Seedance 2.0 fast Native Audio

Seedance 2.0 Fast is ByteDance's accelerated video generation model family for creating videos from text, images, and multimodal references. Atlas Cloud provides separate endpoints for text to video, image to video, and reference to video, all with native audio support.

Choose text to video for prompt based creation, or animate a first frame with the image to video endpoint and optionally define the last frame. Reference to video accepts images, videos, and audio for multimodal generation, video editing, and extension.

Create an API key in the Atlas Cloud dashboard and store it securely in a server environment variable. Select the endpoint that matches your input workflow, then submit a prompt and the required media. Generation runs asynchronously, so save the returned prediction ID.

Send an authenticated POST request to `/api/v1/model/generateVideo` with the exact model ID and endpoint specific inputs. Use the returned prediction ID to poll `/api/v1/model/prediction/{prediction_id}` until the request completes or reaches a terminal error state.

Text prompts are supported across the family, while image to video requires a first frame and accepts an optional last frame. Reference to video can accept up to nine images, three videos, and three audio files. Available controls include duration, resolution, aspect ratio, bitrate mode, seed, audio generation, watermarking, and last frame return.

Native synchronized audio is available for all three generation modes and is enabled by default. Set `generate_audio` to `false` when your workflow requires video without generated audio.

Atlas Cloud lists a standard price of $0.09 per generated video second for the three Seedance 2.0 Fast endpoints. Billing is pay as you go, so total cost scales with output duration and the selected generation settings.

The Fast variant is designed for lower latency workflows such as prompt iteration, drafts, and higher volume generation. Atlas Cloud exposes Fast and standard variants through separate model IDs. Compare both with representative prompts before choosing one for a production workload.

First, verify the Bearer token, exact model ID, required inputs, media limits, and accepted parameter values. Continue polling with the returned prediction ID until the status becomes completed, failed, or timeout. When using reference audio, include at least one reference image or video as required by the schema.