Luma adds ByteDance's Seedance 2.5 video model
Luma now hosts ByteDance's Seedance 2.5, which can generate a 30-second continuous clip in one pass, about twice the length Luma says most models reach before footage has to be stitched together.
Luma announced that ByteDance's Seedance 2.5 video generation model is now available on its platform. The headline feature is native 30-second generation in a single pass. Luma says most models stop at about 15 seconds and then need clips stitched together, and each join is a chance for a character's face or a scene to drift.
Users can also extend a clip forward or backward. Luma says the character, scene and camera stay consistent across extensions.
Other changes Luma lists:
- References: up to 50 inputs per generation, including images, video, audio and style, up from 12 in the previous generation. Luma recommends a tighter setup for the cleanest results: five subjects or fewer, six to ten references in total, and one clear view of each subject.
- Native audio: audio is generated in the same pass as the video and synced to the action. Given audio alone, the model generates matching visuals.
- Style swap: rendering in live-action, illustration, comic or anime styles while keeping the same composition, camera and rhythm, with subject identity held across styles.
- Targeted edits: adding, removing or adjusting an object re-renders only that element. Luma also says the lighting follows real physics more closely.
- Control: about 20% better prompt adherence, a figure that comes from ByteDance, plus native support for more than 10 languages without a translation step.
The post is Luma's announcement of hosting a third-party model, not ByteDance's own release. It includes no benchmarks, pricing or details of how the prompt-adherence figure was measured. Luma says it uses Seedance 2.5 when a request needs one continuous take rather than a stitched clip.