Concept Art to Game-Ready GLB: A Solo Dev Pipeline With Trellis 2

Solo and indie game devs live in a specific kind of squeeze: you can draw concept art fast, but turning that concept into a model you can drop into Unity or Blender is a whole second skill — or a whole second person. Here is the pipeline I now use to close that gap, end to end.

The pipeline

1. Concept to single image. One clear character/prop concept per image, neutral background, 45° angle. If you can draw it in Procreate in 20 minutes, that's your input.

2. Image to 3D. I upload the single image to Trellis 2, an AI image-to-3D generator built on Microsoft's SLAT architecture (2B parameters, trained on 500K+ 3D objects). Single image to previewable GLB in 1-3 minutes, running in the cloud — no GPU required on my end.

3. Preview and prune. Spin the preview before committing. What you are checking:

• Does the AI's guess for occluded sides (back, bottom) make sense?
• Silhouette integrity from the gameplay camera angle
• Whether detail density matches the asset's on-screen size (a background prop doesn't need hero-mesh detail)

4. Export GLB. Trellis 2 exports GLB, which imports directly into Blender, Unity and Unreal. From there it's standard engine work: retopo if needed, LODs, materials.

Why this beats the alternatives at prototype stage

• Outsourcing: great for final assets, too slow and expensive for iteration — you don't outsource 12 variants of a barrel
• Photogrammetry: needs a physical object and 30+ photos; fine for scanning, useless for imagined props
• Local open-source models: need 16GB+ VRAM and a CUDA setup; my laptop has neither

Trellis 2 (https://trellis2.app/) being cloud-based is the practical difference: it runs in the browser, and plans start from $19/mo — less than one hour of outsourcing.

Where AI 3D still needs a human

Be honest about the limits: AI-generated meshes are rarely production-final. Expect to clean up topology on hero assets, redo UVs, and rebuild materials to match your art direction. What you skip is the blockout — getting form and proportion from concept to 3D, which used to be the slowest part of iteration.

For background props, prototypes, level-dressing and jam projects, the output is often shippable as-is.

A realistic day-one test

1. Take three pieces of your existing concept art
2. Generate all three on Trellis 2 (https://trellis2.app/) — under 10 minutes total
3. Import into your engine and drop them into a scene

If the silhouettes read correctly in your lighting, you've just cut days off your asset iteration loop. Official site: Trellis 2.