Submit one to five images. The first image is used for material generation. prompt is optional; when omitted, the service derives guidance from the images.
Always send a Gen-2.5 tier explicitly. If tier is omitted, the endpoint falls back to Regular and generates a Gen-1/1.5 asset.
tier selects the generation family and how much detail the model resolves. It is independent of the mesh and texture controls below: any tier can be combined with any mesh_mode, quality, or texture_mode the schema allows.
Tier
Intended use
Gen-2.5-Extreme-Low
Quick generation of simple assets.
Gen-2.5-Low
Clean assets and small hard-surface props.
Gen-2.5-Medium
Balanced structure and detail.
Gen-2.5-High
Richer structure and smoother surfaces.
Gen-2.5-Extreme-High
High-frequency detail reproduction. This tier also returns an additional mesh with up to 10M of faces.
Creative mode (geometry_instruct_mode=creative) allows more flexible generation while maintaining consistent output quality. It is available for Gen-2.5-Medium and higher tiers.
HighPack upgrades pack textures from 2K to 4K. For Gen-2.5 it does not change geometry or generate an additional high-poly model.
Use jobs.subscription_key for Check Status and top-level uuid for Download Results. HTTP 201 means the endpoint handled the request; acceptance requires an absent or empty error and a non-empty top-level uuid. HTTP 400 indicates invalid fields/files, 401 indicates invalid authentication, and 429 indicates throttling.
Generate a 3D model and textures from images or a text prompt.
images*array<>
Images to be used in generation, up to 5 images. As the form data request will preserve the order of the images, For Image-to-3D generation: required (one or more images are needed, maximum 5 images)
For Text-to-3D generation: omit this field.
prompt?string
A textual prompt to guide the model generation.
For Image-to-3D generation: optional, provide orientation-level guidance (symmetric, sharp, detail...) for generation process
For Text-to-3D generation: required. The prompt is first rendered to an image, so a text-only request can also be rejected with `IMAGE_CONTENT_VIOLATION`.
Tips: it is recommended to omit this field for Image-to-3D generation.
soft?boolean
Optional. Appends smooth edges to the final geometry-generation prompt without changing the generation tier.
Tips: We recommend leaving this parameter unset or omitting it entirely.
detail_level?integer
Optional. Controls how many final geometry-refinement stages use RGB image conditioning. Higher values preserve more visual detail. Omit to use the default refinement path.
Tips: We recommend leaving this parameter unset or omitting it entirely.
tier*string
Required. Select and explicitly send a Gen-2.5 tier. If omitted, the endpoint falls back to `Regular` and generates a Gen-1/1.5 asset.
See [Tier](/en/api-specification/rodin-gen2-5#tier) for more information.
This parameter may affect billing. See [Pricing](/en/api-specification/rodin-gen2-5#pricing).
mesh_mode?string
Optional. Mesh topology of the generated model. `Raw` stands for Triangle face, `Quad` stands for Quad face.
When omitted the default depends on `tier`: Gen-2.5 tiers default to `Raw`, every other tier defaults to `Quad`.
quad_normal?boolean
Optional. Bakes high-resolution geometry detail into a normal map during mesh refinement.
Optional. Texture mode for the generated model.
Higher modes generally produce richer, higher-resolution texture detail but take longer and may affect billing. See [Pricing](/en/api-specification/rodin-gen2-5#pricing).
- `legacy`: Use the legacy texture pipeline.
- `extreme-high`: Can produce up to 12K textures. See [Texture resolution](/en/api-specification/rodin-gen2-5#texture-resolution).
The default value depends on the selected `tier`, see [Texture mode](/en/api-specification/rodin-gen2-5#texture-mode).
material?string
Optional. Material type.
- `PBR`: Physically Based Materials, including base color texture, metallicness texture, normal texture and roughness texture, providing high realism and physically accurate response to dynamic lighting.
- `Shaded`: Only base color texture with baked lighting, providing stylized visuals.
- `Hybrid`: PBR texture maps with a base color texture that includes baked lighting.
- `All`: Both PBR and Shaded will be delivered.
- `None`: Asset without material. This will speed up generation.
quality?string
Optional. Preset target face count for Gen-2.5. Defaults to `medium`. The mapping depends on `mesh_mode`:
- `Quad`: `high` 50000, `medium` 18000, `low` 8000, `extra-low` 4000;
- `Raw`: `high` 1000000, `medium` 500000, `low` 60000, `extra-low` 20000.
quality_override?integer
Optional. Custom target face count from 500 to 2000000 for Gen-2.5. Takes precedence over `quality` when present. `Gen-2.5-High` and `Gen-2.5-Extreme-High` support up to 2000000; all other Gen-2.5 tiers support up to 1000000. `Quad` mode has an additional maximum of 200000.
seed?integer
Optional. A seed value for randomization in the mesh and texture generation.
bbox_condition?array<>
Optional. Controls the maximum dimensions of the generated model. Provide exactly three integers in Width (X-axis), Height (Y-axis), Length (Z-axis) order `(x, y, z)`. See [BBOX Control](/en/api-specification/bbox-control).
addons?array<>
Optional add-on. Upgrades pack textures from 2K to 4K. For Gen-2.5, HighPack does not change geometry or generate an additional high-poly model.
TAPose?boolean
Optional. For a human-like subject, enables rest-pose conditioning so the result is generated in a T-pose or A-pose suitable for animation. When `false`, the input pose is reconstructed as-is.
use_original_alpha?boolean
Optional. If `true`, preserves and uses the uploaded image alpha channel during image preprocessing. Has no effect when the input image has no alpha channel.
preview_render?boolean
Optional. If true, provides an additional high-quality render image in the download list. Defaults to `false`.
If use the [Fast mode](/en/api-specification/rodin-gen2-5#fast-mode), this parameter is ignored.
is_symmetric?string
Optional. Intended symmetry of the generated geometry. Use `symmetric` for objects that should mirror left and right, `asymmetric` for posed or irregular objects, and `balanced` for mostly symmetric objects that should retain small natural differences. Use `unknown` to let the system infer the mode from the prompt and input images.
image_label?array<>
The viewing direction of each uploaded image, positionally matched to images. Entries default to ? (unknown) and the list is padded to the image count. Supplying more labels than images is rejected with IMAGE_LABEL_LENGTH_TOO_LONG. Support direction labels:
- `Front`: `F`
- `Front-Left`: `FL`
- `Front-Right`: `FR`
- `Back`: `B`
- `Back-Left`: `BL`
- `Back-Right`: `BR`
- `Left`: `L`
- `Right`: `R`
- `Up`: `U`
- `Down`: `D`
- `Unknown`: `?`
hd_texture?boolean
Optional. Enables an additional texture super-resolution and enhancement pass. This can improve fine-detail clarity but may also introduce artifacts. It does not select the final 2K or 4K pack resolution; use the `HighPack` add-on for 4K output.
uhd_texture?boolean
is_micro?boolean
Whether to generate micro-scale surface detail. Only allowed when `tier` is `Gen-2.5-Extreme-High`; any other tier is rejected with `INVALID_REQUEST`.
Tips: We recommend leaving this parameter unset or set to default value.
geometry_instruct_mode?string
Optional. Selects the geometry-generation strategy. `creative` favors richer or more imaginative structure with less strict input fidelity; `faithful` stays closer to the provided image and prompt. Availability depends on the selected `tier`.
Tips: We recommend leaving this parameter unset or set to default value.
texture_delight?boolean
Optional. If `true`, removes highlights and baked lighting from texture-reference images before texture generation to reduce lighting artifacts in the generated material.
Tips: We recommend leaving this parameter unset or omitting it entirely unless you have specific lighting requirements.
Images to be used in generation, up to 5 images. As the form data request will preserve the order of the images, For Image-to-3D generation: required (one or more images are needed, maximum 5 images)
For Text-to-3D generation: omit this field.
A textual prompt to guide the model generation.
For Image-to-3D generation: optional, provide orientation-level guidance (symmetric, sharp, detail...) for generation process
For Text-to-3D generation: required. The prompt is first rendered to an image, so a text-only request can also be rejected withIMAGE_CONTENT_VIOLATION.
Tips: it is recommended to omit this field for Image-to-3D generation.
Optional. Appends smooth edges to the final geometry-generation prompt without changing the generation tier.
Tips: We recommend leaving this parameter unset or omitting it entirely.
Optional. Controls how many final geometry-refinement stages use RGB image conditioning. Higher values preserve more visual detail. Omit to use the default refinement path.
Tips: We recommend leaving this parameter unset or omitting it entirely.
Required. Select and explicitly send a Gen-2.5 tier. If omitted, the endpoint falls back toRegular and generates a Gen-1/1.5 asset.
The default value depends on the selectedtier, seeTexture mode.
Optional. Material type.
PBR: Physically Based Materials, including base color texture, metallicness texture, normal texture and roughness texture, providing high realism and physically accurate response to dynamic lighting.
Shaded: Only base color texture with baked lighting, providing stylized visuals.
Hybrid: PBR texture maps with a base color texture that includes baked lighting.
All: Both PBR and Shaded will be delivered.
None: Asset without material. This will speed up generation.
Optional. Preset target face count for Gen-2.5. Defaults tomedium. The mapping depends onmesh_mode:
Optional. Custom target face count from 500 to 2000000 for Gen-2.5. Takes precedence overquality when present.Gen-2.5-High andGen-2.5-Extreme-High support up to 2000000; all other Gen-2.5 tiers support up to 1000000.Quad mode has an additional maximum of 200000.
Optional. A seed value for randomization in the mesh and texture generation.
Optional. Controls the maximum dimensions of the generated model. Provide exactly three integers in Width (X-axis), Height (Y-axis), Length (Z-axis) order(x, y, z). SeeBBOX Control.
Optional add-on. Upgrades pack textures from 2K to 4K. For Gen-2.5, HighPack does not change geometry or generate an additional high-poly model.
Optional. For a human-like subject, enables rest-pose conditioning so the result is generated in a T-pose or A-pose suitable for animation. Whenfalse, the input pose is reconstructed as-is.
Optional. Iftrue, preserves and uses the uploaded image alpha channel during image preprocessing. Has no effect when the input image has no alpha channel.
Optional. If true, provides an additional high-quality render image in the download list. Defaults tofalse.
Optional. Intended symmetry of the generated geometry. Usesymmetric for objects that should mirror left and right,asymmetric for posed or irregular objects, andbalanced for mostly symmetric objects that should retain small natural differences. Useunknown to let the system infer the mode from the prompt and input images.
The viewing direction of each uploaded image, positionally matched to images. Entries default to ? (unknown) and the list is padded to the image count. Supplying more labels than images is rejected with IMAGE_LABEL_LENGTH_TOO_LONG. Support direction labels:
Front:F
Front-Left:FL
Front-Right:FR
Back:B
Back-Left:BL
Back-Right:BR
Left:L
Right:R
Up:U
Down:D
Unknown:?
Optional. Enables an additional texture super-resolution and enhancement pass. This can improve fine-detail clarity but may also introduce artifacts. It does not select the final 2K or 4K pack resolution; use theHighPack add-on for 4K output.
Whether to generate micro-scale surface detail. Only allowed whentier isGen-2.5-Extreme-High; any other tier is rejected withINVALID_REQUEST.
Tips: We recommend leaving this parameter unset or set to default value.
Optional. Selects the geometry-generation strategy.creative favors richer or more imaginative structure with less strict input fidelity;faithful stays closer to the provided image and prompt. Availability depends on the selectedtier.
Tips: We recommend leaving this parameter unset or set to default value.
Optional. Iftrue, removes highlights and baked lighting from texture-reference images before texture generation to reduce lighting artifacts in the generated material.
Tips: We recommend leaving this parameter unset or omitting it entirely unless you have specific lighting requirements.
