Media GenerationModels 🇨🇳 31.07.2026 13:01

Just Now: Jimeng Seedance 2.5 Is Here! I Tested It Like Crazy……

ByteDanceByteDance
ByteDance released its self-developed video model Seedance 2.5, which is now available in the Jimeng AI app. The model claims industry-exclusive native generation of 30-second videos, with a long-video mode supporting up to 3 minutes, video editing capabilities, reference generation from up to 50 assets including green screen and white models, and precise timestamp control. Testing shows high consistency and cinematic quality.
ByteDance has launched its self-developed video generation model Seedance 2.5, and the Jimeng AI app is the official entry point for trying it. The model's headline feature is industry-exclusive native generation of 30-second videos, which was demonstrated through several test clips showing consistent character appearance across shots, high-quality CG-like visual effects, and strong narrative pacing. Additionally, a new ultra-long video generation mode supports videos up to 3 minutes long. Seedance 2.5 also introduces video editing capabilities, allowing users to modify, add, delete, or replace elements within a video, as demonstrated by removing parts of a generated space film. The model can reference up to 50 assets, including green screen and white model footage, and it integrates with Maya and Blender plugins for direct generation. A notable improvement is precise timestamp control, where text instructions can be applied to specific seconds within the video, with accuracy within 1 second. In testing, the model successfully added a brand logo at the exact requested time. The combination of longer generation, editing, and reference features aims to make AI video generation more suitable for professional production in film, advertising, and gaming, reducing the need for repeated generation attempts and manual post-processing.
Source: QbitAI 量子位 — original
Our earlier posts on this topic ↓
Fresh news