Physically Based Rendering (PBR) has become the industry standard for creating realistic materials. But there are several different workflows for getting there — and choosing the right one can save you hours.
In this article, we’ll compare four methods for creating PBR materials: Photogrammetry, Procedural Generation, Hand-Painted, and AI-Assisted Generation. We’ll look at the pros and cons of each, as well as some benchmark data for processing times.
These are the 4 example materials used for the benchmark tests. You can download them and use them in your own projects. These Materials are made with the Photogrammetry method and I recreated them in the other methods to compare the results. The Materials are free to use and are licensed under CC0.
1. Photogrammetry
Photogrammetry involves capturing real-world surfaces with a camera and processing the images to create high-fidelity PBR materials. The results can be stunningly realistic, especially for natural materials like rocks, wood, or other Surfaces. With this method, you can get real and true detail from the real world, including imperfections and variations that are hard to replicate manually. Plus, you get a grate displacement and normal map from the same source. The barrier to entry has also lowered significantly. You can even capture small objects or surfaces with a smartphone, making it accessible for hobbyists and professionals alike.
The downside is the time investment — you need controlled lighting like cloudy outside light, many photos, and dedicated software like RealityCapture, Metashape or Meshroom.
Photogrammetry Processing Time Comparison
Here are some benchmarks with diffrent hardware setups and software for photogrammetry processing times:
| Software | Hardware | Avarge |
|---|---|---|
| RealityCapture | Ryzen 7 5800x, 64GB RAM, NVIDIA RTX 3090 | ~63 minutes |
| RealityCapture | Ryzen 7 5800H Laptop, 32GB RAM, NVIDIA RTX 3060 | ~77 Minutes |
| RealityCapture | Intel Core i7 9700k, 32GB RAM, NVIDIA RTX 3090 | ~66 Minutes |
| RealityCapture Average | / | ~69 Minutes |
| Meshroom | Ryzen 7 5800x, 64GB RAM, NVIDIA RTX 3090 | ~158 Minutes |
| Meshroom | Ryzen 7 5800H Laptop, 32GB RAM, NVIDIA RTX 3060 | ~195 Minutes |
| Meshroom | Intel Core i7 9700k, 32GB ,NVIDIA RTX 3090 | ~235 Minutes |
| Meshroom Average | / | ~197 Minutes |
1. Reality Capture
Benchmark graphs for Reality Capture:
All Reality Capture Avarage processing times with diffrent Hardware Setups:
2. Meshroom
Benchmark graphs for Meshroom:
2. Total Average of all examples, Setups and Programs
2. Procedural Generation
Procedural generation uses algorithms to create textures and materials. This method is highly flexible, allowing artists to tweak parameters like color, roughness, and pattern density. It’s particularly useful for creating materials that need to be tileable or for generating variations quickly.
The downside is that it can be challenging to achieve photorealism, especially for complex surfaces. However, procedural tools like Substance Designer or Blender’s shader nodes can produce impressive results with enough skill and experimentation.
In the following examples, you can see some procedural materials created to match the photogrammetry examples. Each material is fully customizable, allowing for a wide range of variations. I also show the processing time for each material, which is takes significantly more time than photogrammetry with reality capture.
I provide a Picture of the final node setup and the project file for each material, so you can see how they were constructed. The procedural materials are free to use and are licensed under CC0.
4. AI-Assisted Generation
Newer tools use AI to generate seamless textures from text prompts or reference images. While still maturing, they can dramatically speed up early-stage material concepting.
For measuring the processing time of the AI-Assisted Generation method, I used the same 4 example materials and generated them with 2 AI tools. The processing time was significantly faster than the other methods.
Prompts for the 4 example materials generated with both method
Example 1
Example 2
Example 3
Example 4
AI-Assisted Generation Methode 1
For this method, I used Adobe Firefly with Google Gemini (Nanobanana) for the Basecolor and Substance Sampler to generate the other maps. These are the results of the 4 example materials generated with this method.
Processing time for the 4 example materials generated with this method.
AI-Assisted Generation Methode 2
Processing Times
Average and Comparison of the Methods
Conclusion
In conclusion, each method of PBR material creation has its own strengths and weaknesses. It depense on the project requirements, available resources, and desired level of realism.
Photogrammetry is perfect for achieving high realism, especially for natural materials, but it can be time-consuming in processing times. But it has grate ways to automate this process. wile im reasearching this article for a seminar work, I created a Automation Script for Reality Capture, which can automate the whole process of photogrammetry. I will provide the github repository for this script in here
Procedural generation offers flexibility and control, making it suitable for creating tileable textures and variations. However, it may require more time and expertise to achieve photorealistic results. Also the user needs to have experience with the software and the node setups to create the materials.
Ai generated materials are the fastest methode. but the results are not as good as the other methods. The user has less control over the final result and the quality of the materials is not as high as the other methods. But it is a good methode for concepting and generating ideas for materials. The lack of quality is often seen in normal and displacement maps, which are often not as detailed as the other methods. But the user can use the generated materials as a base and then tweak them in a software like Substance Designer or Blender to achieve the desired result.
| Method | Realism | Speed | Control |
|---|---|---|---|
| Photogrammetry | ★★★★★ | ★★☆☆☆ | ★★☆☆☆ |
| Procedural | ★★★★☆ | ★★★☆☆ | ★★★★★ |
| AI-Assisted | ★★★☆☆ | ★★★★★ | ★★☆☆☆ |