tier variants for content lanes the mainstream hosts don't carry. Resolution and duration are node widgets — the same parameters the models page prices against, so the cost is knowable before the queue runs.
The useful pattern: generate the still on the t2i side first with the same prompt voice, then hand it off. Consistent style across the frame boundary is easier than feeding a mismatched source image to i2v.
What it costs
Clips bill per result at the per-resolution/duration price — the catalog lists it before you run. 480p drafts for pennies while you iterate, premium tiers when the shot is locked. Prepaid $10 credits are the whole buy-in; nothing bills per hour.
When to keep it local instead
If you're grinding hundreds of drafts through a local Wan install you already own, per-result pricing won't beat your sunk GPU. The hosted route wins when the hardware isn't there, when you're testing across model lines, or when the video tier you want isn't runnable locally at all.