
Meshy v7 Image-to 3D API by MESHY
Meshy v7 Image-to-3D reconstructs a textured 3D model from a single image in one request. Outputs GLB/OBJ/FBX/STL/USDZ/3MF with optional PBR maps up to 8K, a settable polygon budget, and optional four-angle preview renders.
Meshy v7 Image-to 3D is developed by a third-party provider. Atlas Cloud (operated by Atlas Cloud AI LLC) provides access to it and does not own it. All trademarks belong to their respective owners.
Meshy 7 Image-to-3D
Meshy 7 Image-to-3D reconstructs a textured 3D model from a single photograph or render. One request in, a mesh with PBR textures out — no preview stage, no second call.
Why Choose This?
- One call, one model — geometry and textures are generated in the same task.
- Faithful to your image — turn
image_enhancementoff to keep the input's exact look instead of a stylised interpretation. - Polygon budgets you control — an exact face count, an adaptive decimation level, or the raw high-detail mesh.
- PBR textures up to 8K — base colour plus metallic, roughness and normal maps.
- Review before you download — optional four-angle preview renders and a transparent-background thumbnail.
Parameters
| Parameter | Required | Description |
|---|---|---|
model | Yes | meshy-v7/image-to-3d |
image | Yes | Source image: public URL, base64 data URI, or asset://<ASSET_ID>. jpg/jpeg/png. |
ultra_mode | No | Finer surface detail. Slower, +5 credits. Default: false |
should_texture | No | Generate textures. Default: true; false returns a bare mesh for less. |
enable_pbr | No | Also produce metallic / roughness / normal maps. Default: false |
texture_resolution | No | 2k (default), 4k, or 8k. |
texture_prompt | No | Describe the surface material, up to 800 characters. |
texture_image | No | A separate reference image for the texture (URL, data URI, or asset://<ID>). |
image_enhancement | No | Pre-process the input for a better reconstruction. Default: true |
should_remesh | No | Run the remesh phase. Default: false — leave it off for the best mesh. |
topology | No | triangle (default) or quad. Requires should_remesh. |
target_polycount | No | Face count, 100–300000. Default: 30000. Requires should_remesh. |
decimation_mode | No | Adaptive polycount level 1–4 (ultra → low). Requires should_remesh; overrides target_polycount. |
save_pre_remeshed_model | No | Also return the GLB from before the remesh. Requires should_remesh. Default: false |
pose_mode | No | a-pose, t-pose, or empty for no forced pose. |
target_formats | No | Any of glb, obj, fbx, stl, usdz, 3mf. Default: ["glb"] |
auto_size | No | Scale the model to an AI-estimated real-world height. Default: false |
origin_at | No | bottom (default) or center, when auto_size is on. |
alpha_thumbnail | No | Also render a transparent-background preview. Default: false |
multi_view_thumbnails | No | Also render right / back / left previews (+~3s). Default: false |
moderation | No | Screen the inputs for harmful content before generating. Default: false |
target_formats defaults to GLB alone rather than every format: each extra format is another file to generate and store, so ask for OBJ/FBX/STL/USDZ/3MF only when your pipeline needs them.
topology, target_polycount and decimation_mode only take effect when should_remesh is on — they configure the remesh phase, so with remeshing off they are ignored upstream.
How to Use
curl -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \ -H "Authorization: Bearer $AIP_API_KEY" \ -H "Content-Type: application/json" \ --data-raw '{ "model": "meshy-v7/image-to-3d", "image": "https://static.atlascloud.ai/media/images/sneaker.png", "should_texture": true, "enable_pbr": true, "texture_resolution": "4k", "target_formats": ["glb"] }'
Generation is asynchronous — poll /api/v1/model/prediction/{request_id} until status is completed, then read the model URLs from outputs. files gives the same URLs with their format and MIME type, and thumbnail is a rendered preview image.
Pricing
Billed per request, in Meshy credits:
| Configuration | Credits |
|---|---|
Mesh only (should_texture: false) | 20 |
Mesh + texture at 2k / 4k | 30 |
Mesh + texture at 8k | 35 |
ultra_mode | +5 on top of the above |
Failed tasks are not charged.
Best Use Cases
- Product catalogues — turn existing product photography into web and AR viewers at scale.
- Game and film assets — props and set dressing from concept art, with a polygon budget you set.
- 3D printing —
target_formats: ["stl", "3mf"]plusauto_sizefor a physically-scaled print. - AR try-on and Quick Look —
usdzoutput straight from a single product shot. - Digitising real objects — photograph it once, get a mesh you can edit.
Pro Tips
- One object, centred, evenly lit, plain background. Clutter and hard shadows become geometry.
- A three-quarter view reconstructs better than a dead-on front or a steep top-down shot.
- Turn
image_enhancementoff when the input is already a clean render and you need it reproduced exactly. - Leave
should_remeshoff unless you need a specific face count — remeshing trades detail for polygon control, and with it offtopology/target_polycount/decimation_modedo nothing. multi_view_thumbnailsis the cheap way to check a batch: four renders instead of downloading every GLB.- Use
texture_imagewhen the shape comes from one photo but the material should come from another. - Need a more faithful reconstruction of the back and sides? Shoot several angles and use Multi-Image-to-3D.
Notes
- Supported input formats are jpg, jpeg and png; other formats are rejected upstream.
texture_resolutionmust be spelled exactly as documented (lowercase2k/4k/8k): a request is rejected rather than silently priced at a different rate.enable_pbrrequiresshould_texture: true.save_pre_remeshed_modelonly produces a file whenshould_remeshis also on.symmetry_mode,art_styleandis_a_t_poseare deprecated upstream and are not exposed here; usepose_modefor poses.- Generated files are re-hosted on AtlasCloud storage, so the URLs you receive are ours and stay reachable — Meshy itself deletes API assets after a few days.
Related Models
- Meshy 7 Multi-Image-to-3D — up to four views of the same object for a more faithful reconstruction.
- Meshy 7 Text-to-3D — generate a textured model from a written description, no image needed.


















