Rhys Howell
Nov 18, 2026
"Image to PBR material" describes turning a photo or a textured 3D asset into a set of physically-based rendering maps (separate albedo, metallic, roughness, IOR and normal channels a renderer can actually light) instead of one flat colour image. M-XR's Marso Studio does this for a single image or an existing textured mesh, predicting that stack using image-to-material (I2M), M-XR's foundational material model, rather than baking a guess into a single texture.
Two different searches hide behind this one phrase, though. Some people have a single photo of a material and want a tileable swatch out of it. Others already have a textured 3D asset (scanned, sculpted, or AI-generated) and need the whole thing relit correctly under a moving light. This guide covers both, then walks through the mesh route step by step.
Image to PBR material is the process of predicting a physically-based material (typically albedo, metallic, roughness, normal and, where supported, IOR) from a 2D image, rather than hand-painting or hand-authoring each map. Each channel describes one physical property: albedo is surface colour with lighting removed, metallic marks which texels behave as metal versus dielectric, roughness controls how sharp or diffuse a reflection is, and IOR sets how strongly a surface reflects at normal incidence. A normal map stores fine surface detail, so light catches bumps and grooves the mesh itself doesn't model.
The reason this matters more than it sounds: a single photo already has lighting baked into it. Highlights, shadows and ambient occlusion are all mixed into the pixels. Predicting separate PBR channels means pulling those physical properties back out so the material behaves correctly once your own engine relights it, instead of looking right only under the lighting it was photographed in.
Most confusion in this category comes from two genuinely different tasks sharing the same name. The first is generating a tileable material swatch from a single reference photo (a photo of a wall, a plank of wood, a scrap of fabric) for use as a repeating texture. The second is taking a textured 3D asset you already have and predicting a full, separable PBR stack across its existing base-colour texture, so the whole mesh behaves consistently under dynamic light.
Marso Studio does both, with Generate PBR Textures. Give it a single image and it predicts a PBR material from it, with the option to make the result tileable. Give it a textured mesh and it predicts materials across the entire surface at once, matching however many distinct materials are already laid out on it.
Free single-photo tools generally take one of two approaches, and it's worth knowing which one you're using. Some run entirely client-side using classic image-processing heuristics — edge-detection for surface detail, luminance-based classification for which pixels read as metal, variance-based estimation for roughness — with no upload and no cost. Others use a small trained model instead of pixel heuristics, so the result follows material regions rather than raw brightness.
Adobe Substance 3D Sampler sits a level up from both. Its "Image to Material (AI Powered)" filter recognises shapes and objects in a photo, strips shadows and highlights out of the albedo, and derives roughness and height from the same image. It ships inside a paid Substance 3D subscription.
All three approaches share a limitation for this use case: they work on one photo of one material at a time. None of them take a textured 3D mesh and predict a coherent material stack across its whole surface, which Marso Studio also does.
Browser heuristic tools — Input: Single photo; What it predicts: Albedo, roughness, metallic; Method: Edge-detection and luminance heuristics, client-side; Cost: Free
Substance 3D Sampler: Image to Material (AI Powered) — Input: Single photo; What it predicts: De-lit albedo, roughness, height; Method: ML-assisted, Adobe Substance 3D subscription; Cost: Paid subscription
Marso Studio: Generate PBR Textures (image) — Input: Single image; What it predicts: A PBR material from one image, optionally tileable; Method: Predicted by I2M, trained on real measured material data; Cost: Credits per run, see marso.app
Marso Studio: Generate PBR Textures (mesh) — Input: Textured mesh (GLB, USD, or FBX); What it predicts: Albedo, metallic, roughness, IOR and normal, across the whole asset; Method: Predicted by I2M, trained on real measured material data; Cost: Credits per run, see marso.app
Marso Studio turns a textured mesh into a full PBR stack in a handful of steps, entirely in the browser. Starting from a single image of a material instead? Run Generate PBR Textures on the image and choose whether the result should tile.
Export your mesh as GLB, USD or FBX with an image-based base-colour texture. Vertex colours alone are rejected. No mesh yet? Start from an image instead and Generate 3D Model builds one first.
Load the mesh into Marso Studio at marso.app.
Run Generate PBR Textures. I2M predicts albedo, metallic, roughness, IOR and normal across the whole textured surface at once, not one swatch at a time.
Check it under a few HDRIs and rotate the light. This is the fastest way to catch a material that only looks right under one fixed lighting setup.
Export the individual maps as PNG, or the whole thing as a textured USD, and bring it into your engine or DCC.
Pick by what you're starting from, not by which tool you land on first. A single photo of a repeating surface (a fabric swatch, a ground plane, a brick wall) calls for a photo-to-material tool: Generate PBR Textures in Marso Studio, which can also make the result tileable, a free browser tool for quick iteration, or Substance Sampler if you already pay for the Adobe ecosystem and want manual control over the result. A textured 3D asset that needs a consistent, separable material stack across its whole surface (including AI-generated meshes) is the job Marso Studio predicts for directly.
If your bottleneck is turning a finished or AI-generated mesh into consistent, engine-ready materials, that's the gap Marso Studio is built for: GLB, USD or FBX in, with an image-based base-colour texture; albedo, metallic, roughness, IOR and normal out, as PNG maps or a textured USD; running on Generate PBR Textures, or on Generate 3D Model first if you have no mesh yet. Credits are charged per step and current pricing is on marso.app.
The output is physically-based PBR predicted by I2M, trained on real measured material data captured with Marso Measure, not scraped internet images. Marso Measure is the part of the pipeline that produces actual measured PBR, directly from photographed real-world objects; I2M's predictions inherit their physical plausibility from that measured training data rather than reproducing it directly. For the full picture of how the model works, see What is Image-to-Material (I2M)?, and for the measured-data foundation behind it, What is Measured PBR? and why high-quality 3D data matters for AI training.
What does "image to PBR material" mean? It's the process of predicting a physically-based material — albedo, metallic, roughness, normal and, where supported, IOR — from a 2D image or a textured 3D asset, instead of hand-authoring each map separately.
Is there a free way to convert an image to a PBR material? For a single-photo swatch, yes — free browser tools use image-processing heuristics at no cost, and some hosted tools offer a free tier. For predicting a full, separable stack across an existing textured mesh, Marso Studio runs on credits rather than a flat subscription.
What's the difference between a photo-to-swatch tool and predicting PBR for a whole 3D mesh? A photo-to-swatch tool takes one reference image and produces one repeating material. Predicting PBR for a mesh takes an asset that may already carry several distinct materials and generates a coherent stack across the entire textured surface at once. Marso Studio does both.
Can I do this online without installing anything? Yes. Marso Studio runs entirely in the browser at marso.app, as do several free single-image tools. Substance 3D Sampler's AI Powered filter requires the desktop application.
What file formats does Marso Studio accept? GLB, USD and FBX, each with an image-based base-colour texture. Vertex-colour-only meshes are rejected. For a single material, a single image is enough. You can also start from an image and generate a mesh first.
Try it on your own asset — bring an image or a mesh into Marso Studio and see the full PBR stack in minutes.