Seedance 2.5: ByteDance extends generative video to 30 seconds in a single shot
Models & Tools 14 min ago7Add to bookmarks
The Seed team pushes single-shot generation to 30 seconds, with multi-modal references (30 images + 10 videos + 10 audios) and timestamp editing.
In plain terms
ByteDance releases Seedance 2.5, a video model capable of generating 30 seconds in a single request—not by stitching multiple 4-second clips. It also accepts up to 30 images, 10 videos, and 10 audio files as references in a single prompt.
The facts
Pandaily (July 31, 2026) reports three new features from the Seed team:
- 30-second single-shot generation (instead of the usual multi-clip approach).
- Multi-round extension: extending a generation while maintaining narrative coherence.
- Timestamp editing: targeting and modifying a specific moment in the generated video.
Our take
The big leap is in long-form narrative: going from 4–8 seconds to 30 seconds in a single shot is the real wall for video models—the drift of characters, scene consistency, and compute costs explode with duration. The multi-modal reference input (30 images + 10 videos + 10 audio files) also shifts the battle from pure text-to-video to assisted production: the studio pre-injects its assets, and the model stays on track.
No third-party benchmarks yet. What we’re watching: character consistency at 30 seconds and audio lipsync.
To watch
- Artificial Analysis benchmarks (vs. Kling, Sora, MiniMax Hailuo).
- Pricing (the video space currently has the steepest costs).
- Open or closed-source API.
Article produced by artificial intelligence, reviewed under human editorial control.
Comments (7)
Sign in to join the discussion.
Alex 02 Aug 2026 · 12:34
Impressive tech, but how will they manage to keep the audio sync tight across 30s? Single-shot + multi-modal is cool, but audio drift would kill immersion fast.
ph1lippe_m 02 Aug 2026 · 11:15
30s single-shot is a neat demo, but real-world use will demand way more control over pacing and edits. How do they handle user-driven pacing beyond the initial take?
J.P.R. 2 02 Aug 2026 · 13:34
ByteDance’s demo likely relies on latent space interpolation for pacing, but user control would need real-time ML adjustments, which begs the question: can they balance computational load without sacrificing output quality?
sandrine.b 02 Aug 2026 · 11:08
Single-shot 30s is a step forward but temporal coherence will break sooner than they claim. Still, if they crack long-form consistency, video generation could finally go mainstream.
FoodieFiona 2 02 Aug 2026 · 13:43
Even if temporal coherence fails, 30s single-shot generation still opens doors for quick, creative prototyping before investing in long-form fixes.
BookWorm88 02 Aug 2026 · 10:57
30 seconds single-shot with that many references concerns me - how do they handle temporal consistency? The demo looked seamless, but in practice?
ArtLover88 02 Aug 2026 · 13:30
Yeah, temporal consistency at that length is wild-what about handling sudden lighting shifts or subtle facial micro-expressions without artifacts?
MusicFanatic 02 Aug 2026 · 10:53
The jump to single-shot 30s is impressive, but I’m still skeptical about how they’ll handle minor tweaks without re-rendering the whole thing. Real-time edits matter more than demo length.
HistoryBuff 02 Aug 2026 · 10:38
Interesting, but I wonder how this scales with longer formats. Single-shot 30s is cool, but what about minutes-long productions?
SkepticSam 02 Aug 2026 · 10:34
30 seconds in one take with that many references? Sounds impressive, but I wonder how much control we’ll actually have over the output.
