ByteDance Cracks the 30 Second AI Video Code
Seedance 2.5 creates 30 seconds of smooth AI video in one shot, including scene transitions, 50 concurrent references, and post-processing. Available early July 2026.
At their Volcano Engine FORCE conference on June 23rd, ByteDance revealed five new AI models. The star of the show is Seedance 2.5, an AI video generator set to release in early July. Its main selling point? It can generate up to 30 seconds of video all at once, without the need for post-processing.
What the 30-second barrier actually means
Currently, most AI video generators are used for short snippets. For example, Sora and Runway usually output videos between 4 and 8 seconds long. These shorter clips need to be edited together, which can result in jarring changes like sudden shifts in lighting or inconsistent character design. In order to make your AI videos look good, you have to spend a lot of time post-processing to cover up these rough edges. With Seedance 2.5, you can create an entire 30-second clip at once, complete with scene transitions and tempo changes. You can even use up to 50 concurrent references (images, audio files, etc.) to help steer the results. And you can still edit your generated video afterwards, while maintaining its original aesthetic.
Seedance 2.0, released alongside Seedance 2.5, features native 4k and 10-bit color output, which is great news for those looking to use AI video in a professional setting.
The rest of the announcement: models and pricing
ByteDance also introduced Doubao 2.1 Pro, a language model that they claim is approximately 80% cheaper than Claude Opus 4.6, while still offering similar performance for enterprise use cases. They also released Seedream 5.0 Pro for image generation, and Seed-Audio 1.0 for audio generation. All five models will be available via Volcano Engine, ByteDance’s B2B cloud service, which provides AI services to businesses.
Keep an eye on Doubao 2.1 Pro’s price claim. If it lives up to its promise of being 80% of the cost of Opus-level models, that could be a game changer for those doing large scale inference. Of course, these claims need to be independently verified, but competition in the cutting edge model market is heating up, and we’ve seen this trend accelerate throughout 2026.
When Seedance 2.5 launches in early July, the 30-second continuous generation is the thing to test specifically. Generate a scene that requires a setting change midway — something that would normally force a splice — and see whether the model holds consistency across it. That's the real test of whether the claimed improvement is practical or just a number in a spec sheet.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles