Seedance 2.5
#18 in Text-to-VideoSeedance · v2.5 · since 31. Juli 2026 (Consumer-Launch auf Jimeng AI/Doubao Pro; API-Zugang über BytePlus ModelArk/Volcano Engine Ark ab Anfang · 2× · last seen Aug 13, 2026
Seedance 2.5 is ByteDance's new multimodal text-to-video model, officially launched on July 31, 2026, first rolled out via the consumer apps Jimeng AI (即梦), Dreamina/CapCut, and Doubao Pro. Its headline feature is generating up to 30-second video-with-audio clips in a single pass (up from 15 seconds in version 2.0), with optional multi-round extension for longer sequences. The model supports up to 50 multimodal reference inputs (images, video, audio, text) and enables precise, region-level editing. API access via BytePlus ModelArk (international) and Volcano Engine Ark (China) followed shortly after the consumer launch with token-based billing; native 4K output was announced, but the shipped API currently only covers 480p and 720p.
Features
| Fine-Tuning | No official fine-tuning documented; supports up to 50 multimodal reference inputs (images, video, audio) for control instead of model fine-tuning |
| Max Resolution | Official API currently covers 480p and 720p; native 4K was announced but is not confirmed/available in the shipped API |
| Max Video Length | Up to 30 seconds in a single pass (with audio); beta multi-round extension mode for several minutes |
| Platform | Jimeng AI, Doubao Pro, Dreamina/CapCut (consumer); API via BytePlus ModelArk (international) and Volcano Engine Ark (China) |
| Price | API: $10.70 / M tokens without video input, $6.40 / M tokens with video input (BytePlus ModelArk); example: 5s/480p ≈ $0.51, 5s/720p ≈ $1.16 |
| Release Date | Officially launched on July 31, 2026 (previewed earlier on June 23, 2026 at the Volcano Engine FORCE conference) |