Dreamina and CapCut product pages now reference Seedance 2.5, ByteDance’s next video model, with rumors converging on a possible July 9 release. TestingCatalog reported that the pages describe a beta long-video mode extending generation from a 30-second standard clip to a 180-second output, with a 90-second draft tier in between. That three-minute jump is the new fact here, distinct from the 4K resolution and 50-reference-file inputs AI Insiders covered when Seedance 2.5 first leaked on June 25.
Three minutes is not simply a bigger number on the same problem that earlier leak described. Every additional second a video model generates is another chance for a character’s face to drift, a hand to reshape, or a tracked camera move to lose its bearing. Keeping a character’s look, its movement, the camera’s behavior, and the original prompt’s intent steady over six times the prior duration is a materially harder coherence problem than producing one clean half-minute shot. Dreamina’s own product copy, per TestingCatalog, labels the 180-second mode as beta, which reads as an admission that ByteDance is not yet confident in that consistency holding up.
No independent benchmark accompanies the claim. The evidence is public product pages and a CapCut social post, not a dated release notice from ByteDance, so both the July 9 timing and the 180-second ceiling remain claims rather than confirmed specifications until the company posts one.
The distribution path is the second reason this differs from June’s story. Seedance 2.5 is not landing first as a developer API tier. TestingCatalog reported it will ship inside CapCut, the editing app ByteDance already ties to its broader creator ecosystem, alongside Dreamina and unnamed partner platforms. A frontier-adjacent video model wired into a mainstream editing app reaches advertisers, anime editors, and social accounts who would never call a raw API. That is reach through an installed base, not a benchmark win, and it is a distribution lever a pure API release cannot match.
That reach is what should worry rivals. Google’s Veo and Gemini Omni stay the closest capability competitors, according to TestingCatalog, but neither ships inside a consumer editing app at CapCut’s scale. OpenAI’s position weakens further by default: the company shut down the consumer-facing version of Sora on April 26, leaving a gap in mainstream AI video that a CapCut-native Seedance could fill without having to out-benchmark anyone. Seedance 2.0, per BytePlus listings, still caps near 15 seconds, making 2.5’s jump to three minutes ByteDance’s largest single duration increase to date, independent of what any rival ships next.
Teams evaluating AI video vendors should hold judgment on the 180-second mode until ByteDance posts a dated release notice with disclosed limitations. Then test it specifically for identity drift across a moving camera before greenlighting Seedance 2.5 for any client-facing project.
TestingCatalog reported on July 4, 2026, citing Dreamina and CapCut product pages, that ByteDance’s Seedance 2.5 may launch July 9 with a 180-second beta video mode.
