P Video Animate is motion transfer, not text-to-video. Give it a character image and a reference video — it keeps your image's look and copies the video's motion, timing, camera movement, and scene structure. The instruction prompt is optional: leave it empty and the driver carries the whole performance; fill it to lock identity or tighten lip sync.
P Video Animate takes two inputs: a reference video (the driver) and a single character image (the still). It generates a new clip using the still's look and the driver's motion, acting, timing, camera movement, and scene structure. The output's duration and aspect ratio follow the driver video.
It's strongest on slow, controlled action — talking heads, presenters, walking, product demos — where the motion is clearly visible and the framing is stable. Pricing is per second of the driver's length: 1 credit/second at 720p, 2 credits/second at 1080p. There's no separate prompt-from-scratch step; the performance comes from the video you feed it.
2
Inputs — driver video + still
1 / 2
Credits/sec — 720p / 1080p
Source
Duration follows the driver
The reference video supplies the motion, acting, timing, camera movement, and scene structure. You don't describe the action in words — you hand the model a take whose performance you want, and it retargets that onto your still.
Your character image owns appearance — face, wardrobe, palette, art direction. Use one clean, front-facing still (only the first image is used) and match its framing to the driver: chest-up still for a chest-up driver, full-body for full-body.
Leave the prompt empty for a fast first pass — the driver carries the performance. Add a line to lock identity, name motion beats, or tighten lip sync: "she speaks to camera and lifts a bottle; match lip sync and audio from the source video."
Keep Save Audio on for dialogue-driven clips and the driver's audio rides through to the output, lined up with the motion. Turn it off for a silent clip when you only need the movement.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
720p · driver duration · audio on · lip-sync prompt
Driver: seated presenter talking to camera (chest-up). Still: studio headshot, professional blazer, front-facing. Instruction: "the woman speaks to camera as a presenter; match lip sync and audio from the source video."
Chest-up still + chest-up driver is the cleanest framing match, and the talking-head is exactly the slow, controlled motion the model handles best. The lip-sync line in the instruction prompt is what makes the mouth track the driver's audio instead of drifting.
720p · driver duration · audio on · lip-sync prompt
Driver: seated presenter delivering a line. Still: tight face portrait, golden light, three-quarter angle. Instruction: "the woman talks to camera; match lip sync and audio from the source video."
Tight crops transfer expression and lip motion convincingly because there's nothing else competing in frame. Keep the still front-facing or near three-quarter — extreme profiles give the model less to work with.
720p · driver duration · audio on · empty prompt
Driver: a person seated in an armchair, slow head turns and settling. Still: woman seated by a window holding a cup. Instruction: left empty — the driver carries the performance.
No prompt needed — slow, seated motion is the model's sweet spot, so the empty fast pass already looks natural. Reach for the instruction prompt only if identity drifts across a longer take.
720p · driver duration · audio on · empty prompt
Driver: a single figure with steady full-body motion. Still: full-length fashion editorial figure, front-facing. Instruction: left empty.
Full-body works when the still is framed full-body too, so the model has feet-to-head to retarget. Wardrobe, palette, and styling stay locked to your image while the pose follows the driver.
Chest-up still for a chest-up driver, full-body for full-body. A framing mismatch (full-body still on a tight talking-head driver) is the most common cause of awkward crops or warped proportions.
Walking, talking heads, presenters, and product demos are the sweet spot. Very fast action, heavy occlusion, and extreme camera motion reduce consistency. Prefer stable exposure, minimal motion blur, and clear visibility of the motion you want to copy.
One subject, front-facing or three-quarter, well lit. Only the first image is used. If you need a specific character, generate the still first (e.g. with P-Image), then drive it here.
The driver carries motion and audio on its own — empty is the fastest way to a usable take and a good timing check. Only add an instruction prompt when you need to fix something.
When identity drifts on a longer take or the mouth doesn't track speech, name the subject and motion beats and add "match lip sync and audio from the source video." Change one variable at a time across reruns.
720p is half the cost and plenty for blocking out framing and motion. Re-run the keeper at 1080p. Aspect ratio always follows the driver video, and the output length matches it too — trim the driver to the length you want.
| Setting | Values | Notes |
|---|---|---|
| Reference video | Required (.mp4 URL or upload) | The driver — supplies motion, timing, camera, and scene structure. Up to ~2 min at 720p; output length follows it. |
| Character image | Required (image URL or upload) | Owns the look. One clean, front-facing still; only the first image is used. |
| Instruction prompt | Optional text | Empty = driver carries the performance. Fill to lock identity, name motion beats, or tighten lip sync. |
| Resolution | 720p · 1080p | 720p = 1 credit/sec, 1080p = 2 credits/sec. Aspect ratio follows the driver. |
| Save audio | On / off | On keeps the driver's audio in the output (best for dialogue). Off returns a silent clip. |
| Seed | Optional integer | Set for reproducible A/B tests; change one variable at a time. |
| Pricing | Per second of output | 1 credit/sec at 720p, 2 credits/sec at 1080p — billed on the driver duration. |
P Video Avatar creates new speech from a still — you give it a script (TTS) or an audio file and it generates lip-synced talking. P Video Animate copies motion from an existing driver video onto your still; it doesn't write a new script. Use Avatar for new lines, and Animate when the timing and performance of a take you already have should drive the still.