Skip to content
Model infoModelsSnapshot: 15 Oct 2026

ByteDance / DreamActor

DreamActor-H1

Research model for product-demonstration videos from separate human and product images, with explicit motion guidance and identity preservation.

Video generation
restrictedClosed weightsVideo generation
Official documentation (opens in a new tab)Reviewed 15 Oct 2026
8 of 24 specification fields collected
Output resolutionNot collectedAwaiting source review
Clip durationNot collectedAwaiting source review
Input billingNot collectedByteDance research
Output billingNot collectedNo inferred price

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
Not collected
Batch processing
Not collected
Fine-tuning
Not collected
Context caching
Not collected
Parameters
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Languages
Not collected
Output resolutions
Not collected
Output formats
Not collected
Input modalities
Image, TextSource for Input modalities (opens in a new tab)

Paired human/product images plus semantic descriptions in the research pipeline. No public API input schema established.

Output modalities
VideoSource for Output modalities (opens in a new tab)
Reference images
One human image and one product imageSource for Reference images (opens in a new tab)
Base model
Seaweed-7B image-to-videoSource for Base model (opens in a new tab)

The backbone name does not establish the parameter count of the full adapted system.

Architecture
DiT with masked product attention, separate appearance references, body-mesh and object-box motion guidance, plus semantic text conditioning.Source for Architecture (opens in a new tab)
Published training data
15,000 paired demonstrations, about 80 hours and 5,000 products, provide 80% of samples. Auxiliary live-commerce and Seaweed videos provide the remaining 20%.Source for Published training data (opens in a new tab)
Training hardware
24 NVIDIA H20 GPUs, two weeks, 100,000 stepsSource for Training hardware (opens in a new tab)
Published research configuration
720 × 1280; 65-frame segments chained through the previous latent; appearance CFG 2.5.Source for Published research configuration (opens in a new tab)

Published experiment configuration, not a public API resolution or duration limit.

Access & pricing

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

ByteDance research

Provider source (opens in a new tab)
API model ID
Not collected

No public H1 API selector established.

Access
Research disclosure only; no verified public inference service or weight release.
Regions
Not collected

Endpoint pricing has not been collected. Check the provider for applicable rates.

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.

ByteDance research

Not verified

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

How it works
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

Minimum reusable input
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

Lifetime & refresh
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

How to enable it
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

Usage fields
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

Billing rules
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

Limits & data boundaries
Not collected

No documented discounted prompt-cache contract. Task lifetime and latent reuse do not establish caching support.

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 original ByteDance research project, a complete public repository inventory containing website assets rather than weights, and reviewed provider weight inventories without a corresponding release. No explicit proprietary weight license is asserted.

Classification evidence (opens in a new tab)

Lifecycle dates

Research paper first submitted12 Jun 2025Source (opens in a new tab)
Research paper revised27 Aug 2025Source (opens in a new tab)

Sources reviewed