Video Generation
Generate videos from text, a start frame, interpolated keyframes, references, motion control, or a talking avatar.
Video Generation creates a new clip from scratch — text, a starting image, two keyframes to interpolate between, reference media, a motion source, or an audio performance for a talking avatar. To transform a video you already have (edit, extend, or reframe it), that's the separate Video Edit mode.
For pricing details, see the Pricing page. Full model tables: Video Providers.
Sub-modes
Availability is per model — the picker only shows sub-modes that model supports.
Providers (current)
| Provider | Highlights |
|---|---|
| Veo (Google) | Lite / Fast / Full · up to 4K · 4–8s |
| Gemini Omni Flash | Fast 720p generation · 3–10s |
| Wan (Alibaba) | 2.7 suite: T2V, I2V, R2V · optional audio |
| HappyHorse | 1.1 T2V / I2V / R2V |
| Kling | 3.0 Turbo / 3.0 / 3.0 Omni · Avatar · motion control · up to 4K |
| Seedance | 2.0, 2.0 Fast, 1.5 Pro · multimodal refs |
| OmniHuman | Image + audio → talking portrait |
| LTX | 2.3 Fast / Pro · up to 4K · long Fast durations |
| Grok Imagine Video | 1.0 full suite · 1.5 start-frame only |
| Luma Ray 3.2 | HDR, loop, keyframe interpolation · 540p–1080p |
See Video Providers for the full model list and sub-mode map.
Settings
Resolution
Depends on the model (not plan): commonly 480p, 720p, 1080p, 1440p, or 4K where supported.
Duration
Varies widely:
- Veo — 4 / 6 / 8s (higher res often forces 8s)
- Gemini Omni Flash — 3–10s
- Wan / HappyHorse / Kling / Seedance — typically multi-second ranges up to ~15s
- LTX Fast — up to 20s · LTX Pro — up to 10s
- Grok — 1–15s
- Luma — 5 or 10s (HDR/loop often force 5s)
- OmniHuman / Kling Avatar — driven by audio length
Aspect Ratio
Common options include 16:9, 9:16, and 1:1. Many providers add 4:3, 3:4, 21:9, etc. Some start-frame or edit flows inherit ratio from the source media.
Audio
- Optional generated audio on Wan, Seedance, LTX
- Required audio input for OmniHuman and Kling Avatar
Tips
- Start Frame usually gives the most control — pick a strong still, then describe motion
- Veo 3.1 Lite or Wan 2.7 are good iteration defaults
- Match the sub-mode to the model (e.g. Grok 1.5 is start-frame only)
- Video jobs are slower than stills — expect tens of seconds to several minutes