Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Hyper3D / Deemos / Rodin

Rodin Gen-1 / 1.5

Rodin Gen-1 / 1.5 generates 3D geometry and materials from a prompt or up to five reference images. Mesh and texture controls depend on the generation family.

availableClosed weights3D generation
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
23 of 38 specification fields collected
3D output formatsGLB, USDZ, FBX, OBJ, STLSee specification notes below
Text input limit1K1,024
Input billingSee ratesHyper3D native API · text to 3d
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

API task JSON is not caller-defined structured model output.

Streaming
Not collected

Polling returns job status, not streamed mesh content.

Batch processing
Not collected

Parallel tasks and local request loops do not establish a native Batch API or batch discount.

Fine-tuning
Not collected

Enterprise pricing advertises custom LoRA/fine-tuning, but a model-specific self-service training contract is not documented.

Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Gen-1.5: more than 4 billion (provider-reported). Gen-1 count not established.Source for Parameters (opens in a new tab)

This fact applies only to the Gen-1.5 member of the combined API family.

Architecture
Not collected
Languages
Not collected
Context caching
Not collected

No prompt-cache interface or price documented.

Authentication
Bearer API key. Generation uses multipart/form-data; status/download use JSON.Source for Authentication (opens in a new tab)
Asynchronous generation
SupportedSource for Asynchronous generation (opens in a new tab)

Submit, poll every job to Done, then request download URLs. This is not progressive model-output streaming.

Request quotas
Business pricing lists 120 RPM on the first three subscription tiers and 240 RPM on higher tiers. Actual limits vary by account and endpoint; Retry-After on HTTP 429 takes precedence.Source for Request quotas (opens in a new tab)

Concurrency is separately capped; the public contract does not establish one universal concurrent-task number.

Data retention
API payloads and outputs: seven days in active systems, then deletion. No model training or public ASSETS publication. Necessary processors, legal retention, limited security metadata and protected backups are exceptions. Account and other personal data follow the separate privacy policy.Source for Data retention (opens in a new tab)

Policy effective August 3, 2026; not zero retention. URL expiry and active-system retention are distinct.

Custom model training
Custom LoRA and fine-tuning offered under negotiated Enterprise plans; applicable models, training limits and rates require a separate agreement.Source for Custom model training (opens in a new tab)
Deployment options
Native hosted API requires Business or an applicable Enterprise agreement. Enterprise advertises private on-premise deployment; exact model availability and hardware requirements are not specified.Source for Deployment options (opens in a new tab)
Task refunds
Not collected

No automatic failed-task credit refund guarantee verified in these operation contracts.

Output formats
GLB, USDZ, FBX, OBJ, STLSource for Output formats (opens in a new tab)

Default GLB; Agentic uses export_type, other operations geometry_file_format.

Usage metering
Credits per accepted API task. Business or Enterprise API plan required; subscription costs are separate.Source for Usage metering (opens in a new tab)
Output transport
Use jobs.subscription_key for status and top-level uuid for downloads.Source for Output transport (opens in a new tab)

Wait for every job to be Done. Download URLs can expire; no fixed lifetime is promised in the current download contract.

Maximum prompt characters
1,024Source for Maximum prompt characters (opens in a new tab)

Prompt characters, not LLM tokens. Required without images; optional with images.

Image input limits
1–5 uploaded images. The first image supplies material guidance. For text-to-3D, omit images.Source for Image input limits (opens in a new tab)

No numeric per-image byte limit verified in this generation-family contract; Agentic’s 25 MiB limit is not inherited.

Native model options
Sketch, Regular, Detail, SmoothSource for Native model options (opens in a new tab)

Always send the intended tier. Omission falls back to Regular (Gen-1/1.5).

PBR materials
SupportedSource for PBR materials (opens in a new tab)

PBR default; Shaded, All or None are also accepted. None generates geometry without materials.

Bounding-box placement
bbox_condition: exactly three integers, Width (X), Height (Y), Length (Z), each 1–2048.Source for Bounding-box placement (opens in a new tab)

Proportion control; these are not text context dimensions or verified physical units.

Character pose
TAPose defaults false; enable T/A rest-pose conditioning for human-like subjects.Source for Character pose (opens in a new tab)
Mesh topology and face count
Raw triangles or Quad, default Quad. Both presets: high 50,000; medium 18,000; low 8,000; extra-low 4,000. Custom target 500–200,000.Source for Mesh topology and face count (opens in a new tab)
Texture resolution
2K normally, 4K with HighPack. Texture enhancement and output dimensions are separate controls.Source for Texture resolution (opens in a new tab)
Reference images
condition_mode=concat for multiple views of one object; fuse combines features across images. Default concat.Source for Reference images (opens in a new tab)
Mesh processing controls
Raw mesh_simplify defaults true; Quad mesh_smooth defaults false. HighPack adds a roughly 16×-face mesh only for Sketch/Regular in Quad mode with an explicit quality_override ≤100,000. Detail/Smooth get texture changes only.Source for Mesh processing controls (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.

Hyper3D native API · text to 3d

Provider source (opens in a new tab)
API model ID
Sketch | Regular | Detail | Smooth

Choose one tier value. These are family/quality selectors, not separate immutable checkpoints.

Access
POST https://api.hyper3d.com/api/v2/rodin with multipart/form-data and Bearer API key.
Regions
Processing region not verified
Authentication
Bearer API key. Generation uses multipart/form-data; status/download use JSON.Source for Authentication (opens in a new tab)
Asynchronous generation
SupportedSource for Asynchronous generation (opens in a new tab)

Submit, poll every job to Done, then request download URLs. This is not progressive model-output streaming.

Request quotas
Business pricing lists 120 RPM on the first three subscription tiers and 240 RPM on higher tiers. Actual limits vary by account and endpoint; Retry-After on HTTP 429 takes precedence.Source for Request quotas (opens in a new tab)

Concurrency is separately capped; the public contract does not establish one universal concurrent-task number.

Data retention
API payloads and outputs: seven days in active systems, then deletion. No model training or public ASSETS publication. Necessary processors, legal retention, limited security metadata and protected backups are exceptions. Account and other personal data follow the separate privacy policy.Source for Data retention (opens in a new tab)

Policy effective August 3, 2026; not zero retention. URL expiry and active-system retention are distinct.

Custom model training
Custom LoRA and fine-tuning offered under negotiated Enterprise plans; applicable models, training limits and rates require a separate agreement.Source for Custom model training (opens in a new tab)
Deployment options
Native hosted API requires Business or an applicable Enterprise agreement. Enterprise advertises private on-premise deployment; exact model availability and hardware requirements are not specified.Source for Deployment options (opens in a new tab)
Output formats
GLB, USDZ, FBX, OBJ, STLSource for Output formats (opens in a new tab)

Default GLB; Agentic uses export_type, other operations geometry_file_format.

Output transport
Use jobs.subscription_key for status and top-level uuid for downloads.Source for Output transport (opens in a new tab)

Wait for every job to be Done. Download URLs can expire; no fixed lifetime is promised in the current download contract.

Maximum prompt characters
1,024Source for Maximum prompt characters (opens in a new tab)

Prompt characters, not LLM tokens. Required without images; optional with images.

Image input limits
1–5 uploaded images. The first image supplies material guidance. For text-to-3D, omit images.Source for Image input limits (opens in a new tab)

No numeric per-image byte limit verified in this generation-family contract; Agentic’s 25 MiB limit is not inherited.

Native model options
Sketch, Regular, Detail, SmoothSource for Native model options (opens in a new tab)

Always send the intended tier. Omission falls back to Regular (Gen-1/1.5).

PBR materials
SupportedSource for PBR materials (opens in a new tab)

PBR default; Shaded, All or None are also accepted. None generates geometry without materials.

Bounding-box placement
bbox_condition: exactly three integers, Width (X), Height (Y), Length (Z), each 1–2048.Source for Bounding-box placement (opens in a new tab)

Proportion control; these are not text context dimensions or verified physical units.

Character pose
TAPose defaults false; enable T/A rest-pose conditioning for human-like subjects.Source for Character pose (opens in a new tab)
Mesh topology and face count
Raw triangles or Quad, default Quad. Both presets: high 50,000; medium 18,000; low 8,000; extra-low 4,000. Custom target 500–200,000.Source for Mesh topology and face count (opens in a new tab)
Texture resolution
2K normally, 4K with HighPack. Texture enhancement and output dimensions are separate controls.Source for Texture resolution (opens in a new tab)
Reference images
condition_mode=concat for multiple views of one object; fuse combines features across images. Default concat.Source for Reference images (opens in a new tab)
Mesh processing controls
Raw mesh_simplify defaults true; Quad mesh_smooth defaults false. HighPack adds a roughly 16×-face mesh only for Sketch/Regular in Quad mode with an explicit quality_override ≤100,000. Detail/Smooth get texture changes only.Source for Mesh processing controls (opens in a new tab)
Required request fields
prompt required; omit images; set tier explicitly.Source for Required request fields (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
3D generationBase generation10.5 creditsrequest

Rate conditions

  1. Provider API credits; no universal USD conversion asserted. Current operation pricing lists 0.5 credits. Older HighPack add-on guidance conflicts; confirm consumed on the accepted task.
Pricing source (opens in a new tab)

Data & enterprise terms

API payloads and outputs: seven days in active systems, then deletion. No model training or public ASSETS publication. Necessary processors, legal retention, limited security metadata and protected backups are exceptions. Account and other personal data follow the separate privacy policy.

Read the service policy (opens in a new tab)

Hyper3D native API · image to 3d

Provider source (opens in a new tab)
API model ID
Sketch | Regular | Detail | Smooth

Choose one tier value. These are family/quality selectors, not separate immutable checkpoints.

Access
POST https://api.hyper3d.com/api/v2/rodin with multipart/form-data and Bearer API key.
Regions
Processing region not verified
Authentication
Bearer API key. Generation uses multipart/form-data; status/download use JSON.Source for Authentication (opens in a new tab)
Asynchronous generation
SupportedSource for Asynchronous generation (opens in a new tab)

Submit, poll every job to Done, then request download URLs. This is not progressive model-output streaming.

Request quotas
Business pricing lists 120 RPM on the first three subscription tiers and 240 RPM on higher tiers. Actual limits vary by account and endpoint; Retry-After on HTTP 429 takes precedence.Source for Request quotas (opens in a new tab)

Concurrency is separately capped; the public contract does not establish one universal concurrent-task number.

Data retention
API payloads and outputs: seven days in active systems, then deletion. No model training or public ASSETS publication. Necessary processors, legal retention, limited security metadata and protected backups are exceptions. Account and other personal data follow the separate privacy policy.Source for Data retention (opens in a new tab)

Policy effective August 3, 2026; not zero retention. URL expiry and active-system retention are distinct.

Custom model training
Custom LoRA and fine-tuning offered under negotiated Enterprise plans; applicable models, training limits and rates require a separate agreement.Source for Custom model training (opens in a new tab)
Deployment options
Native hosted API requires Business or an applicable Enterprise agreement. Enterprise advertises private on-premise deployment; exact model availability and hardware requirements are not specified.Source for Deployment options (opens in a new tab)
Output formats
GLB, USDZ, FBX, OBJ, STLSource for Output formats (opens in a new tab)

Default GLB; Agentic uses export_type, other operations geometry_file_format.

Output transport
Use jobs.subscription_key for status and top-level uuid for downloads.Source for Output transport (opens in a new tab)

Wait for every job to be Done. Download URLs can expire; no fixed lifetime is promised in the current download contract.

Maximum prompt characters
1,024Source for Maximum prompt characters (opens in a new tab)

Prompt characters, not LLM tokens. Required without images; optional with images.

Image input limits
1–5 uploaded images. The first image supplies material guidance. For text-to-3D, omit images.Source for Image input limits (opens in a new tab)

No numeric per-image byte limit verified in this generation-family contract; Agentic’s 25 MiB limit is not inherited.

Native model options
Sketch, Regular, Detail, SmoothSource for Native model options (opens in a new tab)

Always send the intended tier. Omission falls back to Regular (Gen-1/1.5).

PBR materials
SupportedSource for PBR materials (opens in a new tab)

PBR default; Shaded, All or None are also accepted. None generates geometry without materials.

Bounding-box placement
bbox_condition: exactly three integers, Width (X), Height (Y), Length (Z), each 1–2048.Source for Bounding-box placement (opens in a new tab)

Proportion control; these are not text context dimensions or verified physical units.

Character pose
TAPose defaults false; enable T/A rest-pose conditioning for human-like subjects.Source for Character pose (opens in a new tab)
Mesh topology and face count
Raw triangles or Quad, default Quad. Both presets: high 50,000; medium 18,000; low 8,000; extra-low 4,000. Custom target 500–200,000.Source for Mesh topology and face count (opens in a new tab)
Texture resolution
2K normally, 4K with HighPack. Texture enhancement and output dimensions are separate controls.Source for Texture resolution (opens in a new tab)
Reference images
condition_mode=concat for multiple views of one object; fuse combines features across images. Default concat.Source for Reference images (opens in a new tab)
Mesh processing controls
Raw mesh_simplify defaults true; Quad mesh_smooth defaults false. HighPack adds a roughly 16×-face mesh only for Sketch/Regular in Quad mode with an explicit quality_override ≤100,000. Detail/Smooth get texture changes only.Source for Mesh processing controls (opens in a new tab)
Required request fields
1–5 images required; optional prompt; set tier explicitly.Source for Required request fields (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
3D generationBase generation10.5 creditsrequest

Rate conditions

  1. Provider API credits; no universal USD conversion asserted. Current operation pricing lists 0.5 credits. Older HighPack add-on guidance conflicts; confirm consumed on the accepted task.
Pricing source (opens in a new tab)

Data & enterprise terms

API payloads and outputs: seven days in active systems, then deletion. No model training or public ASSETS publication. Necessary processors, legal retention, limited security metadata and protected backups are exceptions. Account and other personal data follow the separate privacy policy.

Read the service policy (opens in a new tab)

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.

Hyper3D native API · text to 3d

Sketch | Regular | Detail | Smooth
Not verified

No prompt-cache contract or cached-input discount verified for this API.

How it works
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Minimum reusable input
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Lifetime & refresh
Not collected

No prompt-cache contract or cached-input discount verified for this API.

How to enable it
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Usage fields
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Billing rules
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Limits & data boundaries
Not collected

No prompt-cache contract or cached-input discount verified for this API.

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

Caching reviewed 14 Oct 2026

Hyper3D native API · image to 3d

Sketch | Regular | Detail | Smooth
Not verified

No prompt-cache contract or cached-input discount verified for this API.

How it works
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Minimum reusable input
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Lifetime & refresh
Not collected

No prompt-cache contract or cached-input discount verified for this API.

How to enable it
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Usage fields
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Billing rules
Not collected

No prompt-cache contract or cached-input discount verified for this API.

Limits & data boundaries
Not collected

No prompt-cache contract or cached-input discount verified for this API.

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

Caching reviewed 14 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

Qualified inference from named provider-operated Rodin families, restricted service terms and the reviewed official repository inventory, which identifies integration tools rather than corresponding released checkpoint weights. Exact internal weights and per-checkpoint licenses remain undisclosed; repository inventory is not exhaustive proof of absence.

Classification evidence (opens in a new tab)

Lifecycle dates

Gen-1.5 member officially launched29 Apr 2025Source (opens in a new tab)

Sources reviewed