Picsart adds Seedance 2.5 with 20-language lip-sync
By Steven Van ·
Picsart's AI Playground gains BytePlus's new video model, with realistic physics and 3D-reference VFX support.
Picsart has added Seedance 2.5, BytePlus's new flagship video model, to its AI Playground. The pitch goes well past longer clips: generate and adapt a single video across 20 languages with matched lip-sync, render complex physical interactions like shattering glass and collisions that actually look real, and feed the model a 3D reference to get back fully textured, lit generative VFX. For creators who make ads, product videos, or localized social content, each of those three features solves a real production problem.
One video, 20 languages, matched lip-sync
Localization has always been the expensive part of video. Reshooting is out of the question for most teams, and traditional dubbing leaves mouths flapping out of sync. Seedance 2.5 on Picsart generates and adapts a video across 20 languages with lip-sync matched to each one, so a character delivering a line in English can deliver the same line in Spanish, Japanese, or German with mouth movement that tracks. For brands running international campaigns, one generation can now cover markets that used to require twenty separate productions.
Physics that hold up: glass, collisions, and material impacts
AI video has historically fallen apart the moment objects touch. Seedance 2.5 was trained to handle complex physical interactions: shattering glass, collisions, and material impacts that behave the way physics says they should. Picsart also highlights the model's lighting, which places shadows naturally instead of painting scenes with the flat, rendered glow that gives AI footage away. If your work involves action, product drops, or anything where materials meet, this is the area where the upgrade from earlier models is most visible.
From 3D reference to textured, lit VFX
The most production-oriented feature is 3D reference support. Feed Seedance 2.5 a 3D reference and it returns fully textured, lit generative VFX built around it. That gives motion designers and VFX artists a way to keep control of geometry and blocking while letting the model handle surfacing and lighting, a workflow that previously meant hours in a 3D suite. It fits a broader pattern in this release: Seedance 2.5 accepts up to 50 references per generation, up from 15 in Seedance 2.0, so you can lock a character's face, wardrobe, voice, and world across an entire sequence.
Where to try it on Picsart
Seedance 2.5 is live now in Picsart's AI Playground at picsart.com/ai-playground, where you can prompt it directly and compare its output against more than 150 other AI models from the same interface. Picsart has also published a dedicated model page with examples and details. Beyond the web playground, the model is reachable from Picsart's desktop app or programmatically through the company's CLI, MCP, REST API, and SDK, which matters if you want to wire 20-language video generation into an automated content pipeline rather than clicking through a UI.
Sources: Picsart Seedance 2.5 model page, Picsart announcement on X, The Decoder