Hi everyone! I'm trying to learn how to use Leonardo AI for texturing 3D models. I've heard img2img is the way to go for this, but I'm a bit lost on the actual workflow.
Could someone break down a beginner-friendly step-by-step process? Like, what's a good starting base image? How detailed should my prompts be for textures? I'm also curious if there are any specific settings or models within Leonardo that work best for materials like wood, metal, or fabric. Any tips to avoid the common pitfalls would be awesome too! 😅
You've got the right idea about img2img being central. A solid workflow starts with a good base image. Render a clean, well-lit view of your 3D model from a standard angle. The render should have flat, uniform lighting, no shadows, and use a mid-grey base color. This gives the AI a clear "canvas" without baked-in lighting that would fight your texture.
For prompts, be highly descriptive of the material properties, not the object shape. Instead of "a wooden barrel," prompt with "seamless aged oak wood texture, rough grain, realistic varnish sheen." The detail should focus on surface qualities. For settings, start with a lower Denoising Strength (around 0.3-0.5) to preserve your model's geometry while letting the AI repaint the surface. Use the Albedo model in Leonardo for materials, as it's trained on textures without lighting information. A common pitfall is using a base image with shadows, which the AI will interpret as part of the texture and bake in permanently.
You'll likely need to generate multiple texture passes. Do one for the base color/diffuse, then use that result as a new base image for a second pass to add details like scratches or rust, increasing the denoising strength slightly for that pass. Finally, consider generating a separate specular/roughness map by prompting for "grayscale roughness map for [material]" using a black and white version of your diffuse output as the base.
null
user1191 covered the base render and prompts well, but the Denoising Strength range is too broad for a beginner. Starting at 0.5 will often warp your geometry. You need to treat it like a controlled variable.
Begin at 0.2 and increase in increments of 0.05, assessing surface detail versus shape preservation each time. For the model itself, Albedo is correct for diffuse maps, but for any material with specularity, like metal or wet wood, you should be using Leonardo's Diffusion model. It handles subtle reflectance better.
A major pitfall is inconsistent texture across UV seams. Always generate a texture for your unwrapped UV layout image, not just a beauty shot of the model. This requires a flat, templated base image, which is a different starting point than the well-lit render they described.
p-value < 0.05 or bust
You're absolutely right to call out the UV seam issue. It's the single biggest failure mode I see in these workflows. A beauty render is useless for actual texturing work.
The flat UV template approach is mandatory, but even then, you can get bleed and discontinuities if your denoising is too high. At 0.2, you'll often get a safe but bland texture that lacks the macro-variation you want in, say, stone or leather. The iterative bump-up method you described is correct, but I'd pair it with a mask. Use the UV islands as a mask in Leonardo's inpainting feature at a slightly higher strength (0.4-0.45) just at the seams to force continuity, while keeping the main surface pass more conservative.
Data over dogma
Great question! The biggest hurdle is that "good base image" means two very different things.
For learning the tool, start with a simple beauty render like user1191 said. It's fast and you'll see cool results immediately. That's the fun part. But for actual production texturing, you'll need that boring, flat UV layout snapshot to avoid seams.
For prompts, think "what would a photographer say?" Not "metal barrel," but "rusted iron surface, chipped paint, midday sun." Describe the light hitting the material, not the object.