Buying guide · Generators
Best Image-to-3D AI Tools
Which image-to-3D AI tool to use depends on your input: one photo, several views, or concept art. Picks by documented input modes, credits, formats and rights, Nov 2026.
Choose an image-to-3D tool by the kind of image you have, not by a general ranking. For a single clean image Meshy is the best-documented hosted starting point; if you can supply several views, Rodin, Tripo and Meshy all document multi-image routes; and if you want to run it yourself, TRELLIS.2 takes a single image under an MIT license. This page is about picking the tool by input mode. For how to prepare the image, see image-to-3D input preparation, and for printable files see image to STL.
How the picks were made
The criteria are the input modes each tool documents, the credit price per image-to-3D run, export formats, and what the vendor says about rights in what you upload and what you get back. No output was compared. We read no source that publishes per-image results across tools we could check, so this page says nothing about which tool reproduces your image most faithfully. The pick for each situation is a judgment based on how complete the published prices, limits and rights wording are. See the methodology.
| Tool | Price | Input modes documented | Cost of an image run | Best for |
|---|---|---|---|---|
| Meshy | Free; Pro $20 | Image to 3D; Multi-Image to 3D (API; multi-view not on Free) | 20 credits web (25 on the Ultra 2K default for paid plans); API 20 mesh-only, 30 textured | Single image, published costs |
| Rodin / Hyper3D | Creator $30 | 1 to 5 images on Gen-2.5; multi-image "Limited" on Free, a listed feature on Creator and above | API 0.5 credit base (+0.5 and +2.0 for top tiers); web credits unclear | Multi-image for detailed objects |
| Tripo | Not confirmed | Image to 3D, Image to Multiview, Multiview to 3D (API) | 20 to 30 credits (API, $0.01 each; H3.1 no or standard texture, P2 is 100 to 130) | Pipelines needing generated views |
| TRELLIS.2 | Free, own GPU | Single image; shape-conditioned texturing | None (needs 24 GB GPU) | Self-hosted single image to GLB |
| Hunyuan3D 2.1 | Free, own GPU | Image to shape plus PBR paint; Omni adds point cloud, voxel, box or pose control | None locally; hosted API 25 credits for textured Normal | Self-hosted with extra control (Omni) |
| Sloyd | Plus $15 | Image to 3D; multi-image to 3D | Multi-image 20 credits | Low-poly props from references |
| 3D AI Studio | Not confirmed | Image and multi-image (Studio and above), sketch and sculpt to 3D | 20 to 60 credits by model (FAQ; its pricing page says 20, its homepage 25 to 35) | Trying several models on one image |
| Kaedim | Not confirmed (custom plans) | Sketches, reference packs, product photos with a review loop | Not published | Managed studio-style work |
If you have one photo or one piece of concept art
Meshy is our pick here because its image-run cost, free-tier limits and rights wording are all published. At 20 credits a run, a Pro plan’s 1,000 credits cover 50 runs a month; at the 25-credit default that Meshy’s docs say paid plans start from, 40. That is arithmetic, ignoring retries and texturing add-ons, so it overstates the number of usable models. Meshy’s help center says paid ownership is “provided that you have used materials that do not violate the copyrights of others,” which matters for image input: you need the right to use the image you upload. That applies to every tool here.
Documented limits: Free users can download only untextured Meshy 6 Lite models, and a single image shows only one side, so the back of the object is a guess by any single-image tool, including the self-hosted ones. TRELLIS.2’s model card says raw meshes may contain small holes, which is a documented limitation vendors rarely headline; see repairing holes and floaters.
If you can supply several views
Rodin’s Gen-2.5 documentation allows one to five images, and its plan table lists multi-image as “Limited” on Free and as a feature on Creator and above. Meshy’s multi-image route is documented for the API, with multi-view excluded from Free. Tripo documents multiview-to-3D and a step that generates views from one image, though its consumer plans are unconfirmed. Whether several views actually beat one image is the question of the planned single versus multi-view test; no result exists yet.
If you want free or self-hosted
Free AI 3D generators covers terms. In short, TRELLIS.2 (MIT, at least 24 GB, Linux-tested) and Hunyuan3D 2.1 (10 to 29 GB, territory-limited license) are the open options. Detail is in open-source AI 3D generators.
If you want a managed service
Kaedim’s docs say it offers custom plans only and no free credits, subscriptions or trials; its homepage describes a review loop (“Review,” “Markup,” “Iterate,” “Approve”) for sketches and product photos. Its docs FAQ lists exports including .obj, .fbx, .glb, .gltf and .usd and says models are the user’s IP while subscribed to a plan. We could not confirm any price; its pricing page returned a 404 and third-party price lists conflict. Treat it as an enquiry, not a plan.
Photo scans are a different job
Image-to-3D tools generate a shape from pictures, so unseen sides are inferred, not captured. If you need a measured copy of a real object, photogrammetry is a different method; see photo-to-3D AI versus photogrammetry.
What each pick means for the destination
The tools differ in what comes out, and the documented formats decide how much conversion you do. Meshy exports GLB (textures embedded) and FBX (separate texture files, rig kept for animated models); TRELLIS.2 documents a GLB output, exported opaque by default; Hunyuan3D’s hosted API returns a GLB. Godot’s manual recommends glTF 2.0 and Unity’s recommends FBX, so a GLB-only tool suits Godot directly and Unity only after conversion. For printing, an image-run model needs more than an STL export, which none of these tools’ documents promise to be printable; see image to STL. If you intend to sell the result, the wording that matters is the vendor’s, not ours: Meshy’s help center ties paid ownership to your not violating others’ copyrights, and Kaedim’s FAQ ties IP to an active plan.
Downsides that apply to every pick
Paid hosted tools charge credits per run, and expect to pay again for retries (we did not find a documented per-task retry allowance on the Meshy web-app credit page). Self-hosting trades that for a 24 GB GPU (TRELLIS.2) or a license with territory limits (Hunyuan3D 2.1). Multi-image routes are plan-gated on Meshy (not on Free) and Rodin (Creator and above), and no source we read shows that more views reliably improve the result.
What evidence would change the pick
Per-input-type results for photos, concept art and product shots, with files you can inspect, would reorder this list. They are planned in the photo and concept-art input test; not run yet. The broader shortlist is best AI 3D model generators, and for text input see best text-to-3D tools. Our Meshy review has plan details. Back to best-of guides.