What a Backstage Clip is
A candid, behind-the-scenes motion video of your character. It is the post type that renders true video, with its own controls in the Composer.
Spoken lines
Put lines in "quotes" in What happens and the character says them on camera. The engine creates the mouth movement and the voice together, so there is no separate lip-sync pass. With no quotes you get a wordless clip with ambient sound.
Example: She looks up and deadpans, "We are doing this again? Really?"
The take stretches to fit your lines. The app allows about 2.3 spoken words per second plus one second. If the lines need more time than the duration you picked, it extends to the next length that fits, and the price follows the longer take. The cost line shows this before you click Generate.
Voice description
Voice description says how they sound, not what they say: pitch, pace, accent, energy. Use the same description every time so the character sounds like the same person across clips. It starts filled with the stock description for their voice, or whatever you last used for them in this browser. A Sample button appears when the text exactly matches a stock description.
Engines and prices
Each engine card shows its live price per second at the resolution you choose. Draft on a cheap engine, then re-run the take you like on 2.0 or 2.5.
| Engine | 480p | 720p | With a reference video | Lengths |
|---|---|---|---|---|
| Grok | 4 cr/s | 7 cr/s | not supported | 6-15s |
| Mini | 6 cr/s | 12 cr/s | 8 / 16 cr/s | 4-15s |
| Fast (default) | 17 cr/s | 35 cr/s | 22 / 48 cr/s | 4-15s |
| 2.0 | 27 cr/s | 58 cr/s | 37 / 79 cr/s | 4-15s |
| 2.5 | 40 cr/s | 89 cr/s | 48 / 107 cr/s | 4-10, 12, 15, 20, 25, 30s |
The defaults (Fast, 720p, 8 seconds) cost 280 credits. The same 8 seconds on Mini at 480p is 48; the cheapest possible clip is 24. Resolution is 480p or 720p; Aspect is 9:16 (default), 16:9, 1:1 or adaptive. Rendering usually takes 1 to 3 minutes, or 2 to 4 on 2.0 and 2.5.
A clip that talks costs the same as one that does not. Every rate above is the rate whether your character speaks or stays silent, across Mini, Fast, 2.0 and 2.5. That is unusual and worth knowing: elsewhere on the platform, Wan 2.6 Flash roughly doubles the moment you ask for spoken audio.
It is also the reason a talking character is one pass here rather than two. Rendering silent and syncing a voice on afterwards means paying for a second render, and the mouth is matched by a different engine than the one that drew the face. Put the line in quotes and let the clip speak it. Lip Sync in Brand Studio is for the other case, when the voice has to be a particular recording you already have.
Director options
Start from is one switch, because the engine shoots in one mode or the other:
- References: the character's face plus your images, composed into a new shot. Add reference images (drop, click, or From my gallery), reference videos for motion or style (up to 3, 15 seconds total, 50MB each; a reference video switches to the reference-video rate), and reference audio (up to 3, 15 seconds total). The engine takes up to 9 images including the identity: up to 8 extra with a single face, up to 6 with a locked pack.
- First frame: one image that locks the exact opening shot. References are not used in this mode. On Grok, a first frame runs Grok 1.0.
Switching back to References removes the first frame, so both can never be set. If you choose First frame without adding an image, Generate asks you to add one.
Ambient sound (on by default) controls audio on wordless clips; clips with spoken lines always have audio. Grok takes image references only and has no adaptive aspect (it uses 9:16).