To use Seedance 2.5 on Replicate, get your REPLICATE_API_TOKEN, select the model bytedance/seedance-2.5, set duration and resolution, and estimate the cost using the verified per-second prices below. Rates depend on resolution and input variant: non_video_in starts at $0.1028 per second at 480p, while video_in costs $0.4304 per second at that resolution.
Last verified: 2026-10-23
This guide covers the Seedance 2.5 Replicate setup, output-based pricing, and the distinction between image to video and video-reference workflows. Keep the provider’s available settings separate from the model’s capabilities elsewhere when designing your integration.
How do you set up Seedance 2.5 on Replicate?
Create a Replicate API token and store it in your server environment as REPLICATE_API_TOKEN. Use bytedance/seedance-2.5 as the model identifier. Start with a simple text prompt so you can check authentication and request handling before adding reference files.
The following curl example illustrates a request for a five-second, 480p output. Confirm the current duration constraints in the model’s API schema before adapting it for production.
export REPLICATE_API_TOKEN="your_replicate_api_token"
curl --request POST \
--url https://api.replicate.com/v1/models/bytedance/seedance-2.5/predictions \
--header "Authorization: Bearer $REPLICATE_API_TOKEN" \
--header "Content-Type: application/json" \
--header "Prefer: wait" \
--data '{"input":{"prompt":"A ceramic cup sits on a wooden table. Steam rises gently while the camera slowly moves closer. Keep the cup shape and lighting consistent.","duration":5,"resolution":"480p"}}'
Replicate authenticates requests with a Bearer token. Its HTTP API supports creating predictions through a model endpoint, and Prefer: wait requests a synchronous wait. Your application should still inspect the returned prediction status and retrieve the prediction again if processing remains incomplete. See the Replicate HTTP API reference.
Keep the token on your backend rather than embedding it in browser code. For broader credential guidance, read the Seedance API key guide. Credentials belong to the provider issuing them; do not assume another service’s key will authenticate a Replicate request.
What is the verified Replicate per second cost?
The verified Replicate Seedance 2.5 rates are charged per second of output video. Select the row matching both the input variant and output resolution.
| Input variant | Resolution | USD per second of output |
|---|---|---|
| non_video_in | 480p | $0.1028 |
| video_in | 480p | $0.4304 |
| non_video_in | 720p | $0.2312 |
| video_in | 720p | $0.9676 |
These figures come from the Replicate model listing, verified on 2026-10-23. They are specific to this model and provider.
Use this calculation:
Estimated generation cost = output duration in seconds × applicable USD-per-second rate
For a five-second output, the arithmetic is:
non_video_in, 480p: 5 × $0.1028 = $0.514video_in, 480p: 5 × $0.4304 = $2.152non_video_in, 720p: 5 × $0.2312 = $1.156video_in, 720p: 5 × $0.9676 = $4.838
These are calculated estimates from the verified rates, not additional price tiers. They describe one output of the stated duration, not an entire creative session with multiple attempts.
When planning a budget, include the number of outputs you expect to generate. Ten five-second outputs represent fifty output seconds. Estimate each group separately if your batch mixes resolutions or input variants, then add the totals.
Which resolutions does Replicate currently list?
Replicate’s model page/API enum currently lists 480p and 720p only, verified 2026-10-23. There is no verified Replicate 1080p price in this pricing snapshot.
Seedance 2.5 elsewhere supports 480p, 720p, and 1080p, but that broader capability does not establish availability through this particular endpoint. Your Replicate integration should offer the resolutions listed in its current API schema.
This distinction matters when building a shared interface for several providers. A resolution accepted by one provider may be unavailable through another. Validate the chosen resolution after the user selects a provider, and calculate the estimate from that provider’s matching rate.
For early prompt checks, consider 480p when its detail is sufficient. Move to 720p when the intended output needs it, and recalculate the estimate before submitting the request.
How should you budget for image to video?
For image to video, distinguish still-image references from video references. A still image is not itself a video input. The pricing labels separate non_video_in and video_in; check the actual request inputs against the model’s current billing description before selecting a rate.
Do not choose the video_in rate simply because the result is a video. Both variants produce video output. The relevant distinction is the input configuration.
A practical image workflow starts with a clear reference image and a prompt describing motion, camera behavior, and what should remain consistent. For example, specify that a product should retain its shape while the camera moves closer. Add the reference through the image field documented in the current model schema rather than guessing a parameter name.
For budgeting, record:
- Whether the request includes video input.
- The selected output resolution.
- The requested duration and resulting output duration.
- The applicable rate and the number of generation attempts.
This makes cost differences easier to explain when two requests have similar prompts but different reference materials. Avoid automatically retrying every unsuccessful request without checking its status and error details.
How does Replicate compare with fal and credit billing?
Treat Replicate, fal, and the API described in our documentation as separate integration and billing choices. This article’s dollar figures apply only to the verified Replicate listing.
The Seedance 2.5 fal API guide covers that provider’s setup. Do not reuse Replicate’s rates as a fal estimate; compare each provider using its own current pricing and supported inputs.
For the credit-billed integration, begin with the API documentation, then consult models and pricing. Estimate usage using the documented credit schedule. A Replicate dollar-per-second rate is not a credit conversion formula.
If you are deciding where to integrate, the API providers comparison and BytePlus API guide provide related reading. Compare authentication, available resolutions, reference inputs, and billing units before choosing. Use the API pricing and limits guide when reviewing application constraints.
FAQ
What is the model ID for Replicate Seedance 2.5?
Use bytedance/seedance-2.5. The owner and model name identify the endpoint shown in the setup example. Check its API schema when adding optional inputs.
What is the lowest verified per-second price?
The lowest listed rate is $0.1028 per output second for non_video_in at 480p. It does not apply to every request; resolution and input variant determine the applicable rate.
Can I request 1080p through Replicate?
The verified 2026-10-23 Replicate listing exposes only 480p and 720p. Do not assume 1080p is accepted or estimate a Replicate 1080p price from another provider’s offering.
Does image to video automatically use video-input pricing?
A still-image reference is not a video reference. Check the submitted input configuration and current model billing description. The presence of video in the output does not determine the input variant.
Can I try a video before integrating an API?
The web experience offers free credits on sign-up, enough for one video. This is separate from Replicate API billing and should not be treated as a Replicate account allowance.