ByteDance / Seed3D

Seed3D 2.0

Image-to-3D generation with two-stage geometry refinement and unified PBR material synthesis. Native access exports textured objects in four file formats.

3D generation

availableClosed weights3D generation

Official documentation (opens in a new tab)Reviewed 15 Oct 2026
6 of 20 specification fields collected

3D output formatsGLB, OBJ, USD, USDZSee specification notes below

Polygon count100,000 / 500,000 / 1,000,000 faces; default 500,000See specification notes below

Input billingSee ratesVolcengine Ark

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: 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
Input modalities: ImageSource for Input modalities (opens in a new tab)
Output modalities: 3DSource for Output modalities (opens in a new tab)
Architecture: Coarse-to-fine geometry DiTs with a locality-aware VAE; unified multi-view albedo and metallic-roughness generation using MoE and semantic conditioning.Source for Architecture (opens in a new tab)
Output formats: GLB, OBJ, USD, USDZSource for Output formats (opens in a new tab)
Polygon count: 100,000 / 500,000 / 1,000,000 faces; default 500,000Source for Polygon count (opens in a new tab)
Published research configuration: Research geometry extraction uses a coarse 512³ grid and refinement up to 1536³. Hosted polygon counts are a separate output control.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 Ark

Provider source (opens in a new tab)

API model ID: doubao-seed3d-2-0-260328
Access: Activate the hosted model and use an Ark API key. Supports the model selector or a configured Endpoint ID. No inference performed.
Regions: cn-beijing
API endpoint URL: https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasksSource for API endpoint URL (opens in a new tab)
Authentication: Bearer Ark API keySource for Authentication (opens in a new tab)
Generation mode: Asynchronous POST creates a task ID; GET with the task ID polls status and returns file_url.Source for Generation mode (opens in a new tab)
Supported input workflows: One image plus optional generation parametersSource for Supported input workflows (opens in a new tab)
Image input limits: JPEG, PNG, WebP, BMP, TIFF or GIF; width and height 300–6000 px; aspect ratio 0.4–2.5; under 30 MB per image.Source for Image input limits (opens in a new tab)
Input media transport: Public URL or base64 data URL; request body at most 64 MB. Avoid base64 for large files.Source for Input media transport (opens in a new tab)
Output formats: GLB, OBJ, USD, USDZSource for Output formats (opens in a new tab)
Polygon count: low=100,000; medium=500,000 (default); high=1,000,000Source for Polygon count (opens in a new tab)
Generation controls: file_format and subdivision_level fields. Documentation also describes --fileformat and --subdivisionlevel inside text, with weak validation.Source for Generation controls (opens in a new tab)
Configuration discrepancy: Generic content schema permits text-only input, while model-specific capabilities list single-image generation. Do not assume a text-to-3D contract.Source for Configuration discrepancy (opens in a new tab)
Published request quota: Maximum 300 requests/minute and five concurrent tasksSource for Published request quota (opens in a new tab)
Result and task expiry: Task history: seven days. Generated files are cleared after 24 hours; download or copy them before expiry.Source for Result and task expiry (opens in a new tab)
Cancellation and task deletion: Only queued tasks can be cancelled; cancelled records are deleted after 24 hours.Source for Cancellation and task deletion (opens in a new tab)
Output billing: Count successful output files. Published price basis: 30,000 3D output accounting units per file at CNY 0.80 per 10,000 units.Source for Output billing (opens in a new tab)
Usage accounting: usage.completion_tokens and usage.total_tokens report 3D billing units; input is counted as zero, so both totals are equal.Source for Usage accounting (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
3D generation	Textured 3D asset with PBR materials¹	CN¥2.40	successful 3D file

Rate conditions

Native published price per successfully generated file. Underlying 3D accounting uses 30,000 output units at CNY 0.80 per 10,000 units. This is not an LLM token tariff or a cost-per-question estimate.

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 Ark

doubao-seed3d-2-0-260328

Not verified

No discounted prompt-cache contract established for this 3D generation model.

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 original ByteDance disclosures, the foundation-model repository containing only a report/README, and current Seed3D, ByteDance and ByteDance-Seed weight inventories without a matching 1.0 or 2.0 release. No explicit proprietary weight-license declaration is asserted.

Classification evidence (opens in a new tab)

Lifecycle dates

Technical report published22 Apr 2026Source (opens in a new tab)

Provider release announcement23 Apr 2026Source (opens in a new tab)

Sources reviewed

seed3d2 api (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 guide (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 get api zh (opens in a new tab)Reviewed 15 Oct 2026

volcengine native model list (opens in a new tab)Reviewed 15 Oct 2026

volcengine native pricing (opens in a new tab)Reviewed 15 Oct 2026

volcengine native releases (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 paper (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 fullpaper (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 hf paper (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 launch (opens in a new tab)Reviewed 15 Oct 2026

seed3d2 publication (opens in a new tab)Reviewed 15 Oct 2026

seed3d weight inventory (opens in a new tab)Reviewed 15 Oct 2026

seed3d1 tree (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

Model notes

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

Uses the official ByteDance Seed provider icon. No distinct verified model logo was established.

Seed3D is a multi-stage generation system. Published component sizes do not establish the parameter count of the complete system.

Research components, scene planning and articulation demos are not necessarily exposed by the hosted single-image generation API.

Open checkpoints for Dora-VAE, MagicArticulate and Puppeteer do not constitute a Seed3D 1.0 or 2.0 foundation-model weight release.

Training fine-tuning, mesh formats and asynchronous JSON transport do not establish customer fine-tuning, LLM structured output or streaming support.

The dated API suffix is a version identifier, not proof of a March 28 public launch.

The generic content schema mentions text-only requests, but the model-specific capability and guide specify one-image generation. Text-only generation is not asserted.

Native 3D billing reports output units called tokens. They are not text-token capacity, and do not yield a text cost-per-question estimate.

The published scene and articulation workflows can use other models. Native URDF export, scene-layout inputs and independently callable research components are not established.

Check against your workload

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

Read our methodology