This is the follow-up to generating a model with Tripo AI. It assumes you have a GLB you like and now need it in a real project.

How many triangles is too many?

It depends on the budget, but the numbers that matter for a game prop:

That gap is the whole problem. Decimation closes it for static props.

Should you decimate or retopologize?

Decimate for props that never deform. A Decimate modifier in Blender set to collapse at 10–20% will cut the triangle count and keep the shape recognizable. Check the silhouette from a few angles — that is what the player actually sees.

Retopologize anything that bends: characters, cloth, cables. Automatic retopology tools give you quad-dominant topology you can rig, which decimation cannot. There is no shortcut here; animating a decimated AI mesh produces crumpling.

What is wrong with the original UVs?

They are overloaded and uneven. Stretching shows up around thin features — straps, fingers, antenna. The fix is to unwrap a fresh UV set on the cleaned mesh and bake the original texture onto it.

Baking is the step people skip and then wonder why detail vanishes. Once you decimate, you have thrown away the geometry the texture was painted for; a bake transfers the color and normal detail from the high-poly original to the new low-poly UVs. Skipping it gives you a flat, smeared surface.

What about normal maps?

Bake a normal map from the original mesh. That is how a 1,500-triangle prop looks like it has the detail of the 80,000-triangle original. Without it, the decimated model reads as visibly faceted in a lit scene.

Keep the original high-poly export archived. You will want it every time you re-bake after a texture change.

When is the cleanup not worth the time?

When the asset sits far from the camera or appears once. A fence post in the background does not need retopology. Apply the full pipeline to assets the player inspects up close, and ship the rest decimated after a quick silhouette check.

Quick Summary

• Decimate static props to 10–20% of the raw AI triangle count.
• Retopologize anything that deforms — decimation ruins rigs.
• Rebuild UVs on the cleaned mesh and bake the original texture across.
• Always bake a normal map so low-poly assets keep surface detail.
• Archive the high-poly original for future re-bakes.

It is aimed at people who need an asset now and cannot spend a day modeling. Speed is the selling point, not production-grade topology.

How do you generate a model?

1. Sign in and open the generation panel.
2. Choose image-to-3D or text-to-3D.
3. For image input, upload a single object on a clean background — no clutter behind it.
4. Generate and download as GLB or FBX.

For text input, describe the object and its style: “worn leather satchel, game-ready, low poly”. Vague prompts return generic shapes.

What input gives the best result?

Image-to-3D beats text-to-3D nearly every time, and the photo quality decides the outcome. The inputs that work:

• One object filling the frame, plain background.
• Even lighting with no harsh shadow across the surface.
• A three-quarter angle that shows depth, not a flat straight-on shot.

Multi-view input, where you supply several angles of the same object, improves the back side considerably. If you only have one photo, expect the unseen side to be guessed.

Is Tripo AI free?

New accounts get a batch of free credits, enough for a handful of generations. After that it is a credit system, with higher-resolution or higher-poly output costing more per model. Free credits reset on a schedule for some plans. Check the pricing page for current rates — the per-model cost changes as the models improve.

What is wrong with the exported mesh?

Topology. You get dense, uneven triangles rather than the clean edge loops a modeler would build. That is fine for rendering, painful for animation — a rig will deform badly across messy topology.

UVs are also rough. Textures can stretch at seams, especially around thin features. Plan a cleanup pass in Blender for anything that needs a close-up or a rig.

Who should use this?

Indie game devs filling a scene with props, AR and VR developers who need quick assets, and 3D printing hobbyists who only care about the shape. Skip it if you need animation-ready topology or precise mechanical dimensions.

Quick Summary

• Tripo AI generates textured 3D models from images or text in minutes.
• Image input on a clean background beats text prompts for accuracy.
• Multi-view input improves the parts a single photo cannot see.
• Exports have dense topology and rough UVs — clean up for animation.
• Free credits to start, then credit-based pricing per model.

If you need full environments rather than single objects, see World Labs Marble.