Guides
How do AI 3D generators handle thin parts like straps, antennas and fingers?
Why AI 3D generators thicken, fuse or break thin parts like straps, antennas and fingers, and the input changes and Blender fixes that help.
7 min read

Quick answer
How do AI 3D generators handle thin parts like straps, antennas, or fingers without breaking the mesh?
Mostly by approximating them. Generators predict a shape on a fixed-resolution grid and then extract a surface, so anything thinner than that grid tends to thicken, fuse with a neighbour, break into fragments or vanish. You get better results from inputs that show clear gaps, multi-view images, higher-resolution modes, and generating fragile parts separately.
How do AI 3D generators handle thin parts?
AI 3D generators handle thin parts like straps, antennas and fingers by approximating them, and the approximation is where the mesh breaks. Most image-to-3D and text-to-3D models predict a 3D field (occupancy, or a signed distance to the surface) on a grid of fixed resolution, then extract a mesh from it. A feature narrower than a grid cell has nowhere to live. It gets rounded up to a fatter shape, merged with whatever is next to it, or dropped.
The research papers say this openly:
- The InstantMesh paper lists its 64×64 triplane resolution as a possible bottleneck for high-definition modelling, and notes its mesh extraction is "less effective on modeling tiny and thin structures".
- The Hunyuan3D-1.0 paper says fast feed-forward generators often "fail to generate thin, paper-like structures".
- The ShapeGen paper names "fragmented thin-shell structures" as an open problem, and explains the trade-off: signed-distance representations give smooth surfaces but fragment thin shells, while occupancy representations train more stably but produce "staircasing".
So the honest answer is that no current generator guarantees a clean strap or a separate finger. What you control is how much evidence you give it, and how you repair what comes out.
The four ways thin parts fail
Look at your model from every side and name the failure first, because each one has a different fix.
- Fused. The backpack strap melts into the backpack, the fingers turn into a mitten. The gap between two surfaces was narrower than the generator could resolve.
- Thickened. The antenna becomes a cone or a sausage. The generator kept the part but rounded it up to the smallest shape it could represent.
- Fragmented. The antenna is there, but as floating beads or a broken chain. This is the "fragmented thin shell" the papers describe.
- Missing. The part was hidden or too faint in the input image, so the generator never saw it. A strap behind the object in a front view is a common example.
Fused and thickened parts are usually input problems. Fragmented and missing parts are often easier to rebuild than to regenerate.
What generators do to keep thin parts intact
Vendors attack the problem in three ways. You can use all three when the tool offers them.
- Higher geometry resolution. Meshy's Image to 3D docs describe Ultra 2K and Ultra 4K resolution options for Meshy 7.1 that "generate the geometry at higher precision from the start", aimed at fine detail and printing. A finer grid means a thinner minimum feature.
- Part awareness. The same docs say Meshy 7.1 handles multi-part objects "where each part keeps its position and neighboring parts stay separate", and offer a Split option that segments the model into parts during generation (it can't be combined with Texture).
- More views. Multi-view input lets the generator see the strap from the side instead of guessing. Meshy accepts up to three extra views next to the main image; more on this in multi-view image to 3D.
At the time of writing (November 2026), these are features of specific tools and models, not something every generator has. Check what yours offers before you rely on it.

Input changes that save thin parts
These work with any generator, because they all increase the evidence for the thin part.
- Show the gap. Pick an angle where the strap clearly separates from the body, with background visible between them. A three-quarter view usually does this better than a straight front view.
- Contrast the colours. A black strap on a black bag is one blob to the model. If you control the reference (concept art, an AI image), make thin parts a different colour or value from what's behind them.
- Exaggerate a little. If it's stylised art, draw the antenna slightly thicker than you want. It is easier to slim a part in Blender than to rebuild one that vanished.
- Use a sharp, high-resolution image. Meshy recommends at least 512 × 512 pixels; Tripo's guide to source images suggests 1024 × 1024 or higher, because resolution "preserves seams, grooves" and similar fine detail.
- Open the hands. For characters, spread the fingers and keep hands away from the body. A clenched fist or a hand on a hip is a guaranteed mitten.
- Light evenly. Harsh shadows can be read as geometry and hide the edge you need. How to take a photo that converts well to 3D covers setup.
- Generate fragile parts on their own. A long antenna, a sword's thin crossguard or a chain often comes out better as its own generation, or as a simple primitive you add yourself.
Fixing thin parts after generation in Blender
When the body is good and only the thin parts failed, repair beats re-rolling.
- Find thin areas. In Edit Mode with Solid shading, open Overlays ‣ Mesh Analysis and choose Thickness. The Blender manual describes it as a ray-casting test that shows thin walls in red, built with 3D printing in mind.
- Find fragments. In Edit Mode, Mesh ‣ Separate ‣ By Loose Parts (P) puts every disconnected piece in its own object, so floating beads of a broken antenna become easy to select and delete. Mesh ‣ Clean Up ‣ Delete Loose removes stray vertices and edges.
- Rebuild instead of rescuing. Delete the broken antenna and add a cylinder in its place. For a strap, model a flat strip along the bag and give it depth with the Solidify modifier, which adds thickness to a surface.
- Be careful with voxel remeshing. The Remesh modifier's manual page warns that "thin parts of the input mesh can become loose" and offers Remove Disconnected to clean up the bits. Remeshing a model with thin parts can make them worse, not better.
The full clean-up routine is in editing an AI mesh in Blender and cleaning up an AI-generated model.
Generate, split or model: which approach for thin parts?
- Generate the whole object in one go. Fastest. Fine for chunky props and stylised characters with thick limbs. Expect fused straps and webbed fingers on anything delicate.
- Generate parts separately and assemble. More steps, but each part gets the whole grid to itself. Good for a robot with an antenna, a bag with straps, or a weapon with a thin guard.
- Model thin parts by hand. A cylinder, a curve with a bevel or a solidified strip takes minutes and is always clean. Best for antennas, cables, straps and anything that must print.
When not to rely on AI for thin parts
Some shapes are beyond what single-shot generation does reliably today: chains, wire fences, hair strands, lace, cloth with holes, spokes, thin lettering and see-through grilles. For games, those usually become textures with transparency on simple cards anyway. For printing, anything thinner than your printer can make needs to be thickened regardless of where the model came from; see make an AI model watertight for printing.
Where Vi3W fits
Vi3W turns a text prompt, sketch or image into a textured 3D model in the browser, and its automatic retopology rebuilds the dense result as a lighter low-poly mesh before you export GLB, FBX or OBJ. That makes the Blender repair step lighter. It does not get around the thin-part limit above: retopology simplifies the geometry that was generated, it cannot restore a strap that fused or an antenna that vanished. The input advice in this guide applies to Vi3W as much as any generator. Try Vi3W.
Frequently asked questions
how do ai 3d generators handle thin parts like straps, antennas, or fingers without breaking the mesh?
They approximate them, and often not well. Most generators work on a fixed-resolution 3D grid, so a strap, antenna or finger thinner than one grid cell gets thickened, fused or broken. Help the generator with an input image that shows clear gaps around thin parts, add side or back views if the tool supports multi-view, use a higher-resolution mode, and generate very thin parts as separate objects you attach in Blender.
Last reviewed

