Vitent now supports Seedance 2.5. On September 10, 2026, we verified the integration in the live English Agent interface with a minimal abstract-video task. The test was intentionally simple so it could confirm model selection, submission, billing, completion, playback, and output metadata without relying on external assets.
What we verified in Vitent
The request asked for a polished silver sphere rotating above a dark pedestal while a soft blue light moved across it. It excluded people, dialogue, logos, watermarks, and on-screen text.
Before submission, the approval card showed Seedance 2.5, 5 seconds, 16:9, 480p, audio off, and an estimate of 85 credits. After confirmation, the account balance changed from 812 to 727 credits, matching the estimate.
The task completed successfully and produced a playable five-second video. The result kept the sphere centered above the pedestal, introduced a visible blue highlight across the reflective surface, and followed the exclusions. This confirms the tested path; it does not by itself prove every Seedance 2.5 feature or the quality of more complex multi-character scenes.
What Seedance 2.5 documents
ByteDance Seed describes Seedance 2.5 as an audio-video generation model for longer storytelling, multimodal reference, and editing. Its official release states that the model can generate up to 30 seconds in one pass and supports multiple rounds of extension.
The provider also specifies a per-generation reference limit of up to 30 images, 10 video clips, and 10 audio clips. That is more precise than saying only “up to 50 multimodal references,” because the limit is divided by media type.
Provider-documented editing capabilities include timestamp-level audio and video editing, green-screen work, camera-perspective control, and reference-based editing. The model page also highlights white-model control for blocking and camera planning.
Those specifications describe the model. Our Vitent test verified selection and completion of a basic text-to-video task. It did not test the full reference limits, video extension, green-screen editing, white-model control, or timestamp-level editing inside Vitent.
Use the Agent to prepare a clean generation task
For a production clip, first define the outcome, duration, references, required actions, camera language, audio needs, and exclusions. Review the Agent's approval card before submission, then inspect the completed video for subject consistency, physical plausibility, motion, text, audio, and brand accuracy.
Prepare a Seedance 2.5 video plan for this shot. List the reference media, timing, camera movement, required actions, exclusions, output settings, and live credit estimate before asking me to confirm.
ByteDance Seed notes that complex-motion physics and stability in scenes with multiple interacting subjects still have room to improve. Vitent's project context can organize the inputs and approvals, but the completed output still needs human review.
Sources and test boundary
- ByteDance Seed: Introducing Seedance 2.5
- ByteDance Seed: Seedance 2.5 model page
- Vitent integration test: September 10, 2026, English Agent interface, one 5-second 16:9 480p text-to-video generation with audio disabled.
Try Vitent:app.vitent.ai