Wan 2.2 LoRA is Alibaba's image-to-video model with your own LoRAs. Attach up to three weight files in each of three slots (high-noise, low-noise or both) at strength 0 to 4, for 5 or 8 second clips at 480p or 720p.
Workbench
These examples were written for this endpoint. Replace the prompt and sample media URLs with your own accessible inputs, then check the current quote and model availability before submitting.
- Output5s
neon-rooftop-qipao-anime The still fixes her on the railing in a purple silk qipao under the full moon; the prompt carries the rest — swinging down, a hand along the slit, the walk toward the lens past the neon and a slow, knowing smile. Base run with `loras` left empty, `resolution` `480p`, `duration_seconds` `5`. Restyle by adding a weights URL to `loras`; both passes pick it up.
- 480p
- 5s
- Output5s
rooftop-walk The still fixes her pose at the railing; the prompt carries the rest — the push-off, the blazer resettled on one shoulder, and a walk that lets city lights and the vending machine glow slide past. Base run with `loras` left empty, `resolution` `480p`, `duration_seconds` `5`. Restyle by adding a weights URL to `loras`; both passes pick it up.
- 480p
- 5s
- Output5s
film-grain-low-noise-only The point is where the LoRA is applied, not which one it is. `low_noise_loras` attaches it to the low-noise pass only, so grain lands on texture and background while skin stays clean; `high_noise_loras` and the plain `loras` array target the other pass and both. Each takes up to three. Same weights, different pass, visibly different picture — worth one paired run before you settle on a configuration.
- 720p
- Output5s
ink-brush-sword-form The baseline with no LoRA attached, just the base model on its own, which is what a strength sweep needs to be measured against: an ink painting has to complete a full sword form — blade sweeping down and around, sleeves and hem trailing — without the wet brushwork and paper fibre collapsing into photographic rendering.
- 720p
- 5s
This example starts at $0.024 per output second. The Playground updates the estimate when you change the settings.
