The seedance 2.5 fal api provides text-to-video, image-to-video, and reference-to-video generation through three model IDs: bytedance/seedance-2.5/text-to-video, bytedance/seedance-2.5/image-to-video, and bytedance/seedance-2.5/reference-to-video. Billing is token-based: tokens = (output_height × output_width × duration_seconds × 24) / 1024. For standard 16:9 generation, fal lists approximate costs of $0.2205 per second at 480p, $0.4730 at 720p, and $1.164 at 1080p. These are planning estimates; calculate a specific request from its dimensions, duration, and applicable token rate.
Last verified: 2026-10-22
Which fal Seedance 2.5 endpoint should you use?
Choose the endpoint according to the inputs your application supplies. The model ID becomes the path after https://fal.run/ when making a REST request. Keep the full endpoint ID in your configuration so each generation workflow uses the intended route.
| Workflow | Model ID | Input details |
|---|---|---|
| Text to video | bytedance/seedance-2.5/text-to-video | A required prompt describes the generation. |
| Image to video | bytedance/seedance-2.5/image-to-video | Requires image_url; end_image_url is optional. |
| Reference to video | bytedance/seedance-2.5/reference-to-video | Accepts image_urls[], video_urls[], and audio_urls[]. |
For an image/reference to video API integration, distinguish a starting-image workflow from a multimodal reference workflow. Reference-to-video supports up to 50 multimodal reference inputs, addressed in the prompt as [Image1], [Video1], [Audio1], and so on. Video references also affect billing, as explained below.
For broader integration context, read our Seedance 2.5 API providers guide and the separate BytePlus Seedance 2.5 API guide.
How do you set up your first API request?
Set your fal credential in the FAL_KEY environment variable. Requests use the Authorization header value Key $FAL_KEY and the URL pattern https://fal.run/<model-id>. The following minimal curl example selects the text-to-video endpoint with an explicit resolution, duration, and aspect ratio.
export FAL_KEY="YOUR_FAL_KEY"
curl --request POST \
"https://fal.run/bytedance/seedance-2.5/text-to-video" \
--header "Authorization: Key $FAL_KEY" \
--header "Content-Type: application/json" \
--data '{"prompt":"A slow camera move across a sunlit desk.","resolution":"720p","duration":"5","aspect_ratio":"16:9","generate_audio":true,"end_user_id":"YOUR_END_USER_ID"}'
Replace the credential and end-user placeholders with your own values. The end_user_id parameter is required for B2B use. For credential planning across integrations, see our Seedance API key guide.
The verified text-to-video parameters include resolution values of 480p, 720p, and 1080p, with 720p as the default. Duration accepts auto or a value from 4–30 seconds. Aspect ratio accepts auto, 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. An optional seed is also available.
Audio generation defaults to true, and enabling or disabling it does not change the price. When requesting dialogue for lip-sync, put the dialogue in double quotes within the prompt. To explore how these choices fit into an application, read our video generation API guide.
How does fal token billing translate into dollars?
For standard, non-US hosting, the rate is $0.0214 per thousand tokens at 480p and 720p, and approximately $0.0234 per thousand tokens at 1080p. After calculating tokens, estimate USD cost as tokens / 1,000 × the applicable rate.
| Resolution | USD per thousand tokens | Approximate USD per second, 16:9 |
|---|---|---|
| 480p | $0.0214 | $0.2205 |
| 720p | $0.0214 | $0.4730 |
| 1080p | Approximately $0.0234 | $1.164 |
These per-second figures are approximate published estimates for standard 16:9 output with audio enabled. Audio has no price surcharge. The verified frame grids specify 864 × 496 for 480p 16:9 and 1280 × 720 for 720p 16:9.
Use the dimensions-based formula for a detailed estimate. The published per-second figures do not exactly reproduce every worked example below, so keep the two kinds of estimates distinct. Do not treat a rounded per-second figure as an exact invoice calculation or infer unlisted output dimensions from it.
US-hosted variants have higher rates, approximately 20% above standard hosting. The table above applies to standard hosting.
What do the published worked cost examples show?
The fal text-to-video and image-to-video pages provide the following examples. They offer concrete checkpoints for a cost estimator using the token formula.
| Generation | Published dimensions | Tokens | Approximate cost |
|---|---|---|---|
| 5 seconds, 720p, 16:9 | 1280 × 720 | 108,000 | $2.31 |
| 30 seconds, 720p, 16:9 | 1280 × 720 | 648,000 | $13.87 |
| 10 seconds, 480p, 16:9 | 864 × 496 | 100.44 thousand | $2.15 |
For the five-second 720p example, multiplying 1280 × 720 × 5 × 24 and dividing by 1024 gives 108,000 tokens. Applying $0.0214 per thousand tokens produces approximately $2.31. For budgeting, record the endpoint, resolution, aspect ratio, duration, and hosting choice alongside each estimate.
How do video references change the estimate?
When reference-to-video includes video inputs, the token formula becomes: tokens = (output_height × output_width × (input_video_duration + output_duration) × 24) / 1024. The price is then multiplied by 0.6.
At 720p, the published approximation with video references is $0.2838 per second, applied to both input and output seconds. An estimator that counts only the requested output duration would therefore miss part of the billable duration.
Image and audio references are not billed; input video seconds are billed. Keep these input types separate in your cost calculation. Reference assets and generated audio have different billing roles, even though both may appear in the same generation workflow.
How does our docs API compare with fal billing?
Our API documentation uses the public model ID seedance-2.5. Its billing system uses website credits, while fal uses USD token billing. These identifiers and billing units belong to separate integrations.
For our API, POST /quotes returns cost_credits, and GET /credits/balance provides the credit balance. Consult models and pricing for the credit-based workflow. Do not apply fal token rates to a credit quote or assume a USD-to-credit conversion.
Our documented account limits include one active API generation, ten generation/upload requests per minute, and a daily API generation credit budget of 2,000. These are our API limits; they do not establish fal limits. See the Seedance 2.5 API pricing and limits guide for related planning.
What are the frequently asked questions (FAQ)?
What is the text-to-video model ID?
Use bytedance/seedance-2.5/text-to-video. The corresponding REST URL is https://fal.run/bytedance/seedance-2.5/text-to-video.
Does disabling generated audio reduce the cost?
No. The verified pricing is the same with or without generated audio. The generate_audio parameter defaults to true.
Which durations can I request?
The verified text-to-video duration setting accepts auto or 4–30 seconds. Set a specific duration when you need a concrete cost estimate before generation.
Are image and audio references charged?
Image and audio references are not billed. Video reference inputs contribute billable input duration, and requests with video inputs use the 0.6 price multiplier.
Can I use seedance-2.5 as the fal endpoint ID?
The verified fal routes use the full bytedance/seedance-2.5/ path plus the workflow name. The shorter seedance-2.5 identifier is the public model ID for our docs API.
Which estimate should my application display?
Use the token formula and applicable rate for a request-specific estimate. Label per-second planning figures as approximate, and include input video duration when estimating reference-to-video requests.