LogInLisa Bilai shared with us the creation of optimized and game-ready kitchen assets, including fruits, vegetables, and a bowl, using the photogrammetry technique and RealityCapture. She discussed her setup and how to capture full rotations, along with workflows for items that are tricky to scan.
Introduction
Hi, my name is Lisa Bilai. I’m a 3D Environment and Scanning Artist at ScansMatter. I got into 3D during the COVID lockdowns when I suddenly had more than enough time to dive into a new profession. I quickly landed a job in mobile games, but I was always drawn to a more realistic style. So, alongside my work, I kept practicing and building a portfolio focused on realism.
When the war in Ukraine began, I had to leave my home and job, but that difficult chapter led me to my first AAA studio job at Bloober Team. There, I worked as a Prop Artist on the Silent Hill 2 Remake and contributed to Cronos: The New Dawn.
Still, even while creating realistic props, I felt like something was missing; I wanted even more realism in my work. That’s when I started exploring photogrammetry, which eventually led me to ScansMatter, where I currently work as an Environment / Scanning Artist.
About The Kitchen Food Photogramerty Project
I’ve always been drawn to the kind of art that nature creates — its textures, colors, and shapes are so rich and inspiring. There’s something really special about being able to capture that beauty and bring it into the 3D world. Photogrammetry felt like the perfect way to do that. It lets me preserve all those little details with incredible realism. It’s both a technical challenge and an artistic pursuit, which makes it especially rewarding for me.
While working on a kitchen environment for ScansMatter, I needed to fill the scene and bring it to life. So, I decided the best solution would be to scan real food items, fruits, vegetables, and baked goods. Not only would it add realism, but it also gave me the opportunity to experiment and learn more about scanning organic objects.
In total, the project includes 21 high-quality food assets. For frequently used items like apples, carrots, and potatoes, I scanned two variations to provide more diversity and realism in scenes. Each item brought its own challenges, whether it was the shape, texture or general material response.
Some of my favorite scans from this set are the onion and the baked goods. The onion really surprised me. It has so many layers of color that aren’t immediately visible to the human eye, but the scan captured all of them beautifully. And the baked goods are just full of tiny, intricate details that make them really stand out in a 3D environment. This project really reminded me how much beauty there is in everyday things, especially when you take a closer look.
Since some fruits and vegetables spoil faster than others, I planned the scanning order based on how quickly each item would start to rot. That said, certain fruits like bananas actually looked more interesting and natural after they started to age a bit, so I leaned into that when it made sense.
I started with apples and oranges because they have simple, rounded shapes that are ideal for scanning. My setup included a Sony A7RIV camera, a turntable, two light sources on either side, and a RIKO 400 flash on the camera. Both the lens and flash also have a polarization filter to remove reflections, which is a must! One of the most important parts of the setup was a black backdrop. It made it much easier to isolate the object from the background during processing. They often call this the “Void” Setup. I placed each item on the turntable and captured several full rotations of around 25 photos from different angles. Ensure you have at least 50% overlap on each rotation.
To capture the bottom of an item I simply rotated the item 90 degrees since thanks to the void setup it will still align the pictures properly and combine all angles together.
For longer items like bananas and carrots, I had a very simple but effective rig: a chopstick stuck into a wooden base. I’d skewer the fruit or vegetable onto it, which allowed me to capture all sides of the object with minimal contact points obstructing the view. On average, I took about 100 to 200 photos per asset to ensure full coverage and good detail.
One of the trickiest objects to scan was a wooden bowl. Because of its depth and thin edges, it was hard to keep the entire object in focus. After several tries, I landed on a solid workflow: I marked the bowl with tracking points (like small stickers or dots), then scanned the top and bottom separately. Once both parts were captured, I aligned and merged them in RealityCapture using control points and then merged components. Here is a quick video I found on how to do it as it is quite a specific workflow.
Before importing the images into RealityCapture, I processed them in Lightroom to ensure the exposure was consistent across all photos and converted them from RAW to JPG to make the sizes more manageable. (Make sure not to compress your JPG.) Once the images were ready, I loaded them into RealityCapture and started the reconstruction process there.
For this, I leave almost all settings default except for the texture settings. For those, I make sure RealityCapture produces as many textures as needed to capture the resolution, rather than a single texture per object. For the mesh processing, I also processed this in high quality.
Below you can see my settings if you are interested:
After processing the scan in RealityCapture, I’m usually left with a very dense, high-poly model. So, the next step is optimization. I bring the model into ZBrush and use ZRemesher, which usually does a great job at creating a clean base mesh. Since the meshes I create are often static, the geo can be a bit messy. However, for round objects, I recommend doing manual retopology. I then move to Blender, where I refine the model further, adding or removing polygons manually to get the best balance between performance and quality.
Some objects, like cauliflower, baked goods, or garlic, often need more manual cleanup because of their complex shapes and fine details. In Blender, I also handle the UVs and prepare the model for texture baking. However, if you have a complex model, I highly recommend RizomUV, but for simple fruit, this is often not needed.
In Marmoset Toolbag, I follow a standard baking workflow where I project the high-poly details onto the low-poly mesh. My typical baking setup includes maps like Base Color, Ambient Occlusion, and Normal.
After baking in Marmoset, I move on to Unity ArtEngine to generate the roughness map and do any additional cleanup. (Please note that ArtEngine has been discontinued, but it still works great.) I use ArtEngine regularly to process both object scans and material scans. It’s a really helpful tool for filling in missing data and refining surface qualities. In this case, it allows me to create a detailed and accurate roughness map based on the Albedo, which is super important for achieving realistic surface behavior in-game. Once the roughness map is ready, I bring everything into Substance 3D Painter for the final texturing and cleanup.
Just a quick tip. If you need a highly accurate roughness map, you can also use a technique called “cross polarization” — only this technique is much slower to do.
After that, I bring everything into Substance 3D Painter. There, I use the stamp tool to clean up any artifacts left over from the scanning process or any extra tracking marks. Once all that’s done, the asset is game-ready.
Summary
The whole project took me around two weeks to complete. I tried to optimize my time as much as possible — for example, while one batch of photos was processing or RealityCapture was working, I’d already start scanning the next object. It felt a bit like multitasking on overdrive, but it definitely helped keep things moving quickly.
One of the biggest challenges? Honestly, resisting the urge to eat the assets before scanning them! On a more technical note, one of the trickiest problems I faced was scanning tomatoes and bell peppers. Their surfaces are very smooth, shiny, and almost completely uniform in color, which makes it really hard for RealityCapture to find enough visual reference points to align the photos correctly. I even tried adding extra tracking marks to give the software more to work with, but it still struggled to stitch everything into a single, clean model. Unfortunately, those assets didn’t make it into the final set, but it was a good learning experience.
The most enjoyable part of the project was seeing everything come together — the moment when all the scans are cleaned, optimized, and fully textured is really satisfying. It’s amazing to take something from the real world and turn it into a beautiful, game-ready asset.
As for learning photogrammetry, I highly recommend the Ultimate Photogrammetry Course by FastTrackTutorials. It covers everything you need from scanning objects and materials to processing and optimizing them for production. It’s a solid foundation for anyone getting into the field. I’d also suggest diving into the more technical side of photogrammetry early on, like understanding light behavior, camera settings, and the theory behind scanning, as they can help save a lot of time and headaches later. And most importantly, don’t get discouraged if your first scans aren’t perfect. Every scan is a learning experience.