PromptProp

Review · Generators

Rodin (Hyper3D) Review: Credits, License and Cleanup

Rodin by Hyper3D reviewed from its pricing, API docs and terms: pay-by-result credits, topology options, and a license that never says you own the output.

Rodin, made by Deemos and sold as Hyper3D, documents more topology and texture-resolution options than most hosted generators here, and is the one to be careful with if you need a clear statement of who owns the output. Its pricing shows plans but not the credits inside them, and its Terms promise not to limit your use of output without ever saying you own it.

What stands out: control over topology and detail

Rodin’s documentation describes tiered quality settings (Extreme-Low, Low, Medium, High, Extreme-High), where Extreme-High returns an extra mesh of up to 10 million faces. The Gen-2.5 API accepts one to five images, offers Quad or Raw topology, a Smart Low-Poly mode, 2K and 4K textures, and 6K up to 12K with a HighPack add-on. Vendor claims also include PBR materials and “ready-to-use UVs” in the FAQ. Treat those as claims, not results.

What survives the next step is the key question for a game developer. Quad topology is the option to compare if you intend to retopologize or subdivide, and a mesh of up to 10 million faces would need decimation or retopology before real-time use (we cite no official engine limit). See retopology for AI 3D models for how to choose a method. Plugins are listed for Blender, Unity, Unreal, Godot, Maya, 3ds Max, Cinema 4D, Omniverse and ComfyUI, plus an MCP server and an “Agentic API” announced in September 2026.

Who should use it and who should skip it

  • Use if: you want to preview before spending (the web app is described as “pay by result”, with the preview shown before you confirm), you want documented control over detail tiers and texture resolution, and you can verify the license against your plan.
  • Skip if: you need a documented monthly credit budget, you are working from a Free account for commercial output, or you need a plain statement of ownership for a publisher or a store.

Pricing and credits

As of November 2026 the pricing page lists Free $0, Creator $30 a month ($24 a month yearly, $288 a year), Business $120 ($96 yearly, $1,152 a year), and custom Enterprise. The page labels yearly “Save 20%” and, when read, also showed a banner reading “Prices can’t be shown right now. Try again.”

What is not stated: the monthly credit allowance of any plan. The page gives estimates instead, about 60 models on Creator and about 416 on Business. Dividing the price by those vendor estimates gives roughly $0.50 and $0.29 per model, but they are estimates and the unit is not defined. Direct credits cost $1.5 each and are described as dearer than subscription credits. API pricing is expressed in credits: Gen-2.5 base 0.5, Extreme-High geometry adds 0.5, and texture_mode=extreme-high adds 2.0, so a maxed-out generation lists at 3.0. If those API credits were the same unit as the $1.5 direct credits, a base generation would cost $0.75, but Hyper3D does not reconcile the two, and its changelog tells users to confirm current costs before submitting paid work. The changelog notes add-on and Generate Texture pricing updates in August 2026.

The pricing page’s Business entry lists API rate limits of 120 requests per minute on the first three tiers and 240 on higher tiers; it does not say which tier an account sits in, and the API spec pages we read give no figures.

Output and formats

Format lists conflict. The Gen-2.5 documentation gives geometry_file_format as glb, fbx or usdz; other pages list glb, usdz, fbx, obj and stl, and the homepage adds GLTF. Confirm the one you need on the plan you buy. Rigging and Generate Images API docs were added in October 2026 per the changelog, but the homepage does not mention rigging, and web-app availability is unknown. For rigging routes, see the rigging hub.

License: no ownership statement

This is the most important section. The Terms (section 5 for Rodin; re-read in full this pass) say “we will not limit your use of such Output, subject to any restrictions set forth in these Agreements.” They also say Hyper3D makes “no representations, warranties or undertakings of any kind as to the copyrightability of any Output.” The Terms do not state that users own Output. The pricing page says Creator has “Unlimited export and any use” and lists “Commercial license included” for Business under its ChatAvatar line, but it does not say who owns generated assets.

On a second read, the Terms do not define a Free plan or say Free outputs are public or under a Creative Commons license; the Creative Commons wording in section 5 attaches to ChatAvatar. Nor do they say what happens to exported output after account closure: section 12 only says Hyper3D “may at our option delete any Content or other data associated with your Account.” The Terms add that export may be “for private or commercial use depending on your subscription plan.” Whether Free outputs can be used commercially is not stated. A separate product, ChatAvatar, says users may not use content for commercial purposes without a purchased license; whether that applies to Rodin’s free plan is unclear and we make no claim either way.

Other clauses to know: Hyper3D may use prompts and output to provide, maintain and improve its services and support research; private mode means it will not use output apart from displaying it in your account; section 12 allows deleting account content; and a subscription “cannot be canceled or downgraded within the corresponding subscription period.” The API data retention policy (effective August 3, 2026) says request payloads and generated output stay in active systems for 7 days and are “not used to train models,” with exceptions for legal retention, narrow security and abuse metadata, and temporary backups. It covers API payloads only; account data falls under the Privacy Policy. For game publishers, the practical advice is to buy a paid plan, keep your generation records, and ask Hyper3D in writing if a store or publisher requires an ownership statement. This is not legal advice; see copyright and AI-generated 3D models and the license matrix.

Documented limits and behavior changes

Rodin’s API documentation is unusually specific about tiers, and the changelog shows several defaults moving during 2026. These are from the Gen-2.5 and Gen-2 specifications, the Rigging API page and the changelog, as read on 2026-11-08.

Item What the docs say Consequence
tier Required. “If omitted, the endpoint falls back to Regular and generates a Gen-1/1.5 asset.” Same wording on the Gen-2 page for tier=Gen-2 A script that forgets tier is billed and produces an older-generation asset without an error
Texture resolution 2K or 4K by default; 4K or 8K with HighPack; Extreme-High texture gives 6K, or 12K with HighPack. HighPack “raises texture resolution only” on Gen-2.5 and does not add geometry A 12K texture is a texture-memory problem in an engine before it is a quality gain
Extreme-High tier Returns an additional mesh with up to 10M faces A cleanup input, not a game asset; the 10M figure is the vendor’s cap, not an engine limit
Fast mode Applies only if the tier is not Extreme-High, format is glb, fbx or usdz, mesh_mode is Raw and quality_override is 20,000 or less The quick path is raw triangles, not quads
geometry_instruct_mode faithful is “experimental,” “may be discontinued” and “not recommended for production.” The default changed from faithful to creative in July 2026, and creative was restricted to Gen-2.5-Medium and higher in June Results for the same prompt can differ between a script run in spring and one in autumn
Rigging API Presets: human, quadruped, bird, dragon, kaiju, or auto; custom skeleton by hierarchy or description. A custom model upload does not preserve embedded materials and textures, and needs a reference image. Up to 10 animation clips, 25 MiB each A rigged creature has a documented route, but a textured upload loses its textures
Mesh simplify On Gen-1/1.5, mesh_simplify takes effect only when mesh_mode is Raw (Dec 2024 changelog) Applies to the older generation only; the Gen-2.5 spec we read does not mention it

A third-party host, Runware, documents Gen-2 face budgets by quality tier: roughly 50,000 quad faces at high and 18,000 at medium, 150,000 raw faces at medium, with custom quad counts from 1,000 to 200,000 and raw from 500 to 1,000,000. That page also says HighPack multiplies faces about sixteen-fold on Gen-2, which conflicts with Hyper3D’s own statement that HighPack is texture-only; the host’s table also lists fewer raw faces than quad faces at extra-low (2,000 against 4,000), the reverse of every other tier. Treat those figures as Gen-2 and third-party only, and note they were read from a reseller’s docs, not Hyper3D’s. Hyper3D’s Gen-2.5 page gives no face counts per tier.

What this means for a 40-prop kit: a single base generation is 0.5 API credits, so the batch is 20 credits at the defaults and up to 120 at the maximum setting (3.0 each), for a spread of 6x before retries. The 6x spread is the planning risk; the dollar value of an API credit is not stated. If it matched the $1.5 direct-credit price, which Hyper3D does not confirm, the range would be $30 to $180, and the plan allowance that would cover it is not published. For props that will get an LOD pass anyway, the Medium tier with the default texture mode is the documented cheap end, but nothing we read says how it looks. For a rigged character, the documented route runs from a completed Rodin asset into the Rigging API, which accepts rig templates beyond humanoids but is not described in the web app.

Documented weaknesses

  • The pricing page showed a price-unavailable banner when read, and credits per plan are blank.
  • Format and endpoint differences are not reconciled across docs.
  • No published polycount limits for the web app.
  • Effect of cancellation on already-exported assets is not covered by the Terms excerpt we read; see what happens when you cancel.
  • We found no independent output-quality data we trust, and no forum or GitHub thread about Rodin that we could read in full. Review pages we saw in search results were vendor-hosted or SEO-style and are not relied on.
  • Defaults moved during 2026 (geometry instruct mode, add-on pricing, texture parameters), so a result from an older tutorial may not reproduce today.

Alternatives

Head-to-head pages are planned on Meshy vs Rodin and Tripo vs Rodin. For a vendor with a readable credit table, see the Meshy review. Our methodology describes the planned checks; none has been run.

More generators are listed in the AI 3D generator reviews hub.

Sources

  1. Hyper3D pricingPrimary · read 2026-11-08
  2. Hyper3D TermsPrimary · read 2026-11-08
  3. Rodin Gen-2.5 API specificationPrimary · read 2026-11-08
  4. Hyper3D docs changelogPrimary · read 2026-11-08
  5. Hyper3D API data retention policySecondary · read 2026-11-08
  6. Hyper3D homepagePrimary · read 2026-11-08
  7. Rodin Gen-2 API specificationPrimary · read 2026-11-08
  8. Hyper3D Rigging APIPrimary · read 2026-11-08
  9. Runware docs: Rodin Gen-2 topology and polygon budget (third-party host)Secondary · read 2026-11-08