How to Use the Seedance 2.0 API: Pricing, Variants, and Global Access

ByteDance opened the Seedance 2.0 API on April 16, 2026, through Volcano Engine in China and BytePlus everywhere else. The early production numbers are the kind that make a render farm nervous: studios report going from 100+ attempts for a 2-second shot to 3 attempts for a 15-second long take, and one brand campaign passed 20 million plays in 60 hours.

It ships as four variants (I2V, Fast I2V, V2V, Fast V2V), and picking the wrong one costs you latency, quality, or both. Below: what each variant is for, how the async lifecycle works on Modellix, what each tier costs, and how to call it without a China-region account.

Seedance 2.0 Capabilities Explained: I2V, V2V, and 4 Key Production Advantages

Most AI video APIs accept a text prompt or a single image. Seedance 2.0 accepts four modalities simultaneously: text, image, audio, and video. That changes the class of applications you can build.

Practically speaking, this means you can pass a reference image of a character, an audio track, a rough video clip as motion guidance, and a text prompt , and get a single coherent output. The model's instruction-following in complex, multi-subject scenes is measurably better than previous generations, which matters when you're producing at scale and can't afford high discard rates.

Key capabilities that translate directly to production workflows:

• Long-take consistency. Seedance 2.0 can generate extended single-take sequences and continue them coherently from a prior clip. The model reads the motion rhythm of the previous segment and maintains it , no jump cuts from model drift.
• Physical accuracy. Lighting, collision dynamics, and real-world physics hold up across complex scenes. This is what enables use cases like robotics simulation data generation, where physical plausibility isn't aesthetic , it's a hard requirement.
• Portrait and copyright safety infrastructure. ByteDance built face-verification and portrait authorization flows into the platform. For enterprise workflows that deal with identifiable persons, this matters for compliance.
• High-fidelity virtual human library. Over 10,000 pre-cleared virtual avatars covering different ages and professions are available via the API, removing a major production bottleneck for teams that need licensable characters.

See it in action in ByteDance's official Seedance 2.0 demo:

4 Variants Compared: I2V, Fast I2V, V2V, and Fast V2V Decision Guide

Seedance 2.0 is not a single model , it's a family of four API endpoints, each serving a different input mode and speed/quality tradeoff. Calling the wrong one for your use case wastes cost or produces lower quality output than the model is capable of.

A few things worth noting from the API spec:

I2V vs Fast I2V: Both accept 1-9 reference images and up to 3 audio tracks alongside a text prompt. Fast I2V trades some quality for significantly lower latency. Use I2V for final production renders, Fast I2V for development passes and high-concurrency pipelines where you're running many jobs in parallel.

V2V vs Fast V2V: Both accept 1-3 reference videos, 1-9 images, and 1-3 audio tracks. V2V is the full-quality variant , this is what you use for the long-take extension use case where you're feeding a prior output back in to continue a sequence. Fast V2V is appropriate for rapid iteration on motion retargeting workflows.

Resolution cap: All four Seedance 2.0 variants currently max at 720p via the Modellix API. The resolution parameter supports 480p and 720p. If your pipeline requires 1080p output, plan for post-processing upscaling.

Duration control: The duration parameter accepts -1 for auto-length or explicit values from 4 to 15 seconds. For the 42-second long-take use case, chain multiple V2V calls rather than setting a single long duration , feed each output back as the reference video for the next call.

Audio generation: All variants include generate_audio: true by default. Set to false if you're handling audio separately in your pipeline to avoid redundant processing cost.

Newer routes: text-to-video and Mini. The 2.0 family on Modellix has grown beyond the four variants above. Text-to-video (seedance-2.0-t2v) takes a prompt with no reference media, and the Mini tier (seedance-2.0-mini-t2v, -mini-i2v, -mini-v2v) is the lowest-cost option for drafts and bulk jobs. The request lifecycle is identical, so adding either one is a slug change. Check each model page for the resolutions and per-second rate it currently exposes.

API Request Lifecycle: The 3-Step Submit, Poll, and Retrieve Pattern

This section is for developers integrating Seedance 2.0 into production media workflows. Creative and product readers can use the model guidance above; backend teams should use the submit, poll, and retrieve pattern below for reliable automation.

Every Seedance 2.0 variant on Modellix uses the same async job pattern: submit, poll for status, retrieve results. The endpoint structure is consistent across all four variants, so the integration code you write for I2V works unchanged for V2V.

Step 1 , Submit the Job

Real Modellix endpoint for I2V (image-to-video):

curl --request POST \
  --url https://api.modellix.ai/api/v1/bytedance/seedance-2.0-i2v/async \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "image": [
      {
        "url": "https://example.com/character.jpg",
        "role": "first_frame"
      }
    ],
    "prompt": "A swift over rooftops at golden hour, cinematic long-take",
    "duration": 10,
    "resolution": "720p",
    "generate_audio": true
  }'

For V2V (video continuation), swap the endpoint and pass a video reference instead:

curl --request POST \
  --url https://api.modellix.ai/api/v1/bytedance/seedance-2.0-v2v/async \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "video": [
      {
        "url": "https://example.com/previous-clip.mp4"
      }
    ],
    "prompt": "Continue the motion, same lighting, camera pulls back slowly",
    "duration": 10
  }'

The response is identical across all variants:

{
  "code": 0,
  "message": "success",
  "data": {
    "status": "pending",
    "task_id": "task-abc123",
    "get_result": {
      "method": "GET",
      "url": "https://api.modellix.ai/api/v1/tasks/task-abc123"
    }
  }
}

Pass large artifacts (videos, high-res images) as URLs rather than inline. For multimodal inputs combining image + audio + text, keep all references under the same request body , the model processes them as a unified context, not sequential inputs.

Step 2 , Poll for Status

Use the get_result.url from the submission response. Categorize responses into three buckets:

Recommended cadence: first check at 15 seconds, then exponential backoff starting at 5s, capped at 30s, maximum 12 total attempts. Add ±20% jitter when running concurrent jobs.

Step 3 , Retrieve and Validate Results

Log at minimum: task_id, your correlation ID, input_hash, output URL, estimated cost, and wall-clock time from submit to terminal state. If you're chaining V2V calls for long-take generation, store the output URL immediately , it becomes the video.url for the next call in the sequence.

Seedance 2.0 API Pricing: Per-Second Billing, Resolution Tiers, and Cost Breakdown

Seedance 2.0 API pricing on Modellix is billed per second of generated video. The rate varies by variant and resolution tier. Key things to understand:

Standard vs Fast variants have different price points. Fast I2V and Fast V2V are cheaper per second than their standard counterparts , you're trading output quality for cost. For prototyping and high-volume pipelines, default to Fast variants. Switch to standard I2V or V2V only for final production renders.

You pay for output, not input. Passing 9 reference images or 3 audio tracks doesn't change your cost , only the generated video seconds are billed. A job that fails at content-policy check costs a small fraction of a completed render. This means your discard rate directly drives your monthly bill, which is why Seedance 2.0's higher consistency has real financial impact beyond just quality.

Resolution is currently capped at 720p. All four Seedance 2.0 variants on Modellix support up to 720p. If 1080p is a requirement, factor in upscaling costs when modeling your pipeline economics.

Duration range is 4 to 15 seconds per call. For longer sequences, you chain V2V calls. Each call in the chain is billed independently , a 45-second sequence = three 15-second V2V calls = 3× the per-call rate.

For current per-second rates across all Seedance 2.0 variants, check Modellix pricing directly. Rates change, and stale numbers are worse than no numbers.

BytePlus vs Volcano Engine: How to Access the Seedance 2.0 API Without a China Account

ByteDance's international cloud arm, BytePlus, has launched Seedance 2.0 API access simultaneously with the domestic Volcano Engine release. This matters for international teams: no Chinese entity, no Volcano Engine account, no China-region infrastructure required.

BytePlus operates under a different endpoint base URL and uses separate API keys from Volcano Engine. If you're building globally distributed products, this is the path you want , compliance, data residency, and support all differ from the domestic offering.

Practical limitations to verify before committing:

• Portrait authorization and face verification features may have different availability or require separate setup on BytePlus vs. domestic
• The virtual human library (10,000+ pre-cleared avatars) access terms differ between regions , confirm licensing for your specific use case
• Rate limits and SLAs on BytePlus may differ from domestic Volcano Engine; benchmark your specific workload before finalizing architecture decisions

What Direct Access Requires: Activation, Model IDs, and Endpoints

On the official route, an account and an API key are not enough to call Seedance 2.0. The models sit behind an activation step: BytePlus requires an account balance above USD 30 or a purchased Seedance resource pack before the 2.0 series can be activated in the console, and Volcano Engine applies the same structure domestically with a balance of at least 200 CNY. Key creation and model activation are separate console actions, which is why a valid new key can still fail.

The specifics to pin down before a direct integration (per the BytePlus docs):

• Base URL: https://ark.ap-southeast.bytepluses.com/api/v3 (region ap-southeast-1). Video jobs are async: POST .../contents/generations/tasks, then GET .../tasks/{id} until the status is succeeded or failed.
• Model IDs are version-suffixed: dreamina-seedance-2-0-260128 (2.0), dreamina-seedance-2-0-fast-260128 (Fast), and dreamina-seedance-2-0-mini-260615 (Mini). ByteDance rotates the suffix when a model is updated, so read the ID from configuration rather than hard-coding it.
• Reference limits: up to 9 images, 3 videos (2 to 15 seconds each), and 3 audio clips per request. Audio cannot be the only reference; include at least one image or video with it.
• Real faces: reference images or videos containing real human faces are rejected when uploaded directly. Portrait workflows go through the platform's asset library (asset://<ASSET_ID>).
• Output URLs expire: the returned video_url is time-limited, so download the file as soon as the task succeeds.

Single-Endpoint Access: One API Key for All Major AI Video Models

Running Seedance 2.0 via Volcano Engine or BytePlus means managing separate accounts, separate API keys, separate billing dashboards, and separate retry/polling logic for each model you integrate. If your product uses Seedance 2.0 for video plus Seedream for image generation plus Wan for audio-video sync, you're juggling three vendor relationships, three cost centers, and three debugging surfaces.

Modellix solves this with a unified AI media API: one endpoint, one API key, one billing dashboard , and consistent async job patterns across every model. Seedance 2.0, Kling, Seedream, Hailuo, Wan, Flux , all accessible via the same integration pattern. The async polling logic, idempotency key architecture, and observability setup you built for one model works for all of them.

For international teams specifically, Modellix handles BytePlus and Volcano Engine routing transparently. You don't manage regional endpoints or account splits , the platform abstracts that layer.

From an enterprise compliance perspective: Modellix's parent company JG Group is NASDAQ-listed, which matters when procurement teams ask about vendor stability, audit trails, and data handling commitments. Full billing history, per-job cost logging, and transparent per-model pricing are available in the dashboard without filing a support ticket.

Troubleshooting Seedance 2.0 API Key and Access Errors

Most failed first calls are access problems, not model problems. Keys are not interchangeable: a BytePlus key only authenticates on the BytePlus endpoint, a Volcano Engine key only on the domestic endpoints, and a Modellix key only on api.modellix.ai.

Treat the key like a password: keep it in an environment variable, never commit it or a .env file to source control, never ship it in a client bundle, rotate it on a schedule and immediately after any suspected leak, and check per-key usage daily so a compromised key is caught before it burns budget. There is no legitimate "API key generator": keys are only issued from a provider console you log into.

4 Reliability Patterns for Production Async Video API Pipelines

Once you're past the proof-of-concept stage, these patterns reduce operational pain significantly.

Separate Your Retry Buckets

Keep two distinct retry queues: transient failures (5xx, network timeout before server acknowledgment) and permanent failures (invalid input, quota exceeded, content policy). Auto-retry the first queue with backoff. Alert and stop on the second. Mixing them is how you build a silent money-burning loop.

Validate Inputs Locally Before Submitting

Before any network hop, run three checks:

1. Schema: Required fields present, types correct, enums valid
2. References: Do all URLs resolve? Do they meet size and type constraints? Run a preflight HEAD request and attach content-length
3. Logical consistency: If your prompt says "close-up portrait," don't also pass wide_angle: true. These contradictions generate bad outputs, not errors , and you'll burn render budget diagnosing them

Log Correlation IDs Through the Full Job Lifecycle

Tag every submit, poll, and result-fetch call with the same correlation ID. When a job silently fails at step 2 of a 5-step pipeline, you need to find it in logs without rebuilding the crime scene. OpenTelemetry semantic conventions are a good reference if you're setting this up fresh.

Monitor Cost Slope, Not Just Cost Total

Log estimated cost per job including inputs, outputs, and retries. Roll it up with P50 and P95 views weekly. Total monthly spend is a lagging indicator. P95 cost-per-job tells you if a new input type or resolution setting is quietly becoming expensive before it shows up on the invoice.

4 Verified Production Deployments: Studio, Brand, Robotics, and Digital Human Workflows

The model isn't theoretical. As of early 2026, production deployments include:

Short drama and AI film production. Content studios report 80-90% efficiency gains across the full workflow from script to edit. Renders that previously required 100+ attempts for a 2-second shot now average 3 attempts for 15-second long-takes.

Brand campaign video. Seedance 2.0's precise lighting generation and dynamic control have enabled product marketing videos with physics and motion that would require heavy-rig physical shoots. One OPPO product launch video hit 20M plays within 60 hours of release.

Robotics and autonomous vehicle training data. Multiple enterprise teams use Seedance 2.0 to generate physically accurate simulation data covering edge-case scenarios , extreme weather, complex collisions , that are expensive or impossible to capture in real-world recording.

Digital human content creation. Workflows that previously required image generation → animation → lip-sync as three separate model calls collapse to a single Seedance 2.0 call with a character reference image and audio input.

Modellix collection

Restyle and Edit Existing Video

Compare video-to-video models for controlled restyling, scene edits, and reusable production workflows.

Good first tests: Seedance 2.0 Mini V2V / Gemini Omni Flash Video Edit / HappyHorse Video Edit

Explore the collection

How to Test Seedance 2.0 in 30 Seconds

Every method starts with a Modellix account. Sign in with email, Google, or GitHub, add credit, and pay only for what you generate. There is no subscription.

Create your Modellix account →

Method 1: Playground

No code. Open the Seedance 2.0 T2V model page, paste a prompt, and review the clip in your browser.

Method 2: API, Through Your Coding Agent

Paste this into Claude Code, Codex, or Cursor. One sentence and one API key are all your agent needs: change the model name and docs link, and the same prompt batch-generates assets with any model on Modellix.

My Modellix API key is mdlx-••••••••••••••••••••••••. Read https://docs.modellix.ai/get-started/index.md and https://docs.modellix.ai/bytedance/seedance-2-0-t2v.md, then use Seedance 2.0 T2V to generate one 5-second 16:9 clip for each prompt in prompts.txt. Poll each task until it finishes and save the MP4 files to ./clips.

Swap the masked key for your own from the Modellix console. Treat it like a password and share it only with agents you trust. The API guide covers the raw request format.

Method 3: Skill

Install the Modellix skill once. After that, your agent finds the right model and request fields by itself.

npx skills add https://github.com/Modellix/modellix-plugin --skill modellix

Use the Modellix skill to generate a 5-second clip with Seedance 2.0 T2V, then download it to ./clips.

Setup for each agent host is in the Skill guide.

Method 4: CLI

For scripts and scheduled jobs. One command submits the task, waits, and prints the result URL.

npm install --global modellix-cli
export MODELLIX_API_KEY="your_api_key"

modellix-cli model run --model-slug bytedance/seedance-2.0-t2v \
  --body '{"prompt":"A chef plates a dessert in a busy kitchen, warm cinematic lighting"}' \
  --wait --timeout 15m --quiet

For batches, put one request per line in a JSONL file and run modellix-cli model batch. See the CLI guide.

Frequently Asked Questions About Seedance 2.0 API (2026)

What is the difference between Seedance 2.0 I2V, Fast I2V, V2V, and Fast V2V?

The four variants split along two axes: input mode and speed tier. I2V (image-to-video) takes reference images as primary input; V2V (video-to-video) takes reference video clips. Within each mode, the Fast variant is optimized for lower latency at some quality tradeoff. Use standard variants for final production renders, Fast variants for development and high-concurrency pipelines.

What's the difference between Seedance 2.0 on Volcano Engine vs. BytePlus?

Volcano Engine is ByteDance's domestic China cloud. BytePlus is the international offering. Both carry Seedance 2.0 API access. If your infrastructure is outside China, BytePlus is the correct direct path. Modellix abstracts both , you use a single endpoint regardless of region.

Does Seedance 2.0 support image-to-video?

Yes. The I2V and Fast I2V variants accept 1-9 reference images alongside a text prompt and up to 3 audio tracks. Images can be assigned roles (e.g., first_frame) to control how the model uses them.

Can I extend a Seedance 2.0 video output with another API call?

Yes, using the V2V variant. Pass your previous output URL as the video reference in a new V2V call. The model reads the motion rhythm and visual style of the input clip and continues it. Chain multiple V2V calls to build sequences beyond the 15-second single-call limit , this is the mechanism behind 42-second single-take outputs.

What content is blocked by Seedance 2.0's safety system?

ByteDance has built portrait and copyright safety standards into the platform covering all input modalities. Real-person likenesses require face verification and portrait authorization flows before they can be used as reference inputs. Plan your content pipeline around this if you're building any product involving identifiable individuals.

Why does my Seedance 2.0 API key return 403?

On the official route, almost always because the model was never activated. Activation requires the balance or resource-pack condition plus a separate console step; until it completes, a valid key still cannot call the model. The fix is the activation step, not a new key.

How do I access Seedance 2.0 on Modellix?

Get an API key at modellix.ai/console/api-key. All four Seedance 2.0 variants are available immediately under the bytedance/ namespace. The endpoint pattern is https://api.modellix.ai/api/v1/bytedance/seedance-2.0-{variant}/async. No separate BytePlus or Volcano Engine account required.

How does Modellix pricing compare to going direct to BytePlus?

Every Seedance 2.0 variant has its price on the Modellix pricing page. The bigger win is boring in the best way: one API key, one bill, and the same async job pattern whether you call Seedance 2.0, Kling, Seedream, or Hailuo.