Skip to content

Model routing

How multi-model AI video routing works

The best model is the model that fits the production need, not a permanent winner for every shot.

Producers, creators, creative technologists, and AI platform evaluators

Start from the shot

Define duration, aspect ratio, camera and subject motion, reference needs, continuity, audio, dialogue, extension, and finishing target before selecting a provider.

Route deliberately

Seedance, Kling, and Veo expose different strengths and limits. A production can choose one model family for a long coherent take, another for specialized motion or lip sync, and another for reference-guided extension or finishing.

Keep fallback honest

If a requested backend is unavailable or incompatible, the caller should receive the reason and the supported fallback. Returning a WebGL, Canvas, or different provider result while claiming the requested backend succeeded destroys useful evidence.

Keep the workflow stable

Providers will change. The brief, assets, sequence, edit commands, review state, output specification, and lineage should remain durable around them.

Read the live model catalog.