Generative AI is increasingly becoming a staple in game development pipelines, but text-to-3D generation still lags behind its 2D counterpart in practical usability. However, by embracing a retro aesthetic, we can turn current limitations into stylistic strengths. This guide walks through a step-by-step workflow to create PS1-style 3D assets using open-source tools and AI models.
Prerequisites
Basic familiarity with Blender and concepts like materials and UV mapping is assumed.
Step 1: Generate a 3D Model
Start by visiting the Shap-E Hugging Face Space, which uses OpenAI's open-source Shap-E diffusion model. Enter a prompt (e.g., "Dilapidated Shack") and click 'Generate'. Download the resulting model.
Step 2: Import and Decimate in Blender
Open Blender (3.1+) and import the GLTF file. The model likely has a high polygon count. Apply a Decimate modifier (ratio ~0.02) to drastically reduce polygons. The result may look rough, but that's intentional for a low-poly style.
Step 3: Install Dream Textures
Add the Dream Textures Blender addon, a Stable Diffusion–based texture generator. After installation, download the texture-diffusion model from the addon preferences.
Step 4: Generate a Texture
In the UV Editor, open the 'Dream' tab. Set the model to texture-diffusion, enable seamless mode, and enter a subject prompt like "Wood Wall". Generate and save the texture. Apply it to your model by creating a new material and assigning the image texture to the base color.
Step 5: UV Mapping
Enter Edit Mode (Tab), unwrap the model with 'Smart UV Project'. Switch to rendered view to preview. Scale the UV map to tile the texture seamlessly. Keep it simple to maintain the retro vibe.
Step 6: Export the Model
Export as FBX (File > Export > FBX). Your low-poly, textured asset is ready.
Step 7: Import into Unity
Bring the FBX into Unity (or your engine). To emulate a PS1 aesthetic, use custom vertex-lit shading, disable shadows, add heavy fog, and apply glitchy post-processing. Resources for recreating the PS1 look are available here.
Conclusion
This workflow shows how to turn AI-generated 3D models into usable game assets by leaning into low fidelity. As text-to-3D models improve, these techniques could scale to realistic styles. Share your creations or join the discussion on the Hugging Face Discord.