Media models (preview)

Beyond text, the catalog includes image, video, audio (text-to-speech) and embedding models. They share the same account, credits and project scoping as text models, but each modality bills on its own meter instead of input + output tokens. Current models and list prices are on the pricing page.

Video

Video models: seedance-2-0 (standard), seedance-2-0-mini (lighter, lower cost) and seedance-2-0-fast (faster rendering). Per-model rates are on the pricing page.

Video generation bills on video tokens, computed from the pixels and seconds you render:

Per-second quick reference at list rates (16:9, no video input):

Worth knowing before you render:

• Aspect ratio changes the price. Cost follows actual width × height — a 1:1 clip at 720p costs ~44% less than 16:9 at the same tier.
• Auto duration bills actual seconds. With duration: auto, you are charged for the seconds the task actually renders, not the request.
• Reference video switches tiers. Adding input video moves the task to the with-input rate, and input seconds enter the formula.
• 4K is not yet available; requests above 1080p are rejected before any task is created.
• Failed tasks are not billed.

For the current video model parameters and examples, see Seedance 2.0.

Image

Text-to-image generation runs synchronously — prompt in, image URL back. List pricing will be published at launch; the model appears on the pricing page as Preview until then.

Audio (text-to-speech)

TTS bills per character of input text rather than per token. List pricing will be published at launch.

Embeddings

Embedding models bill on input tokens only — there is no output meter. Rates are on the pricing page.