SceneWish Reference to Video: How Many References Each Model Takes
Page: Reference to Video · Site: SceneWish
Reference limits vary a lot. Some models also take reference videos or audio.
- Seedance 2.0 Fast: up to 9 images, optional 3 videos (2–15 s each, 15 s total) and 3 audio. 5–15 s, 480p/720p. Lower cost.
- Seedance 2.5: up to 30 images, optional 10 videos (30 s total) and 10 audio. 4–30 s, up to 1080p.
- Veo 3.1 Fast: 1–3 images. Fixed 8 s, up to 4K.
- Kling v3 Omni: 1–7 images. 3–15 s, up to 4K.
- Wan 2.7 Reference: up to 5 images, optional 5 videos (1–30 s each, 150 s total) and 5 audio. 2–15 s.
- Vidu Q4 Preview: 1–15 images, optional 3 audio. 3–16 s, up to 4K.
- Gemini Omni Flash Preview: 1–16 images. Length chosen by the model, 720p.
For the Seedance and Wan models, at least one image, video or audio reference is required.
Make references easy to read
One subject per image, a visible face or label, and a few different angles help the model recognise the subject. Then describe the new scene.
SceneWish team.