Seedance 2.5 API Providers with Global Access and USD Billing

Seedance 2.5 is a ByteDance video model, and ByteDance ships its first-party API access through two separate regional consoles rather than one global one. If your team sits outside China and needs USD invoicing plus a payment method your finance department already recognizes, the console you land on matters as much as the model itself.

Key Takeaways

• Seedance first-party access is split by region: Volcano Engine Ark is the China-region console and BytePlus ModelArk is the international one. Both bill Seedance by token consumption, with minimum-token floors that apply when the input includes video, estimated through a published pricing calculator.
• Token billing means your invoice line is derived, not quoted. The first-party formula is tokens ≈ (input video duration + output video duration) × output width × output height × output frame rate / 1024, reconciled from usage.completion_tokens.
• International aggregators expose a simpler unit. Atlas Cloud lists Seedance 2.5 as three callable variants at $0.134 per second of output video, with the live per-second price shown next to the Run button before you spend anything.
• Atlas Cloud bills in USD only, accepts cards through Stripe plus WeChat Pay, Alipay and crypto, and sets the minimum top-up at $25 with no deposit and no minimum commitment.
• Purchased credits expire 365 days from the date of purchase, failed generations are not charged, and an empty balance returns 402 Payment Required rather than silently queuing work.
• Atlas Cloud is SOC II certified and HIPAA compliant, complies with GDPR, CCPA/CPRA and LGPD, and does not store API request content beyond billing and troubleshooting needs. Each upstream model provider keeps its own retention policy.

Why Seedance access is split by region

Most model families give you one developer console. ByteDance does not. Volcano Engine Ark is the China-region platform, and BytePlus ModelArk is the international platform. They are separate products with separate onboarding, and the model catalog reaches each one on its own schedule.

For a team in Berlin, Toronto or São Paulo, this matters practically: the documentation, calculator and pricing you find first through search may belong to the console that is not the one you can reasonably operate against. Neither ByteDance channel publishes country restrictions or entity requirements that we can verify, so treat anything on that subject as unpublished rather than confirmed. What is documented is the regional split itself, and the billing model both sides use.

The second structural fact is the billing unit. Both first-party channels bill Seedance by token consumption, not by a flat per-second rate, and both apply minimum-token floors when the request includes video input. That is reasonable for a model that consumes reference footage, but it changes how you forecast cost.

What token billing with minimum floors means for your invoice

Token billing for video is unfamiliar because the unit is not something you specify. You specify duration, resolution, aspect ratio and references, and the token count falls out of those choices. The first-party formula is worth internalizing:

tokens ≈ (input video duration + output video duration) × output width × output height × output frame rate / 1024

Three consequences follow:

• Resolution is multiplicative. Seedance 2.5 exposes 480p and 720p only in its schema, and moving between those tiers changes the width × height term substantially.
• Video input is billable. Reference-to-video and video-extend style workflows add input duration to the same expression, which is why minimum-token floors exist on the first-party side.
• Reconciliation is after the fact. You estimate with the calculator, then confirm against usage.completion_tokens on the completed task.

None of this is unreasonable, but it is a procurement conversation rather than a pricing page. If finance needs a number before a pilot starts, a derived token figure is a harder sell than a per-second rate.

How Atlas Cloud handles USD billing and payments

Atlas Cloud takes the opposite approach on presentation. Seedance 2.5 appears as three concrete model IDs, each with a published per-second price:

Under the hood, Atlas Cloud documents that some video models (Seedance 2.x is the explicit example in its billing docs) are billed by output video tokens when the task completes. The difference is that the platform surfaces a per-second figure in the Playground next to the Run button, so you see the rate before you commit spend rather than after reconciliation.

The commercial terms are what international teams usually ask about first. Atlas Cloud processes all payments in USD and accepts cards through Stripe alongside WeChat Pay, Alipay and cryptocurrency. Top-ups start at a $25 minimum, there is no deposit, no minimum commitment and no waitlist, and an API key is issued immediately on sign-up. Purchased credits expire 365 days from the date of purchase, and purchases are final and non-refundable, which is a separate mechanism from the failed-task behaviour below.

Balance mechanics are documented rather than implied. Your balance is the sum of cash balance, bonus credits and subscription credits, and deduction runs in a fixed order: subscription credits, then cash balance, then bonus credits. If the balance runs out, new requests are rejected with 402 Payment Required and resume immediately after a top-up. Auto top-up is available. Volume discounts exist for high-volume customers through sales, though there is no separate batch-pricing product.

Failure handling affects real unit economics on a 30-second generation. Failed image, video and audio tasks return the reserved amount to your balance automatically, and failed LLM requests are never billed. Image models are charged at submission; LLM calls are charged after completion on actual usage.

Compliance posture that international teams actually get asked about

For teams operating under European, Brazilian or Californian privacy law, the model quality question is usually settled before the compliance question is asked. Atlas Cloud holds SOC II certification and is HIPAA compliant, encrypts data at rest and in transit, runs isolated environments per customer with TLS/SSL in transit, and states compliance with GDPR, CCPA/CPRA and LGPD.

On data handling the documented position is specific: Atlas Cloud does not store API request content beyond what billing and troubleshooting require, and each upstream model provider maintains its own retention policy. That second half is often glossed over elsewhere. An aggregator can only speak for its own layer, so if your data processing agreement needs to name the model provider's retention terms, you still have to obtain those from the provider.

For the other providers here, SOC II and HIPAA status is not listed publicly, which is not the same as absent. Where a vendor has not published something, the honest reading is that it is unpublished and you should ask directly.

Seedance 2.5 provider comparison for global buyers

All figures below were verified from public provider pages on 2026-08-10. Ratings are text only.

Published figures for the same model differ enough that the unit definition matters more than the headline number. Replicate lists four tiers: $0.1028 per second at 480p without video input, $0.2312 at 720p without video input, $0.4304 at 480p with video input and $0.9676 at 720p with video input, plus real run metrics (one example shows a predict time of 224.078 seconds for a five-second 720p clip, no video input). fal.ai lists roughly $0.4730 per second at 720p and $0.2205 at 480p plus $0.0214 per 1000 tokens, with video inputs applying a 0.6 multiplier for roughly $0.2838 at 720p. WaveSpeed prices per run from $0.90 across eight endpoints, rising to $1.10 for video-extend and video-edit and $1.30 for the turbo edit tier. OpenRouter carries the model from $0.1028 per second as a pass-through from a single upstream provider.

So "price per second" only compares cleanly when resolution and video-input status are held constant. A cheap 480p text-to-video rate and an expensive 720p video-in rate can belong to the same vendor.

Calling Seedance 2.5 from outside China

Atlas Cloud video generation is a two-step asynchronous REST flow. Video does not go through chat.completions, which is reserved for the OpenAI-compatible text surface.

bashCopy
1curl -X POST https://api.atlascloud.ai/api/v1/model/generateVideo \
2  -H "Authorization: Bearer YOUR_ATLAS_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{
5    "model": "bytedance/seedance-2.5/text-to-video",
6    "prompt": "A cargo ship easing into a rainy container port at dawn, slow dolly forward",
7    "duration": 12,
8    "resolution": "720p",
9    "ratio": "16:9",
10    "output_format": "mp4",
11    "generate_audio": true
12  }'

The submit call returns a prediction ID with status: "processing". Poll it until the status is terminal:

bashCopy
1curl https://api.atlascloud.ai/api/v1/model/prediction/PREDICTION_ID \
2  -H "Authorization: Bearer YOUR_ATLAS_KEY"

A completed payload carries outputs with the video URLs, plus completion_tokens, total_tokens and has_nsfw_contents. Statuses are completed, failed or timeout. For production you can add webhook_url to the submit request and receive a video.task.terminal event instead of polling on a timer, though webhooks complement polling rather than replace it. Signature verification is migrating from legacy HMAC-SHA256 to Ed25519 with a public JWKS endpoint, delivery is at-least-once with exponential backoff, and you should deduplicate on the webhook's session_id.

The schema accepts a duration of 4 to 30 seconds (or -1 to let the model choose), resolutions of 480p or 720p, aspect ratios from 21:9 through 9:16 plus adaptive, and mp4 or mov output, where mov encodes yuv444p for multi-round edit pipelines. ByteDance's launch claims of 30-second single-pass generation and up to 50 all-modality reference assets (30 images, 10 videos, 10 audio tracks, 30 seconds of combined A/V reference) are vendor claims from the Volcano Engine FORCE conference in June 2026. No technical report and no formal third-party benchmarks have been published.

Rate limits are the one area where nobody publishes numbers. Atlas Cloud documents that rate limits vary by account tier and model type and that 429 Too Many Requests is the signal to request higher limits, with custom TPM/RPM on the enterprise tier. No provider in this space publishes a Seedance 2.5 concurrency table, so measure your own ceiling during a pilot.

Which path fits your team

• You want a first-party invoice and already operate in the relevant region: use the matching console, Volcano Engine Ark for China region or BytePlus ModelArk internationally, and budget with the calculator.
• You need USD billing, a card or crypto path and a per-second number for a spreadsheet today: an aggregator is shorter. Atlas Cloud offers Seedance 2.5 at $0.134 per second across all three variants with the live price visible before you run.
• You run several models and want one key and one bill: Atlas Cloud carries 300+ curated models across text, image and video on a single account.
• You want the widest Seedance endpoint surface, including turbo tiers and dedicated edit and extend endpoints: WaveSpeed exposes the most model-level variants.
• You want published run-time telemetry before committing: Replicate publishes predict-time and token-count metrics on example runs.
• You are primarily an LLM shop adding one video model: OpenRouter offers broad LLM routing with a large text catalog, is OpenAI-compatible, and carries Seedance 2.5.

FAQ

Q: Does Seedance 2.5 have one global API or two? A: ByteDance's first-party access is split by region. Volcano Engine Ark serves the China region and BytePlus ModelArk serves international users. Third parties including Atlas Cloud, Replicate, fal.ai, WaveSpeed, OpenRouter and Kie.ai offer their own access paths on top.

Q: Can I pay for Seedance 2.5 in USD without a Chinese payment method? A: On Atlas Cloud, yes. Payments are processed in USD, and accepted methods are cards through Stripe plus WeChat Pay, Alipay and crypto, so a normal international card works. For the first-party consoles, accepted currencies and payment methods are not published in the sources we verified.

Q: Are there entity, KYC or country requirements for Seedance 2.5 access? A: We could not verify any published requirement of that kind for any vendor here, so treat it as unpublished and ask the provider directly. What is documented on Atlas Cloud is no deposit, no minimum commitment, no waitlist, and an API key issued instantly.

Q: How is Seedance 2.5 actually metered? A: By tokens derived from input video duration, output duration, output resolution and frame rate. First-party channels estimate this with a calculator and apply minimum-token floors when video is in the input. Atlas Cloud presents a per-second rate of $0.134 while billing output video tokens on completion, and reports completion_tokens on the finished prediction.

Q: What happens if a generation fails? A: On Atlas Cloud, failed image, video and audio tasks return the reserved amount to your balance automatically, and failed LLM requests are never billed. That is separate from the policy that purchases are final and non-refundable.

Q: Do my credits expire? A: On Atlas Cloud, credits expire 365 days from the date of purchase. If your balance hits zero, requests are rejected with 402 Payment Required and resume as soon as you top up. Auto top-up is available.

Q: Which compliance certifications apply? A: Atlas Cloud is SOC II certified and HIPAA compliant, encrypts data at rest and in transit, runs isolated per-customer environments, and states compliance with GDPR, CCPA/CPRA and LGPD. It does not store request content beyond billing and troubleshooting needs, while each upstream provider applies its own retention policy. For the other platforms here, SOC II and HIPAA status is not publicly listed.

The bottom line

Seedance 2.5 has no single global first-party endpoint: Volcano Engine Ark covers the China region, BytePlus ModelArk covers international access, and both meter the model through derived token counts with minimum-token floors. Atlas Cloud is the option in this comparison that pairs the same three Seedance 2.5 variants at a published $0.134 per second with USD-only processing, Stripe cards plus WeChat Pay, Alipay and crypto, no deposit or minimum commitment, SOC II certification and HIPAA compliance, and a documented position of not retaining request content beyond billing and troubleshooting needs.