Guides
Best image-to-3D AI for 2027: which model to use for one photo, and which for several
Image-to-3D AI to use going into 2027, sorted by single image or multi-view input: Hunyuan 3D 3.1, Rodin, Meshy, Tripo, Seed3D, TRELLIS.2 and SAM 3D.
5 min read

Quick answer
What is the best image-to-3D AI going into 2027?
It depends on how many pictures you have. From one photo, Meshy 7.1, Tripo P2.0 and the open TRELLIS.2 are strong picks as of November 2026. With several views of the same object, Hunyuan 3D 3.1 (up to eight views) and Rodin Gen-2.5 use them best. Test the same image across two or three models before you commit.
At a glance
| Model | Input | Output notes | Where it runs | Free or licence |
|---|---|---|---|---|
| Hunyuan 3D 3.1 (Tencent) | One image, up to 8 views, sketch, text | PBR; part generation, smart topology, UV unfold, auto rigging add-ons | Web, Tencent Cloud API | Web free daily generations (check); API 200 one-time credits |
| Rodin Gen-2.5 (Hyper3D) | One or several images, text | High-poly quads, Smart Low-Poly, textures to 8K (12K beta), part splitting | Web, API (Business+) | Free previews; Creator $30/mo |
| Meshy 7.1 | One image, multi-view, text | AI texturing, Smart Topology low-poly | Web, API (Pro+) | 100 credits/mo; CC BY 4.0 with attribution |
| Tripo P2.0 | One image, multi-view, text | Native quads, Smart UV, up to 4 face-count versions | Web, API | Check; free output non-commercial per secondary sources |
| Seed3D 2.0 (ByteDance) | One image | Geometry plus unified PBR; parts and joints; URDF export | Volcano Engine API | Billed by usage |
| TRELLIS.2 (Microsoft) | One image | PBR GLB (base colour, roughness, metallic, opacity) | Your NVIDIA GPU (24 GB official), ComfyUI, HF Space demo | MIT |
| SAM 3D Objects (Meta) | One image plus an object mask | Gaussian splat by default; mesh via a code flag | Linux, 32 GB NVIDIA GPU | SAM License, gated weights |
What this page ranks, and what it can't
This is a list of what to use going into 2027: the image-to-3D picks as of November 21, 2026, sorted by the one question that changes the answer most, which is how many pictures you have. We'll update it as 2027 releases land. Nobody has 2027 data yet, so there are no predictions here about models that haven't shipped.
It's a ranking by documented features, inputs, outputs and licences. We didn't run a quality test, and we don't publish scores. If you're starting from a prompt rather than a picture, see best text-to-3D AI for 2027 instead. For the full field, see the hub: best 3D AI for 2027.
Single image or multi-view: why it matters
A single photo shows one side of an object. The model has to invent the back, the underside and anything hidden. That's why the back of a generated model is often where things go wrong.
Multi-view input gives the model more of the truth: a front, a side and a back of the same object, ideally with the same lighting and scale. It doesn't make the model perfect, but it takes guessing out of the parts you care about. Multi-view is worth it when:
- the back of the object differs from the front (a backpack, a character with a cape)
- you need proportions to match a real product
- the single-image result keeps getting the depth wrong
If you only have one good photo, don't fake extra views. A clean single image beats three inconsistent ones. Tips for both are in best photos for image to 3D and multi-view image to 3D.
Best when you have several views
Hunyuan 3D 3.1
Tencent's HY 3D 3.1 (January 2026) takes up to eight views, the most of any model here, plus single images, sketches and text. Post-processing add-ons include part generation, smart topology, texture editing, UV unfolding and auto rigging. It runs on the web and through the Tencent Cloud API, which is prepaid at $15 per 1,000 credits. HY 3D 3.x isn't open weights; the newest open asset model is still Hunyuan3D 2.1.
Rodin Gen-2.5
Rodin Gen-2.5 (May 2026) accepts several images, high-poly quads or Smart Low-Poly output, textures up to 8K (12K in beta) and "BANG to Parts" part splitting. It added rigging on October 23, 2026, and prompt-editable rigging with motion presets on November 3. Creator is $30 a month. The free tier gives previews on some models and exports only from legacy ones.

Best when you have one photo
Meshy 7.1
Meshy takes single images and multi-view sets. Meshy 7.1 (September 2026) comes in Standard, Ultra 2K and Ultra 4K variants. It's the easiest single-image tool to try for free with commercial use in mind: 100 credits a month, and free assets are CC BY 4.0 (credit Meshy). Meshy 6 and Meshy 7 are the only Meshy models on the community leaderboard below.
Tripo P2.0
Tripo P2.0 (September 2026) gives native quad meshes from an image, Smart UV and up to four versions at different face counts. It also takes multi-view input. Tripo's pricing page blocked our fetch; secondary sources say free output is non-commercial, so check before you use it for client work.
Seed3D 2.0
ByteDance's Seed3D 2.0 (April 23, 2026) builds geometry with unified PBR materials from one image, splits parts with joints and exports URDF for simulators like Isaac Sim. It's available through the Volcano Engine API, billed by usage. ByteDance warns that some third-party sites sell credits under the Seed3D name without any official tie.
TRELLIS.2 (open)
Microsoft's TRELLIS.2 is MIT-licensed, single-image only (no text mode) and writes a PBR GLB. Officially it needs a 24 GB NVIDIA GPU on Linux; ComfyUI has supported it natively since v0.34. There's also a Hugging Face Space demo. Hardware routes are in image to 3D AI, locally.
SAM 3D Objects (open, with caveats)
Meta's model takes an image plus a mask of the object, so it can pull one object out of a busy photo. By default it outputs a Gaussian splat, not a mesh, and it needs a 32 GB GPU on Linux. Weights are gated.
What the community leaderboard says
3D Arena on Hugging Face asks people to vote between two outputs from the same image, with Elo ratings starting at 1200. On November 21, 2026 it listed 24 models and 114,236 votes. The top five: PicGen3D, TRELLIS-3DGS, Meshy-6, 404_GEN and Meshy-7. TRELLIS.2-4B was 18th. Two cautions: current Tripo, Rodin and Hunyuan 3.x models aren't listed, and new entries have few votes, so ranks move.
How to test image-to-3D models yourself
- Pick three images: a hard-surface product, an organic creature and something with a thin part.
- Use the same image in every model. Remove the background first.
- For multi-view models, run once with one view and once with all views, so you can see what the extra views bought.
- In Blender, check the back, the wireframe, the UVs and the texture maps. Then check the export format and the licence on your plan.
Trends going into 2027
- Edit, don't regenerate: Tripo Mesh Edit, Meshy Edit and Rodin's partial redo fix one region instead of starting over.
- Topology at generation time: native quads (Tripo P2.0), set face counts (Meshy T2.5), high-poly quads (Rodin).
- Rigging in the generator: Rodin, Hunyuan's API and Meshy now rig the result.
- Churn: Meshy retired meshy-5 from its API on October 10, 2026, and Tencent retired legacy endpoints (check). Re-test your favourite model every few months.
Run one image through many models
Upload a photo to Vi3W, run it through Meshy, Tripo, Rodin, Hunyuan 3D or TRELLIS without switching apps, and download the one you like as GLB, FBX, OBJ, STL, PLY or glTF. Try Vi3W.
Frequently asked questions
best image to 3d model ai
For one image, start with Meshy 7.1 or Tripo P2.0; for several views, Hunyuan 3D 3.1 or Rodin Gen-2.5. Results vary by object, so run the same photo through two of them and compare the topology, not just the preview.
what is the best image to 3d model generator
The one that does best on your kind of object. Community votes on 3D Arena put PicGen3D, TRELLIS-3DGS and Meshy-6 at the top in November 2026, but Tripo, Rodin and Hunyuan 3.x aren't on that board, so treat it as a partial view.
best ai for image to 3d
If you want to choose per image instead of per subscription, use a tool that runs several models. Vi3W runs 25 image-to-3D models, including Meshy, Tripo, Rodin, Hunyuan 3D and TRELLIS, from one upload.
Last reviewed

