ByteDance / FlowAct

FlowAct-R1

Streaming character video from a reference image, live audio and optional action prompts. Native access requires reserved concurrency and ByteRTC integration.

Video generation

restrictedClosed weightsVideo generation

Official documentation (opens in a new tab)Reviewed 15 Oct 2026
8 of 22 specification fields collected

Output resolution480pSee specification notes below

Clip durationNot collectedAwaiting source review

Input billingSee ratesVolcengine realtime avatar API

Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit: Not collected
Maximum input tokens: Not collected
Maximum output: Not collected
Reasoning: Not collected
Tool calling: Not collected
Structured output: Not collected
Streaming: SupportedSource for Streaming (opens in a new tab)
Batch processing: Not collected
Fine-tuning: Not collected
Context caching: Not collected
Parameters: Not collected
Architecture: Seedance-based MMDiT adapted with chunkwise diffusion forcing, latent memory and few-step distillation.Source for Architecture (opens in a new tab)
Knowledge cutoff: Not collected
Release date: Not collected
Languages: Not collected
Output resolutions: 480pSource for Output resolutions (opens in a new tab)
Output formats: Not collected
Output modalities: VideoSource for Output modalities (opens in a new tab)
Input modalities: Image, Audio, TextSource for Input modalities (opens in a new tab)
Base model: Seedance family; exact native checkpoint not specifiedSource for Base model (opens in a new tab)
Frame rates (fps): 25 fpsSource for Frame rates (fps) (opens in a new tab)
Published research configuration: Paper reports approximately 1.5-second first frame on NVIDIA A100 at 480p and 25 fps.Source for Published research configuration (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Volcengine realtime avatar API

Provider source (opens in a new tab)

API model ID: flowact-r1
Access: Commercial resource confirmation and a reserved-concurrency purchase are required. ByteRTC integration is separate from the model identity. No session was opened.
Regions: Not collected
API endpoint URL: wss://openspeech.bytedance.com/virtual_human/avatar_live/liveSource for API endpoint URL (opens in a new tab)
API model selector: avatar.avatar_type=flowact-r1; input_mode=audioSource for API model selector (opens in a new tab)
Authentication: Volcengine account plus signed URL token and ByteRTC application/room credentials.Source for Authentication (opens in a new tab)
Input audio format: 16 kHz mono 16-bit PCM; chunks at least 40 ms / 1280 bytes; binary or base64 JSON frames.Source for Input audio format (opens in a new tab)
Output formats: ByteRTC video stream; WebSocket carries audio, controls and status.Source for Output formats (opens in a new tab)
Connection heartbeat: Server ping every five seconds; client responds with pong. Browser clients also receive text heartbeat events.Source for Connection heartbeat (opens in a new tab)
Live session controls: Unique live_id. Disconnect keep_alive_duration defaults to 0; positive values set seconds and negative values retain the concurrency indefinitely.Source for Live session controls (opens in a new tab)
Prompt controls: Action prompts can accompany audio; optional text_list cycles preset actions.Source for Prompt controls (opens in a new tab)
Image input limits: JPEG/JPG/PNG; 480 × 480 through 1920 × 1080.Source for Image input limits (opens in a new tab)
Aspect ratios: 1:1, 3:4, 4:3, 9:16, 16:9Source for Aspect ratios (opens in a new tab)
Output resolutions: 480pSource for Output resolutions (opens in a new tab)
Frame rates (fps): 25 fpsSource for Frame rates (fps) (opens in a new tab)
Configuration discrepancy: First-frame latency: about two seconds in introductory claims versus about five seconds in the output table; excludes the speech-conversation module.Source for Configuration discrepancy (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
Category	Usage	Rate	Unit
Reserved concurrency	One concurrent stream reservation¹	CN¥30,000.00	concurrent stream / month
Reserved concurrency	One concurrent stream reservation²	CN¥300,000.00	concurrent stream / year

Rate conditions

Published reservation price for one API concurrency, subject to commercial resource confirmation. Not a per-minute, per-second or token price; SDK bundle pricing is distinct.
Published annual reservation price for one API concurrency, subject to commercial resource confirmation. No utilization-based conversion or SDK-bundle rate is inferred.

Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Volcengine realtime avatar SDK bundle

Provider source (opens in a new tab)

API model ID: Not collected
Access: Activate the realtime avatar SDK service and reserve concurrency after commercial resource confirmation. Bundle includes FlowAct-R1, Doubao realtime speech and RTC; it is not a separately identified model release.
Regions: Not collected; iOS configuration exposes domestic and overseas environments without an enumerated serving-region contract.
Supported input workflows: Reference image, conversational audio and a voice selection; optional action descriptions.Source for Supported input workflows (opens in a new tab)
Languages: Chinese, EnglishSource for Languages (opens in a new tab)
Output resolutions: 480pSource for Output resolutions (opens in a new tab)
Frame rates (fps): 25 fpsSource for Frame rates (fps) (opens in a new tab)
Output formats: Synchronized realtime avatar video and speechSource for Output formats (opens in a new tab)
Image input limits: JPEG/JPG/PNG; 480 × 480 through 1920 × 1080; clear frontal half-body images recommended.Source for Image input limits (opens in a new tab)
Aspect ratios: 1:1, 3:4, 4:3, 9:16, 16:9Source for Aspect ratios (opens in a new tab)
Live session controls: Long sessions can degrade visually. Set a maximum session duration and reconnect or split sessions as appropriate; no numeric duration ceiling is published.Source for Live session controls (opens in a new tab)
SDK support: Web, iOS and Android integrations linked in the product overview. iOS requires version 12.0 or later and BDAvatarKitToB.Source for SDK support (opens in a new tab)
Authentication: iOS recommends short-lived server-issued WSS tokens. Permanent AK/SK and temporary AK/SK + STS are also documented. Credential refresh affects the next session.Source for Authentication (opens in a new tab)
API model selector: SDK session uses avatarImageUrl and speaker. The overview identifies FlowAct-R1; the bundled speech checkpoint is not pinned.Source for API model selector (opens in a new tab)
Prompt controls: iOS: optional userPrompt and actionList. speechRate and loudnessRate range from -50 to 100, default 0.Source for Prompt controls (opens in a new tab)
Generation controls: iOS enableWebsearch defaults to false. Search is a bundle feature, not evidence of native FlowAct tool calling.Source for Generation controls (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
Category	Usage	Rate	Unit
SDK bundle reserved concurrency	One concurrent bundled avatar session¹	CN¥42,000.00	concurrent stream / month
SDK bundle reserved concurrency	One concurrent bundled avatar session¹	CN¥420,000.00	concurrent stream / year

Rate conditions

Reserved SDK-bundle concurrency, including avatar, realtime speech and RTC integration. Resource confirmation required; peak-time queues possible. Not the avatar-only API tariff and not convertible to per-token or per-second pricing without a separate contract.

Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Volcengine realtime avatar API

flowact-r1

Not verified

Session memory, retained media and model-internal latent buffers do not establish discounted prompt caching.

How it works: Not collected
Minimum reusable input: Not collected
Lifetime & refresh: Not collected
How to enable it: Not collected
Usage fields: Not collected
Billing rules: Not collected
Limits & data boundaries: Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

Volcengine realtime avatar SDK bundle

Not verified

No prompt-cache contract established for this SDK bundle.

How it works: Not collected
Minimum reusable input: Not collected
Lifetime & refresh: Not collected
How to enable it: Not collected
Usage fields: Not collected
Billing rules: Not collected
Limits & data boundaries: Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Inferred from the ByteDance model disclosures, official project repositories containing website assets rather than weights, and the reviewed ByteDance/ByteDance-Seed weight inventories with no matching checkpoint. No explicit weight-license declaration is asserted.

Classification evidence (opens in a new tab)

Lifecycle dates

Research paper first submitted15 Jan 2026Source (opens in a new tab)

Sources reviewed

volcengine flowact overview (opens in a new tab)Reviewed 15 Oct 2026

flowact native api docs (opens in a new tab)Reviewed 15 Oct 2026

flowact project (opens in a new tab)Reviewed 15 Oct 2026

flowact paper (opens in a new tab)Reviewed 15 Oct 2026

flowact fullpaper (opens in a new tab)Reviewed 15 Oct 2026

flowact tree (opens in a new tab)Reviewed 15 Oct 2026

flowact hf paper (opens in a new tab)Reviewed 15 Oct 2026

volcengine weight review seed (opens in a new tab)Reviewed 15 Oct 2026

volcengine weight review bytedance (opens in a new tab)Reviewed 15 Oct 2026

Volcengine realtime avatar SDK bundle: product and pricing (opens in a new tab)Reviewed 15 Oct 2026

Volcengine realtime avatar iOS SDK contract (opens in a new tab)Reviewed 15 Oct 2026

Model notes

67 reviewed records, including historical and research releases. Remaining model and detail coverage is under review.

Research publication dates are distinct from API release dates. No exact API launch date or model parameter count was established.

The linked Hugging Face paper pages and public website repositories are not model-weight releases. Website licenses do not determine the weight license.

No text-token price or text-question estimate is inferred from media duration or reserved-concurrency pricing.

A language model or speech generator can be connected externally. The avatar model is not thereby a TTS model, a chat model or a tool-calling agent.

Research latent memory and connection keep-alive are not evidence of prompt-cache discounts.

The native overview cites roughly two seconds in its introductory claims and about five seconds in its output table. Neither is an SLA; the paper reports a separate research measurement.

Uses the original FlowAct-R1 artwork from its research project. The research claim of arbitrary-length output is not a guaranteed session duration or quality commitment.

A separate SDK bundle adds Doubao realtime speech and RTC. Its CNY concurrency rates and Chinese/English conversation support are endpoint-specific; they do not change the native avatar model into a TTS or chat model.

Check against your workload

Try representative inputs, measure task success and total cost, then check the exact endpoint’s data policy.

Read our methodology