Up to 2 Million Faces
Standard geometry supports up to 1.5 million faces. Detailed geometry raises the ceiling to 2 million, giving close-up objects, sculptural forms, and print-oriented parts more room for fine surface structure.
curl --request POST "https://gptproto.com/api/v3/tripo3d/tripo3d-h3.1/image-to-3d" \
--header "Authorization: Bearer $GPTPROTO_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"image": "",
"texture": true,
"pbr": true,
"texture_quality": "standard",
"texture_version": "v3.0-20250812",
"delight": true,
"geometry_quality": "standard",
"quad": false,
"smart_low_poly": false,
"generate_parts": false,
"face_limit": null,
"texture_alignment": "original_image",
"enable_image_autofix": false,
"auto_size": false,
"export_uv": true,
"model_seed": null,
"texture_seed": null,
"enable_base64_output": false,
"enable_sync_mode": false
}'Start from the cost of a single sample and pick a testing budget.
| Scenario | Tripo3d list | OpenRouter | GPTProto | You save / mo |
|---|---|---|---|---|
| Quick tests & prototypes100 images / mo | $36.00 | $37.98 | $36.00 | −$0.00≈ $0.00 / yr |
| Product & e-commerce photos1,000 images / mo | $360.00 | $379.80 | $360.00 | −$0.00≈ $0.00 / yr |
| Ad / marketing creatives2,500 images / mo | $900.00 | $949.50 | $900.00 | −$0.00≈ $0.00 / yr |
| Social media content10,000 images / mo | $3600.00 | $3798.00 | $3600.00 | −$0.00≈ $0.00 / yr |
Use the Tripo3D H3.1 API to create close-up 3D assets with dense geometry, textured surfaces, and controls for different downstream pipelines. Build from a text prompt, a reference image, or multiple views, then choose detailed geometry, PBR materials, quad output, or a lighter topology according to the asset's final use.
Up to 2 Million Faces
Standard geometry supports up to 1.5 million faces. Detailed geometry raises the ceiling to 2 million, giving close-up objects, sculptural forms, and print-oriented parts more room for fine surface structure.
Tripo H3.1 is the current high-precision model in Tripo's H Series. Its official model version is v3.1-20260211. It accepts text, single-image, and multiview inputs and returns a mesh with optional textures and PBR materials. The model is aimed at work where geometry fidelity matters more than minimum generation time, including hero assets, close-up product visuals, 3D printing, and detailed scene elements.
H3.1 is more than a higher polygon cap. Compared with the legacy v2.5 model, it adds the newer geometry and texture controls introduced in the H3 generation: Detailed geometry, quad output, smart low-poly topology, automatic real-world sizing, segmented parts, and geometry compression. Developers can therefore start from a dense asset and choose how it should enter a DCC, rendering, print, or real-time pipeline.
| Specification | Verified value |
|---|---|
| Official model version | v3.1-20260211 |
| Input routes | Text, single image, multiview images |
| Output | 3D mesh with optional textures and PBR materials |
| Standard geometry | Up to 1,500,000 faces |
| Detailed geometry | Up to 2,000,000 faces |
| Quad output | FBX; Tripo recommends no more than 150,000 faces |
| Smart low-poly range | 500-20,000 faces; 500-10,000 when combined with quad output |
| Texture tiers | Standard, detailed, and extreme 8K; a fast tier requires the newer texture model |
| Reproducibility | Separate geometry and texture seeds are documented by Tripo |
| Published typical time | About 40 seconds without texture and about 120 seconds with texture; actual latency varies |
The main Tripo3D H3.1 API upgrades are visible in geometry density and pipeline control. Detailed mode lifts the geometry ceiling to 2 million faces, while v2.5 is capped at 500,000. H3.1 also supports advanced controls that the official API does not accept for v2.5, including geometry_quality, quad, smart_low_poly, auto_size, generate_parts, and compress.
These controls solve different problems. Use Detailed geometry when the asset will be viewed close up or printed at a larger size. Use quad FBX output when a downstream artist needs a deformation-friendly starting point. Use smart low-poly for simpler assets that must fit a tighter real-time budget. Generate parts only when an editable, segmented object matters more than receiving textures in the same request.
H3.1 is the better default for new high-detail work. The older model remains useful for reproducing an existing workflow while it is being migrated, but Tripo's official deprecation notice sets December 1, 2026 as the end date for v2.5 model and texture support.
| Comparison | Tripo H3.1 | Tripo3D v2.5 |
|---|---|---|
| Official version | v3.1-20260211 |
v2.5-20250123 |
| Positioning | Current high-precision H-Series model | Legacy model for historical compatibility |
| Maximum triangle faces | 1.5M standard; 2M detailed | Up to 500K |
| Detailed or Ultra geometry | Supported | Not supported |
| Quad and smart low-poly generation controls | Supported | Not accepted by the official generation API |
| Auto sizing, segmented parts, and compression | Supported | Not accepted by the official generation API |
| Best fit | Close-up assets, printing, detailed rendering, and higher-control pipelines | Temporary compatibility with existing v2.5 projects |
| Official status | Current stable model | Official support ends December 1, 2026 |
If you still maintain an H2.5 integration, review the existing Tripo3D v2.5 API page, then test H3.1 with the same source inputs, face limits, and texture settings before changing a live asset pipeline. Do not compare only the preview render: inspect mesh density, topology, UVs, texture maps, scale, and export format.
Hero assets and close-up renders: Use Detailed geometry when silhouettes, carved edges, seams, or surface relief must hold up under a close camera.
3D printing and manufacturing previews: The higher face ceiling can preserve curved surfaces and small structural details before print cleanup. Use real-world sizing when the exposed request schema supports it.
Product and e-commerce assets: Start from one clean image or several consistent views, then generate PBR materials for relighting and interactive viewers.
Game and VFX source assets: Generate a dense source mesh, a quad FBX starting point, or a smart low-poly variant according to the next tool in the pipeline.
Editable multi-part objects: Segmented generation can help when doors, lids, wheels, or other components need separate editing. It cannot be combined with textured/PBR or quad generation in the same native request.
H3.1 is not automatically the right setting for every job. If the asset will be small on screen, generated in large batches, or used only for early blocking, begin with standard geometry and a practical face limit. Increase detail only after the silhouette and proportions are correct.
A useful 3D prompt describes the object as something that must be modeled from every angle, not as a single cinematic image. Put the shape and construction first, then add materials, surface condition, style, and intended use.
Prompt structure
[object] + [silhouette and major parts] + [materials] + [surface details] + [style] + [intended use or scale]
Example prompt
A compact retro-futurist field radio with a rectangular aluminum body, rounded protective corners, one large tuning dial, two smaller switches, a folding antenna, and a woven speaker grille. Brushed metal with worn paint along exposed edges, realistic proportions, freestanding product asset, no background or display stand.
Negative prompt example
fused controls, floating parts, warped symmetry, paper-thin walls, unreadable surface detail, extra antennas, display stand
For repeatable tests, keep the geometry seed fixed while changing one part of the prompt at a time. If the shape is correct but the surface is not, keep the geometry seed and change the texture seed. When accurate unseen sides matter, use multiview input instead of adding camera language to a text prompt. When a clean reference image already exists, use the image-to-3D route rather than trying to describe the image again in text.
GPTProto is an API provider for developers who want Tripo H3.1 access without maintaining a separate key and balance for every model vendor. One GPTProto key and one pay-as-you-go balance can be used across 200+ text, image, video, audio, and 3D models. The reason to choose this route is account and billing consolidation, not an unverified discount claim.
Use the live API Usage panel on this page as the source of truth for the current GPTProto model string, endpoint, request fields, and polling flow. Use the Pricing panel for the current per-generation rate. To compare other available versions, browse the Tripo3D model hub.
Guides, comparisons, and updates related to this model.
All Articles
ByteDance's Seedance 2.5 Flash runs 2x faster at ~¥0.99/s (~$0.14/s) for 30s 720p clips—no 4K. See real trade-offs, Kling 4.0 Flash vs 2.5 specs & API code.

See a cat sketch turn into a looping GIF with ChatGPT Images 2.6. Compare Flare and Sunburst and use either on GPTProto at 5% below OpenAI API rates.

Learn how to face swap yourself into a video using a photo or reference clip, fix flicker and side-angle errors, and choose the right workflow.

See Nano Banana 2.1 and Nano Banana Pro compared in five real ecommerce tests covering product photos, editing, packaging, multilingual text, and pricing.
Input
Output