Compare video models
AI Video offers many models. They differ in how long a clip can be, the highest resolution, how many references they accept, whether they make sound, and price. This guide helps you pick one.
Quick picks
- Not sure? Start with the default, Wan 3.0: up to 30 seconds and 1080p, with many image, video and audio references.
- Trying ideas cheaply — Wan 2.2 Turbo or P-Video-2, then make the final version with a larger model.
- Long clips with references — Seedance 2.5 or Kling 4.0, up to 30 seconds.
- Highest resolution — Kling 3.0, Gemini Omni Flash 1.1, Vidu Q4 or Veo 3.1, up to 4K.
- Photos of real people — if a model refuses your photo, try another model, or use a Character instead.
Model guide
| Model | Length | Up to | Good to know |
|---|---|---|---|
| Wan 3.0 | 2–30 s | 1080p | The default. Start and end frames; up to 10 images, 5 videos and 5 audio clips as references. Reference video plus result can be up to 30 s in total. |
| Kling 4.0 | 3–30 s | 1080p | Sound is always generated. Start and end frames; up to 10 reference images (JPG or PNG). |
| Seedance 2.5 | 4–30 s | 1080p | The most references: up to 30 images, 10 videos and 10 audio clips. Start and end frames. |
| Seedance 2.0 | 4–15 s | 1080p | Rich references (9 images, 3 videos, 3 audio clips). Start and end frames. |
| Seedance 2.0 Mini | 4–15 s | 720p | A lighter Seedance 2.0 with the same references, up to 720p. |
| Gemini Omni Flash 1.1 | 3–10 s | 4K | Start and end frames; 6 images and 3 short videos as references. |
| MiniMax H3 Max | 5–15 s | 1080p | Synchronized sound; many references. Doesn’t use Characters. |
| Kling 3.0 | 3–15 s | 4K | Optional sound; up to 7 reference images. |
| Kling 2.5 Turbo | 5 or 10 s | 1080p | Fast, from a start frame or text. |
| Vidu Q4 | 3–16 s | 4K | Up to 12 reference images and short voice clips. |
| Vidu Q3 | 1–16 s | 1080p | Turbo or Pro quality in text and start-frame modes (start frame only). |
| Vidu Q2 | 2–8 s | 1080p | Turbo or Pro quality in Frame to Video (start frame only). Generate Video also accepts reference images and videos. |
| Veo 3.1 / Veo 3.1 Fast | 4, 6 or 8 s | 4K | Optional sound (on by default); from text or a start frame of at least 720 pixels. Fast is the lower-cost version. |
| P-Video-2 | 5, 10, 15 or 20 s | 1080p | From a start frame; optional sound; low price. |
| Wan 2.7 | 2–15 s | 1080p | Start and end frames; negative prompt. |
| Wan 2.2 Turbo | 5 s | 720p | Fixed 5-second clips at a low price. |
The model list changes as new models are added and older ones are retired, so the app always shows the current models, their options and prices. Limits shown when you upload (number of references, file size, length) are the ones that apply.
What the terms mean
- Start and end frames — in Frame to Video, the clip begins on your first image and, with an end frame, finishes on your second.
- References — images, videos and audio the model uses as material in Generate Video. Point to them in the prompt with @.
- Sound — some models make speech, effects and ambience with the video; others are silent.