Generating video
Video mode turns a prompt, and optionally a start and end frame, into a short generated clip. It runs on Qwen, fal.ai, or BytePlus, the three providers that generate video. See the provider matrix for what each one covers.
Quick overview
- Switch the composer to Video mode.
- Write a prompt and set the aspect ratio, resolution, and duration.
- Optionally pick a start and, on some providers, an end frame.
- Check the cost preview, then generate.
Controls
| Control | Options |
|---|---|
| Aspect ratio | 16:9, 9:16, 1:1 |
| Resolution | 480p, 720p, 1080p |
| Duration | 5 or 10 seconds |
Duration in the composer is a simple 5 or 10 second choice, the underlying model may clamp slightly differently.
| Model family | Actual accepted range |
|---|---|
| Qwen, most fal.ai models | Roughly 2 to 15 seconds |
| BytePlus | Roughly 4 to 15 seconds |
| The Veo-based fal.ai model | Fixed to 4, 6, or 8 seconds only |
Start and end frames
Support for guiding motion with images depends on the provider.
| Provider | Start frame | End frame |
|---|---|---|
| Qwen | Yes | Yes |
| fal.ai | Yes | No |
| BytePlus | Yes | No |
Where an end frame is supported, the model animates between your two images, useful for a controlled camera move or a product reveal. Where it isn't, only the opening frame steers the result and the rest of the motion follows your prompt. The result plays in a built-in player and saves to your gallery.
Providers and their models
| Provider | What it offers |
|---|---|
| Qwen | Its own Wan video models, for both text-to-video and image-to-video |
| fal.ai | A hosted roster of third-party models on one key. The Video mode dropdown offers Seedance 2.0, Seedance 2.0 Fast, Veo 3.1, Veo 3.1 Fast, and Kling 2.6 |
| BytePlus | Seedance video, direct from BytePlus rather than through a hub |
Lip-sync is applied automatically, not picked here
fal.ai's roster does include a lip-sync model behind the scenes, but it is not one of the models you choose in plain Video mode. Lip-sync is applied automatically by Marketing Studio for talking-avatar scenarios: a talking-head clip is generated first, then passed through the lip-sync model to lock the mouth to the audio. If you want a presenter whose lips match the voice, go through Marketing Studio rather than looking for a lip-sync option in the Video dropdown.
Cost preview
Because video can be expensive, a cost preview estimates the price before you generate, roughly the model's per-second rate multiplied by your chosen duration, so there are fewer surprises. This inline dollar estimate is shown for fal.ai and BytePlus models. Qwen's own Wan models do not show a dollar estimate in the composer, so for those check DashScope's own pricing page directly.
Rates vary a lot by model, so check the estimate each time you change models, not just once. As an illustration only, and these can change, a fast fal.ai model can run near $0.07 per second of clip while a premium one runs closer to $0.40 per second, with others in between, and BytePlus's Seedance is a flat estimate of roughly $0.14 per second. Providers on fal.ai in particular span a wide range on the same key, which is exactly why it is worth comparing before you commit to a longer or higher-resolution clip.
Step by step
Open AI Studio from the app sidebar (the item tooltipped Create with AI), then switch the composer's mode control to Video. Pick a video-capable provider (Qwen, fal.ai, or BytePlus). If none is connected, add one from Settings, AI Settings, Provider & Model tab (see Connecting a provider).
1. A first prompt-only clip
- Write a prompt, for example
a reusable steel water bottle rotating slowly on a white pedestal, soft studio light, seamless white background. - Set the aspect ratio (for example 9:16 for vertical), the resolution (start at 720p), and the duration (5 seconds).
- If you are on fal.ai or BytePlus, read the cost preview, roughly the per-second rate times your duration. On Qwen, check DashScope's pricing page instead, since no dollar estimate is shown in the composer.
- Generate. The finished clip plays in a built-in player right in the thread, and saves to your gallery.
2. An image-to-video clip with a start frame
- Add a start frame, for example your product hero shot.
- On Qwen, you can also add an end frame, and the model animates between the two, useful for a controlled camera move or a product reveal. On fal.ai and BytePlus only the start frame is used, and the rest of the motion follows your prompt.
- Write a prompt that describes the motion, for example
the bottle turns a quarter rotation to reveal the label, camera slowly pushes in. - Generate and review in the player.
3. Compare two providers or models before committing
- Run your prompt on one fal.ai model, for example Seedance 2.0 Fast, and note its cost preview.
- Switch to another model, for example Veo 3.1, run the same prompt, and compare both the result and the estimate.
- Because a fast model can cost a fraction of a premium one per second (see Cost preview), test the direction cheaply at a short duration and lower resolution first, then commit to a longer or higher-resolution clip only on the model you liked.
Common tasks
Concrete prompts to copy and adapt. Keep the first pass short and cheap, then scale up.
| Goal | Example prompt |
|---|---|
| Product spin | a matte black water bottle slowly rotating on a marble pedestal, soft rim light, seamless white background, subtle reflection. |
| Appetizing close-up | a cup of coffee with steam rising, close-up, warm morning light through a window, gentle camera push-in. |
| Product reveal (start and end frame on Qwen) | the closed box opens and the sneaker rises out, smooth motion, studio lighting. |
For a talking presenter whose mouth matches a voiceover, do not look for a lip-sync model here, use Marketing Studio, which adds lip-sync automatically for that scenario.
After generating
A finished clip plays back in a built-in player right in the thread, and saves to your gallery for reuse. From there it can be downloaded or referenced again later, the same gallery Marketing Studio scenarios draw on.
Video costs add up
You are billed by your provider for what you generate, and video is more costly than images. Start with a short clip at a lower resolution to test a prompt, then scale up once you like the direction.
Try more than one provider
Since fal.ai alone hosts several distinct models, and Qwen and BytePlus each bring their own, the same prompt can look noticeably different across providers. If a result isn't landing, switching models is often faster than rewriting the prompt.
Troubleshooting
I do not see a cost estimate. The inline dollar preview is shown for fal.ai and BytePlus only. On Qwen's own Wan models, no dollar estimate appears in the composer, so check DashScope's pricing page directly for those rates.
I cannot find a lip-sync model in the Video dropdown. The Video mode dropdown for fal.ai offers Seedance 2.0, Seedance 2.0 Fast, Veo 3.1, Veo 3.1 Fast, and Kling 2.6, and none of these is a lip-sync model. Lip-sync is not something you select yourself in plain Video mode. It is applied automatically by Marketing Studio for talking-avatar scenarios: a talking-head clip is generated first, then passed through the lip-sync model to match the mouth to the audio.
My end frame is being ignored. End frames are supported on Qwen only. On fal.ai and BytePlus, only the start frame is used and the rest of the motion follows your prompt. To animate between two specific images, use Qwen.
The clip length does not match the 5 or 10 seconds I picked. The composer's 5 or 10 second choice is clamped by the model. The Veo-based fal.ai model is fixed to 4, 6, or 8 seconds, Qwen and most fal.ai models accept roughly 2 to 15 seconds, and BytePlus accepts roughly 4 to 15 seconds, so the final duration may land near, but not exactly on, your choice.
Related
- Connecting a provider
- Provider and model matrix
- Marketing Studio
- Avatars and products for keeping a subject consistent across clips