Text to video vs image to video: which gives more control over the final shot

Image to video gives more control over the final shot because it anchors the output to a fixed visual reference, while text to video relies only on prompt description. On RelayDance, image to video uses reference media placed in metadata.content[] and cited in the prompt as @image1 to @imageN, supporting up to 9 reference images per request. Both methods run on the same Seedance models and pricing, so control differs by input type, not cost.

How image to video anchors the final shot

Image to video anchors composition, subject and framing by feeding concrete reference media into the request. On RelayDance you place assets in metadata.content[] and reference them in the prompt with tokens @image1 to @imageN. A single request accepts up to 9 reference images, 3 reference videos and 3 audio tracks, with clips up to 15 seconds. Because the model starts from your supplied frame instead of interpreting text alone, the layout, character appearance and color remain closer to your intent. This reduces the variance seen when a model must invent every visual element from a description, giving predictable results across repeated generations.

When text to video is enough

Text to video is sufficient when you do not have a reference frame and want the model to compose the scene from the prompt. You submit a task with POST /v1/video/generations, passing model, prompt, seconds and metadata (ratio, resolution, generate_audio, callback_url). Without reference media in content[], the model generates every element from language, which suits exploratory work and quick drafts. The trade off is lower shot control: framing and subject identity may shift between runs. Since RelayDance bills pay-as-you-go and, according to the official relaydance.com docs, 「Failed or errored requests are never billed」, iterating on text prompts carries no charge for failed attempts.

Pricing is identical across both methods

Pricing depends on the model and resolution, not on whether you use text or image input. The rates below are drawn from relaydance.com/models.

Because both input methods share these rates, choosing image to video for greater control does not add a surcharge.

Steps to run image to video with reference frames

You add control by attaching reference media and citing it in the prompt.

1. Set your client base URL. According to the official relaydance.com docs, 「Set base_url to https://relaydance.com/v1 and keep your OpenAI SDK」.
2. Add reference assets to metadata.content[] (up to 9 images, 3 videos, 3 audio tracks).
3. Cite each asset in the prompt using @image1 to @imageN.
4. Submit with POST /v1/video/generations, including model, prompt and seconds (clips up to 15 seconds).
5. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed, then read the video url.

FAQ

How many reference images can I attach? Up to 9 reference images per request, plus 3 reference videos and 3 audio tracks, cited as @image1 to @imageN.

Do failed generations cost money? No. Billing is pay-as-you-go per generated video, and failed or errored requests are never billed (see relaydance.com/docs).

Do I need to change my SDK to switch input methods? No. The API is OpenAI-compatible; set base_url to https://relaydance.com/v1 and both text to video and image to video use the same endpoint.