Guides
Image to 3D inside Claude (and ChatGPT) with the Vi3W MCP server
Turn a photo, sketch or concept image into a textured 3D model inside Claude or ChatGPT with Vi3W's MCP server: setup, prompts and fixes.
6 min read

Quick answer
Can Claude turn an image into a 3D model?
With Vi3W's MCP server connected, yes. You drop a photo, sketch or concept image into the chat and Claude sends it to one of Vi3W's image-to-3D generators, then returns a textured model you can download as GLB, FBX, OBJ or STL. Without a connected 3D tool, Claude can describe the picture but can't build a mesh.
What the Vi3W MCP server adds to Claude
Claude reads images well. It can tell you a sketch shows a lantern with a hexagonal cap. What it can't do alone is build that lantern as a mesh. The Vi3W MCP server gives Claude the tools for that: image-to-3D generation with the same generators as the Vi3W app (TRELLIS, Hunyuan 3D, Meshy, Tripo, Rodin and others), and then the follow-up tools to rig, animate, retexture and clean up the result.
You stay in the conversation. Attach a picture, say what you want, and Claude handles the calls. The model is saved to your Vi3W account, where you can open it in the 3D viewer at any time.
If you've never connected an MCP server before, MCP for 3D artists explains the idea in five minutes.
Connect Vi3W to Claude
Claude's custom connectors use remote MCP and work on Free, Pro, Max, Team and Enterprise plans, per Anthropic's connector guide.
- Find Vi3W's MCP server URL in the Vi3W app.
- In Claude, go to Customize ‣ Connectors, click + Add, then Add custom connector.
- Name it "Vi3W" and paste the URL exactly, including any path.
- Review the authentication step and sign in to your Vi3W account when asked.
- Click Add. In a new chat, open the "+" menu and check Vi3W is switched on.
On Team and Enterprise plans, an Owner adds the connector once, and everyone else just clicks Connect.
For Claude Code, add the same URL as an HTTP MCP server in its configuration. For ChatGPT, use Plugins ‣ + ‣ Add custom MCP server, then select Vi3W with @ in a chat. OpenAI's custom MCP article lists which plans get full write access.
Your first image-to-3D request
Attach an image and be specific about what you want back:
"Make a 3D model of the lantern in this sketch. Game prop, stylised, about 40 cm tall. Use TRELLIS first. Show me the result and give me a GLB link."
Claude asks for permission before it calls a connector tool (unless you've set the tool to always allow), runs the generation and reports back. Open the link or the Vi3W app to look at the model from every side. A chat thumbnail won't show you the back.

Sketches vs photos vs concept art
- Sketches are rendered into a clean, shaded reference image first, then built in 3D. Ask Claude to show you that image before it generates: a bad render means a bad mesh.
- Photos work best of one object, whole in frame, on a plain background, with even light. Best photos for image to 3D has the checklist.
- Concept art works well if it shows one object from a three-quarter view. A busy illustration with five characters will confuse every generator.
Choosing a generator in a chat
You can name the model or let Claude pick. Useful phrasing:
- "Try TRELLIS and Hunyuan 3D on the same image so I can compare them."
- "This is a character, use the generator that's best for people."
- "Cheapest option first. If it looks rough, try a better one."
TRELLIS is the free default. Others cost XP and are often better at specific subjects. The Vi3W app shows each model's cost before you run it.
After generation: the rest of the pipeline in the same chat
This is where doing it in Claude pays off. Once the model exists, follow-ups are one line each:
- "Rebuild it at about 6,000 faces." Clean shape makes a lighter mesh (500 to 20,000 faces) with the textures baked on.
- "Repaint it as rusty iron." New paint job changes the textures and keeps the shape.
- "Add bones, it's a person, then make it wave." Add bones and Make it move. See auto-rigging through MCP.
- "Give me FBX for Unity." Downloads are free in GLB, FBX, OBJ, STL, PLY and glTF. Exporting from a chat covers formats.

Claude Code: straight into your project
In Claude Code, the image can come from your repo and the model can land back in it:
"Generate a 3D model from art/concepts/crate_front.png, clean it to about 3,000 faces, download it as GLB into Assets/Props/, and add it to the prop list in props.json."
That's the real benefit of MCP over the web app: the generated file joins the rest of your work without a manual download, rename and drag.
When the result looks wrong
Most bad image-to-3D results come from the image, not the generator. Before regenerating, check:
- Is the object cut off? Generators invent what they can't see, and they invent badly.
- Is it one object? A sword in a hand is two objects fused together.
- Is the background busy? Retake the photo against a plain wall, or crop tightly around the object.
- Is the view straight on or three-quarter? Extreme angles distort proportions.
Why image to 3D looks wrong goes through each failure with fixes.
What it can't do
- It won't keep a likeness of a real person reliably. Faces from photos come out approximate. Don't rely on it for portrait work.
- It won't build a scene from a scene image. A photo of a room gives you a lumpy room-shaped object, not separate furniture. Crop to one object per generation.
- It can't see the model the way you do. Claude can report face counts and dimensions, but you need to look at the model to judge it. Always open the 3D view before you build on a result.
- It doesn't replace hero-asset modelling. For background props and blockouts, it's fast and good. For the character on the box art, treat it as a starting point.
Try it with one picture
Pick one clean photo of an object on your desk, connect Vi3W to Claude, and ask for a GLB. Open Vi3W.
Frequently asked questions
claude image to 3d model
Connect a 3D generation tool such as Vi3W as a custom connector, attach the image, and ask Claude to make a 3D model from it. Claude passes the image to the generator and gives you back the model and a download link.
claude code image to 3d
Claude Code can use the same MCP server. Add it to Claude Code's MCP configuration, then point Claude at an image file in your project. It can generate the model and, because it works in your terminal, save the downloaded file straight into your assets folder.
chatgpt image to 3d model
ChatGPT works the same way once Vi3W is added as a custom MCP server. Full MCP with write actions is currently for Business and Enterprise/Edu workspaces on ChatGPT web, and Pro accounts can connect with read and fetch only.
Last reviewed

