Mixer AI
Seedance 2.5 vs Wan 3.0 Prime / Video: model comparison
Seedance 2.5 and Wan 3.0 are currently available in Mixer AI and both reach 30 seconds. Wan 3.0 has Video and Prime variants; Prime is the high-speed variant with the same current input and control contract. The models were tested separately, so this page does not invent one overall winner; it compares product capabilities with long-scene limits.
Key differences
Specifications
| Spec | Seedance 2.5 | Wan 3.0 |
|---|---|---|
| Model variants | Seedance 2.5 | Standard / Prime |
| Generation modes | from text / from first and last frames / from references / video editing | from text / from first and last frames / from references |
| First frame | Yes | Yes |
| Last frame | Yes | Yes |
| Input photos | up to 30 | up to 10 |
| Input videos | up to 10 | up to 5 |
| Input audio files | up to 10 | up to 5 |
| Input video length | 2–30 s | 1–15 s |
| Input audio length | 2–30 s | 1–15 s |
| Output audio | Can be enabled or disabled | Can be enabled or disabled |
| Minimum duration | 4 s | 2 s |
| Maximum duration | 30 s | 30 s |
| Resolutions | 480p / 720p / 1080p | 1080p / 720p / 480p |
| Aspect ratios | 9:16 / 16:9 / 1:1 / 21:9 / 4:3 / 3:4 / adaptive | 9:16 / 16:9 / 1:1 / adaptive / 4:3 / 3:4 |
| Prompt limit | 20,480 characters | 20,000 characters |
| Price, 1 s · 1080p | ≈ 101.19 coins | ≈ 22.25 coins · Standard |
Seedance 2.5: long multi-reference workflow
- The reviewer builds a long scene around complex motion and notes some morphing in the most intense motion-blurred moments.
- With a crowded reference set, the model swapped in the wrong character variant and the reviewer regenerated.
- The reviewer calls out a '30-second trap': the ability to generate 30 seconds does not mean every scene should use the entire duration.
Wan 3.0: one 30-second story
- The reviewer presents a 30-second story as one continuous generation rather than six clips stitched together.
- In the text-to-video test there is no character sheet: Wan has to create the character and preserve its details across different shots.
- The reviewer then moves to image/reference workflows to compare motion and character preservation against the fully text-generated sequence.
How to read this comparison
The same 30-second ceiling does not make the models identical. Seedance 2.5 exposes a larger reference budget in Mixer AI; Wan 3.0 is also multimodal but with smaller limits. Independent tests of both show that long generation is a test of accumulated consistency, not only frame quality.
- Seedance 2.5: more references and a larger prompt limit.
- Wan 3.0: the same 30-second ceiling and its own reference-to-video path.
- For both, inspect character identity and object state at the beginning, middle and end of the clip.
Sources
Prices, availability and specifications come from the Mixer AI catalog. This is not an independent test: try both models on the same task before choosing.
What we compare
FAQ
Do both models really reach 30 seconds in Mixer AI?
Yes. The current Seedance 2.5 and Wan 3.0 product specs both list a 30-second maximum.
Which one preserves a character better for 30 seconds?
The sources are not a direct A/B, so they do not support a fair winner on that criterion. Both reviews treat long-scene consistency as a separate challenge.