Guide
Tripo H3.1 Getting Started Guide: Choosing Among Three Input Modes, Parameters, and Pricing
A practical guide for developers on 3D generation integration decisions and cost control
On this page
What Tripo H3.1 Can Do and Where Its Boundaries Lie
Tripo H3.1 is a 3D generation model available through HiAPI.ai's unified interface; a single API key is all you need to call it. It supports three input forms: natural-language descriptions (text-to-3d), a single object image (image-to-3d), and 2–4 multiview images of the same object (multiview-to-3d).
The output is a fixed deliverable bundle: one GLB model file + one preview image. There is no extra texture-pack bundle, no project file, and no rigging or animation. GLB is a universal format that can go straight into Blender, Three.js, Unity, or any pipeline that supports glTF.
The boundaries deserve equal clarity:
- It generates at the object level, not the scene level. Give it a picture of a chair and you get a chair, not a room with a chair in it. If you need a scene, combine multiple models yourself.
- It does not understand "motion." Static assets are its comfort zone; rigging, skinning, and animation sequences are out of scope.
- Source quality sets the output ceiling. Blurry inputs, multiple objects, and complex occlusion significantly degrade geometry quality, and this holds true on all three paths.
- Billing is per call: one call produces one model, with no intermediate token-based or per-second state. This is good for cost estimation—the number of calls is the billing unit.
Once you understand this scope, the next choice becomes meaningful: the three paths are not three entrances to the same thing; their unit prices, source-material requirements, and use cases all differ.

How to Choose Among the Three Input Paths
The first principle for choosing a path is not "which one produces better results," but what source material you have on hand.
Only a text description → text-to-3D. Suitable for concept exploration, placeholder assets, and early prototypes. You have full control over the generation direction, but you also bear all the uncertainty: the model's inference of face count and form comes entirely from your descriptor words. It has the lowest starting price tier and is suitable for large-volume trial and error.
A clean product image or object photo → image-to-3D. This is the path most developers will actually take. A single image facing the object with a clean background carries far more information than a paragraph of text, and geometry fidelity is noticeably more stable. The source-material bar is also low: just take a photo and cut out the background.
2–4 images from different angles → multiview-to-3D. When you already have multi-angle material (such as the standard four-view set from a product photo studio, or turntable frames rendered from an existing 3D asset), multiview offers the highest accuracy ceiling. It solves the structural problem in single-image generation where "the back is guessed."
A notable pricing fact: image-to-3D and multiview are the same price, while text-to-3D is cheaper. So when material is complete, multiview carries no extra premium—if you have the material, use multiview; there is no reason to fall back to single-image.
The decision can be reduced to one sentence: if you have enough material for multiview, use multiview; if you have one good image, use image-to-3D; only when you have nothing else should you use text-to-3D.
Input Parameters and Source-Material Preparation Tips
Each of the three paths has different input fields, and mixing them up will fail outright.
text-to-3D (model page) accepts a natural-language description. The key to writing a good prompt is specificity: object type, material tendency, and form characteristics are more effective than piling on adjectives.
image-to-3D (model page) uses image_url, a single HTTPS image. Source-material requirements:
- A single object; do not put multiple objects in one frame
- A clean background; avoid cluttered environments that share the subject's color
- Let the object's main body fill the frame as much as possible
- The URL must be publicly accessible (HTTPS); do not pass private addresses that require authentication
Multiview-to-3D (model page) uses image_urls, 2–4 images of the same object, with the order fixed as front, left, back, right. Getting the order wrong is one of the most common failure sources: the model assumes the viewpoint by position, so putting an image in the wrong slot amounts to giving it the wrong geometric constraint.
There are two general recommendations for source-material preparation. First, try to keep consistent lighting and scale across the multiview images; mixing photos from different sources creates ambiguity at the seams for the model. Second, image resolution need not be extremely large; clarity and a well-defined subject matter more than pixel count.
Requests with invalid parameters (missing fields, malformed URLs, out-of-range view counts) are rejected outright and are not charged—this point will come up again in the troubleshooting checklist in the next section.
face_limit and Geometry Quality: Controlling Model Complexity
face_limit is the core parameter controlling the output face count, with a range of 1000–1,500,000; when geometry_quality is set to detailed, the upper limit is relaxed to 2,000,000.
The key design point: this parameter can be omitted. When omitted, the model adaptively decides the face count, usually giving a reasonable value matched to the object's complexity. You only need to specify it manually when you clearly know the downstream face-count budget:
- Mobile or Web real-time rendering: on the low side, kept within tens of thousands of faces
- Mid-ground assets in games: a medium tier
- Film-grade or offline rendering: close to the upper limit, paired with detailed geometry
geometry_quality has two values: standard (standard geometry, the default) and detailed (detailed geometry). detailed delivers higher geometric detail and a higher face-count ceiling, but it is an option that incurs an additional charge, on all three paths. The judgment criterion is straightforward: turn it on only if the asset will be viewed up close or if detail itself is the selling point; enabling detailed for far-background or placeholder assets is pure waste.
A practical tip for using both together: first run one version with the default configuration (omit face_limit, use standard for geometry_quality) to see the result, then tune parameters for the specific asset. Do not max out parameters from the start—that both costs more money and may not yield better results.
Breaking Down the Pricing Structure: Which Tier the Money Goes To
Tripo H3.1 is billed per call, with each call producing a model + preview image. Price is determined jointly by four variables: input path, Texture toggle, PBR toggle, and geometry quality. Once you understand this structure, consult the full list prices below to choose a tier.
The current test-environment list prices are as follows, billed per call, per model, in USD.
text-to-3D (tripo-h3.1/text-to-3d)
| Specification | Standard texture | Fine texture |
|---|---|---|
| PBR off · Texture off | $0.150 | $0.150 |
| PBR off · Texture on | $0.290 | $0.430 |
| PBR on | $0.290 | $0.430 |
image-to-3D (tripo-h3.1/image-to-3d) and multiview-to-3D (tripo-h3.1/multiview-to-3d) are the same price
| Specification | Standard texture | Fine texture |
|---|---|---|
| PBR off · Texture off | $0.290 | $0.290 |
| PBR off · Texture on | $0.430 | $0.580 |
| PBR on | $0.430 | $0.580 |
All tiers above incur an additional $0.290 when detailed geometry is selected.
Layer one: input path. text-to-3D has the lowest starting tier; image-to-3D and multiview are the same price and higher than text-to-3D. This is intuitive: image-to-3D and multiview feed in more information, and the model does more work.
Layer two: the Texture and PBR toggles, the main price amplifiers. With both off, you are at the base tier; turning Texture on moves the unit price up a step; turning PBR on lands on the same or a higher tier. In other words, going from "pure geometry" to a complete "textured, PBR-enabled" asset can more than double the unit price. This is the largest price variable in the whole table, far more significant than the difference in input path.
Layer three: texture fineness. With textures enabled, standard and fine textures are two separate prices. The fine-texture premium is especially pronounced on the text-to-3D path.
Layer four: geometry quality. Choosing detailed fine geometry adds an extra charge on top of the above prices, and the extra charge is the same amount on all three paths.
This yields a few pricing judgments:
- If downstream only needs the geometry (for example, to make a collision body or a base mesh for subsequent hand-painted textures), turn off Texture and PBR and the unit price immediately falls back to the lowest tier.
- If you are only doing prototype validation, text-to-3D + all toggles off is the most economical combination.
- What is truly expensive is the complete delivery with all three—"fine texture + PBR + fine geometry"—fully enabled. That should be a decision made only after you have confirmed the asset will go live.

Recommended Configurations and Cost Estimates for Typical Scenarios
Let us ground the structure in concrete combinations. The schemes below are all per single call; actual cost accumulates with the number of calls.
Prototyping and concept exploration
text-to-3D, Texture off, PBR off, geometry_quality set to standard, face_limit omitted. This is the lowest tier across the whole model and is suitable for generating a dozen directions in one go for quick screening. After screening, re-run the shortlisted concepts once with textures.
E-commerce / product display assets
If you have a product image, go with image-to-3D; if you have standard four-view material, go with multiview. Texture on, PBR on (PBR has real value for conveying material feel, and display scenarios are worth the spend), standard texture, and standard geometry. If the product will be viewed in close-up, then consider upgrading to fine texture.
Game / real-time rendering assets
image-to-3D or multiview, Texture on, and PBR depending on the rendering pipeline (turn it on if the pipeline supports PBR). Set face_limit manually according to the target platform; do not leave it to adaptive. Geometry quality defaults to standard—under real-time rendering, the benefit of detailed usually does not show.
Geometry base mesh only
Any path, with Texture and PBR both off, and geometry quality chosen as needed. This is the way to push cost down to the structural minimum, leaving subsequent texturing to the art pipeline.
A general money-saving rhythm: first validate direction with a low-price tier, then re-run confirmed assets with a higher tier. Do not pay full price for every trial and error.
Integration Troubleshooting and Verification Checklist
The most important mechanism during integration: requests with invalid parameters are rejected outright and are not charged. This means you can boldly trial-and-error during development—misspelling a field name, using a malformed URL, or exceeding the view-count range costs you an error, not a bill. Make good use of this feature: get parameter validation working before discussing output quality.
Pre-launch verification checklist:
- URL accessibility: test
image_url/image_urlswith an anonymous request to confirm they are HTTPS and accessible without authentication. Being able to open it locally does not mean it can be fetched from the public internet. - Multiview order: check each image to confirm it is front, left, back, right. This is the most frequent silent error—the request succeeds, but the result is worse, making it hard to attribute.
- View count: strictly keep multiview to 2–4 images.
face_limitbounds: if specifying manually, confirm it falls within 1000–1,500,000; only withdetailedis the upper limit relaxed to 2,000,000. Out-of-range values will be rejected.- Toggle consistency with expectations: check whether the Texture / PBR toggles match the price tier you expect. The most likely discrepancy between the default configuration and the final configuration is "thinking it is off when it is actually on."
- Output verification: each call should yield one GLB and one preview image. Validate at the integration layer that both exist before proceeding downstream.
- Failure retry strategy: distinguish "parameter rejected" (change the parameters; do not retry) from "network/service exception" (retryable), to avoid repeatedly making billed calls for requests that are bound to fail.
Go through this checklist once, and the vast majority of integration-period issues can be resolved before the first launch.
The three input paths correspond to three states of source material, the two toggles determine the price magnitude, and face_limit only needs to be touched when you have a clear budget. Decide in this order, and integrating Tripo H3.1 will be quite straightforward—just start with your first call from the model page.
Related articles
View all articles →-
Guide
Using qwen-image-2.1-turbo for E-commerce Product Images via the HiAPI API
A developer-first workflow for fast catalog visuals, transparent PNG output, and a working API request
9 min read -
Guide
Using topaz-upscale/video via the HiAPI API: A Short-Form Video Upscaling Workflow
Frame interpolation, enhancement settings, and a working HiAPI request for vertical video
10 min read -
Guide
Using gpt-image-2.5-sunburst/image-to-image for E-commerce Product Images via the HiAPI API
Reference images, background swaps, lifestyle scenes, and a working HiAPI request
13 min read