Seedance API Production Guide for References, Shots, and Audio

Build a real evaluation across text, image, reference, and audio workflows without relying on launch demos.

Key takeawayFor the Seedance family, input composition and shot complexity matter more than one overall score.

Why this deserves its own decision

One model family may include standard, fast, mini, and multiple input modes. A single text-to-video test cannot reveal how references, boundary frames, audio, and complex action change the result—or whether a lower-cost variant is already sufficient.

Decision framework

Separate evaluation by input mode instead of combining text, image, and reference generation.
Compare standard and fast variants on the same shots for yield, wait time, and spend.
Score audio and visuals separately, then decide whether joint generation reduces post-production.

Putting it into a ModelRush workflow

Treat input assets, shot brief, audio need, and budget as ModelRush task fields rather than scattering version names through product code. Run a compact evaluation, create quality and value routes, then add webhooks, polling reconciliation, and idempotency for long jobs.