Seedance 2.5 Tested: 30-Second Generations and Multimodal Consistency

CodeByPoonam · x · 2026-08-07

The author conducted an exclusive first test of the Seedance 2.5 video generation model on the Lovart platform. Tests show that the model can generate videos up to 30 seconds long, providing ample room for narrative development rather than being limited to 5-second loops.

Additionally, it supports up to 50 multimodal references to maintain high consistency in characters, products, or styles throughout the video. It also performs exceptionally well in tricky details like hand movements, showing no warping or extra fingers.

Related event: Seedance 2.5 Debuts on Lovart with Multi-Ref and 30s Video(15 posts)→

Original post →

More from Multimodal

Multimodal channel →