Seedance VIP logoSeedance VIP

Latest · July 2026

Seedance 2.5

Single-pass 30-second video with native audio, 4K output and 50 multimodal references.

Seedance 2.5 is ByteDance's flagship video generation model, released in early July 2026. It is the first widely available model that generates a continuous, unbroken 30-second clip in a single pass — no segment stitching, no visible seams — with scene changes and tempo shifts inside one generation. Audio is co-generated in the same pass, so music, sound effects, dialogue and lip-sync land in sync with the picture. Region-level local editing lets you fix one detail (a character's hair color, a prop) without regenerating the whole clip, preserving the performance and lighting you already liked.

Specifications

Max clip length30 s single pass (3 min long-video beta)
Resolution480p / 720p / 1080p / native 4K, 10-bit color
Reference inputsUp to 50 multimodal references (image / video / audio)
AudioNative synchronized audio, co-generated
EditingRegion-level local editing after generation
ArchitectureSparse Diffusion Transformer, unified audio-video latent
AvailabilityDreamina / Jimeng, Volcano Engine API

Highlights

30 seconds, one pass

Generates a coherent 30-second clip with scene changes in a single generation — the hardest duration problem in AI video, solved without stitching.

50 multimodal references

Feed up to 50 images, video clips and audio tracks as references — enough to lock multiple characters, wardrobe, sets and a musical mood at once.

Local editing

Fix one region of the frame without a full regenerate. Keep the acting, expressions and lighting from the take you liked.

Native audio

Dialogue, sound effects and score are generated in the same latent pass as the picture, so lip-sync is native rather than dubbed on afterward.

Other models