Rodin pricing, free plan and limits

Last updated: 2026-10-13

Rodin Gen-2.5 generates textured 3D meshes from text or up to 5 reference images, then exports GLB, OBJ or FBX files. Guidance options include bounding-box, voxel and point-cloud inputs, a Smart Low-Poly mode and BANG partial editing. It runs in a browser workspace, through a REST API, in engine and Blender add-ons, and over a hosted MCP server.

About Rodin

Rodin is the 3D generation product of Deemos, a Shanghai-based company founded in 2020 and operated through hyper3d.ai. Deemos announced a funding round of several hundred million Chinese yuan on 23 June 2026, led by Cathay Capital and SSCI Leading Private Equity Fund Management, and said the money would speed the worldwide rollout of Hyper3D Rodin. Yicai reports a client base that includes ByteDance, Unity, Figma, Canva.

You describe an object in text or upload reference images, and Rodin returns a mesh with PBR materials in formats that game engines and slicers read. The current family is Gen-2.5, while Gen-2 and the older Gen-1/1.5 tiers stay reachable through the API. Rodin itself is a hosted, closed-source service. Hyper3D says its products draw on research that includes CLAY, a 1.5-billion-parameter geometry generator presented as a SIGGRAPH 2024 technical paper co-authored by Deemos co-founder and CTO Zhang Qixuan.

Rodin sits between a concept image and a finished asset. Concept art made in Midjourney or Flux can serve as the input image, the output opens in Blender or a game engine through Deemos's own add-ons, and agents like Claude Code or OpenAI Codex can call it over MCP. Deemos also publishes official skills and a command-line client on GitHub, and its API page cites ISO 27001 plus GDPR certification. Zhang told Yicai the product is aimed mainly at modelers, game developers and 3D-printing hobbyists who clearly want to pay for it.

Within the media generation group on HokAI, the closest listed neighbour is Hunyuan 3D from Tencent. Two world builders sit nearby: Marble from World Labs makes persistent scenes, Project Genie from Google DeepMind makes explorable ones, and both aim at whole environments instead of single props. Hyper3D publishes no standard benchmark score that we could find, so the fair test is to run your own reference images through each candidate before you commit to a plan.

Pricing

50 each and you pay only after previewing a result. Creator is $30 a month, or $24 a month billed yearly ($288 up front). Business is $120 a month, or $96 a month billed yearly ($1,152 up front), and is the first plan with API access.

Enterprise is quoted by Deemos, and Education needs student or educator verification. 5-credit base for a task on the newest family, with extra credits for the Extreme-High tier and for the extreme-high texture mode, while the API marketing page says it starts from 1 credit, so confirm the live rate before budgeting. ai in October 2026.

For a price-ladder comparison, open the Hunyuan 3D record first, then Marble.

Key Features

• Text and multi-image input: A task takes a text prompt or one to five images, and the first image drives material generation, according to the API reference.
• Five quality tiers: Tiers run from Extreme-Low for simple assets to Extreme-High, and Extreme-High also returns an extra mesh with up to 10 million faces.
• PBR texture resolutions: Texture output starts at 2K and reaches 12K when the HighPack add-on is combined with the extreme-high texture mode.
• 3D ControlNet guidance: Bounding-box, voxel and point-cloud inputs steer the shape of a generation instead of leaving it to the prompt alone.
• BANG partial editing: BANG edits selected regions of a generated asset without rebuilding the whole model, and it is exposed as its own API task.
• Smart Low-Poly: Detailed assets can be reduced to cleaner low-poly meshes with smoother surfaces, a feature the pricing page marks as paid.
• Engine and agent connectors: Official add-ons cover Blender, Unity, Unreal, Godot, Maya and 3ds Max, and a hosted MCP server works from Claude Code, Codex and Kimi Code CLI.
• Export formats: The workspace exports GLB, OBJ and FBX, and the API documentation also lists USDZ and STL where the generation path supports them.

Pros

• Deemos lists SAML 2.0 single sign-on, role-based access and audit logs for team accounts, which matters to studios with procurement rules.
• The Free plan is pay by result, so the generated asset can be inspected before any credits are committed.
• An API data policy effective 3 August 2026 states that payloads and outputs are not used for model training and never reach the public gallery.
• The docs changelog runs back to July 2024 and records parameter changes month by month, which makes integration upkeep predictable.

Cons

• Trustpilot shows 2.9 out of 5 from 63 reviews (read in October 2026), split 43% one-star and 33% five-star, and the one-star reviews we read describe trouble cancelling after a trial and getting a refund.
• The terms of use call every payment final, label AI services non-refundable and bar cancelling or downgrading inside a paid period, so a wrong plan choice costs a full cycle.
• API access starts at the Business plan, and API payloads and outputs leave active systems after 7 days, so downloads must be stored by you.
• On-premise deployment, custom LoRA and fine-tuning sit on the Enterprise plan behind a sales contact, leaving smaller teams no way to tune the model.

Visit Rodin Official Website