DreamActor V2: the still is the contract (5cr)

DreamActor V2 is image-to-video. If the label is wrong, every second is waste.

ByteDance files it under motion-control, 5 credits, requires_image: true. The catalog description is not a scene generator: "AI-powered face reenactment model that transfers facial expressions and head movements from a driving video to a reference image." There is no listed duration enum, no listed max resolution, and audio is unset. You are not buying a soundtrack or a 10-second cinematic. You are buying a face performance mapped onto a still you already locked.

If that still is the wrong person, the wrong crop, or a three-quarter that does not match the driving clip, 5 credits per second of mapped motion is how you pay to watch the wrong identity act.

Reenactment is not a promptable scene

This row transfers facial expressions and head movements. It is not text-to-video. It is not a product orbit. The still is identity. The driving video is performance. Both are required in practice: the record requires an image, and it requires a video input on video_url.

There is no duration menu to "try 10 seconds and see." Length follows the driving clip. If the driver is a messy phone take with a face leaving frame, the output will be a messy reenactment of that. Prompting a cinematic establishing shot into a model whose job is face transfer is how you spend 5 credits on seconds that were never going to be that shot.

The still-to-motion door is image to video. The motion-control tool is motion transfer. The agent phrasing is transfer a dance or motion onto a photo. DreamActor is narrower than a full-body dance: it is face and head. ByteDance's roster also holds Seedance and Seedream. Those are not this contract. Best motion-control model is the shelf. This page is the reenactment SKU.

The still is identity. Inspect it like a contract.

Lock the still first. Front-facing, lit like the driving clip, head and shoulders actually in frame. A beauty crop that hides the jawline is not a reference for head motion. A group photo is not a reference for one face.

Then lock the driver. One subject, face visible, expression actually changing. A clip of someone walking away from camera is not a facial performance.

Only then run DreamActor V2. 5 credits is cheap per second until you multiply it by a driver you have not watched. The catalog does not list native audio. Do not expect a talking-head mix to come back with the picture. If you need a mouth pass against a track, that is a lipsync row, not this one.

The motion-control glossary is the mechanic: the driving clip sets timing. You do not describe the smile in prose and hope the still invents it.

If the label is wrong, stop

Wrong labels that waste this SKU:

• "I need a full scene from a sentence" → text-to-video, not DreamActor.
• "I need the body to dance" → a body motion-control row, not face reenactment.
• "I need the still to speak this VO" → lipsync, and the catalog here does not claim audio.
• "I need 1080p cinematic B-roll" → no resolution is listed on this record. Pick a generator that publishes one.

Every one of those can be run through DreamActor if you ignore the description. Every second of the result is then a transfer of the wrong job onto a face. 5 credits still leaves the register. The still being "kind of the right person" does not make the job image-to-video reenactment. It makes it a miss you will recut.

FAQ

Does DreamActor V2 require an image?

Yes. requires_image is true. The reference image is the identity. Without it you do not have this job. The driving video is the performance. Both files are the prompt.

How long is the output?

The catalog does not list a duration enum. Treat length as inherited from the driving video. Trim the driver to the expression you actually need before you spend 5 credits per second of transfer.

Does it output native audio?

The record does not set audio: true. Do not plan a mixed talking-head delivery on this SKU. Reenact the face, then finish sound on a lipsync or edit row if the plate needs a track.

Is this the same as Kling motion control?

No. Kling's motion-control rows transfer motion from a driving video onto a still with a different contract (and a listed resolution on the Standard SKU). DreamActor's description is face reenactment — expressions and head movement — at 5 credits, no listed resolution. Pick the row whose description matches the body part you actually need to move.