3D Graphic Designer Hary G discussed his approach to photorealistic environments: texturing, render setup, post-processing, and advice for novice artists.
I'm Hary, and I've studied in the field of computer engineering along with software studies (which I've lost interest in ever since I started working on 3D projects). I haven't worked professionally in the field yet but I am freelancing for now as this is the job I'd like to pursue.
I'm self-taught in the field of 3D design. I started working with 3D around mid-2017; from there on, I watched tons of tutorials, checked out how people did their art, got inspired by video game trailers and the details in them, and just wanted to do something similar or just close to that!
I haven't contributed much to any serious projects as I've basically been doing 3D art as a passion for a long time (other than contributing to little things like helping out with cosplays where people needed something modeled for 3D printing).
Why Photorealistic Scenes?
I started to work on photorealistic environments mainly because I got inspired by the amount of renders in that style done by other people and also by watching a few of the CGI trailers for video games. I really wanted to try out something that would make people question whether it's real or not without any context.
When it comes to detailing objects, I go for voxel modeling in Cinema 4D which helps to get those smooth edges. Also, connecting certain parts gives the models a more realistic touch compared to what you'd get with normal beveling and SubD. For things like creases and wrinkles, normal maps help out a lot and it's easy to make those in Photoshop or even take real-life shots and convert them to normals and bumps. The rest is up to mapping.
I can also use ready assets when photogrammetry/3D scans have better details than hand modeling. For example, the best cases I used those in were the fireplace in one scene, flowers, rocks, etc. – that sort of stuff. Apart from that, I've modeled a lot of objects since I started doing 3D, so sometimes I pick a few of those, rework them if they are too outdated, and use them in the scene. It saves a lot of time.
Texturing is more important than modeling to me as it gives life to every object regardless. I take my time to make the textures perfect, experimenting with roughness maps, scratches, bumps, fingerprints, scaling it all up and down to see what fits. Each of those adds detail to the scene. I do use Substance at times but mostly, I do all the work with Octane materials and messing around with dirt nodes and gradients. It's possible to get really detailed and perfect textures with Octane materials, even without using Substance.
Rendering and Post-Processing
The rendering setup I use is Octane's Path Tracing and a very high sample to drain out the noise and get a more realistic look with the PT. I haven't tried out other render engines because I felt comfortable with Octane as soon as I began working with it but I will give engines like Redshift a shot in the future.
Lighting is the most important part of any scene. Depending on the project, I always have an HDRI for the render. Sometimes, if Octane Daylight does a better job (for example, in the cyberpunk city render), I'll use an HDRI just for the sky visibility and the daylight sun with tweaks as the primary source. For the last photorealistic project, I went with an HDRI for both visibility and lighting.
As you may have seen, I've got kind of an overcast sky there, so it was more taxing to get that ambient light into the room. In this case, it's easier to use normal sunlight or even an HDRI that has sunlight in it and cast the light into the room; this is less expensive than an overcast sky. For that, I make two copies of an Octane Sky object, one for visibility and one for lighting. The power of the lighting one is cranked up to about 25 so that the light falls into the room and the visible one is set to around 10 creating a slight effect of overexposure.
Along with lighting, the color choice is also important for the image as it changes lighting drastically. In Octane, I use the Camera Imager and change the settings to linear which gives a more natural look, especially with a few tweaks of gamma and power. Octane's Camera Imager also has a few post-processing options such as bloom, glare, saturation, and contrasts that are really helpful in giving the scene a little boost in realism. It's like having lens flares and glow of a real camera.
Other than that, I process the image a bit with Photoshop's Camera Raw filter just to give it a boost adjusting clarity, contrast, highlights, blacks and whites, and so on. Nothing that ruins the image, just a little to bring out the details even further.
Rigs and Render Time
I've got two rigs, one with 1050ti and the other with 2070. For the last project, rendering took around 5-6 hours to get rid of the noise just enough for it not to be too overwhelming. I used around 8000-10000 samples for that in Octane PT. Usually, with these interior scenes, it takes a lot of time depending on how the lighting is done, what the textures and their physical properties are, if there are any displacements. But overall, for realistic interiors, I'd say the average render time for the final high-resolution output and a very high sample rate is around 4-6 hours.
With photorealistic scenes, the main challenge for me is to get the texturing part right! You've got no idea how many times I lost my patience while mapping each texture to an object perfectly along with its roughness and normals aligned. It's all about testing, re-doing, testing, re-doing again, and so on until it looks right from all angles. This stage is especially challenging in the case of large-scale scenes when you have a huge texture library and finding each texture to tweak it later is at times frustrating.
Overall, to craft an entire scene, it takes me around 90 hours, 120 maximum. The cyberpunk one though took nearly or more than 150 hours considering the number of objects that had to be made, with all the tiny tweaks, texturing neon signs and advertisements, scaling, making emissives, etc. That was a real nightmare but I gained a lot of experience which helped me to work faster on simpler projects.
Advice for Beginners Looking into Photorealistic Art
For beginners, it's possible to get a realistic look in a few tries if you focus on single-object scenes at first and work with HDRIs for the background with DOF. You just have to focus on getting the lighting right and make the textures look as perfect as possible with details like dirt and scratches. Next comes the modeling part, and trust me, it took a long time for me to get a grasp of it. But starting with simpler objects such as a wine bottle, a glass, a book, or a table would be a great way to set off on the photorealistic journey! And once you're done with single objects, the next step would be to combine all of those into a simple room and you've got your first simple interior scene!
I have watched many tutorials on YouTube that helped me out A LOT. I can name such channels as eyedesyn, greyscalegorilla, Arthur Whitehead (a great channel to learn about modeling and texturing for realism), Nikolaus Schatz (an in-depth look at modeling and texturing for single objects), and many more; I still watch them from time to time to gather some tips! Sometimes, I also watched tutorials for specific things such as how to get the water or fog right and then put my own spin on it after learning the basics.
Another thing that helped me get better at photorealism was to ask non-professional people about your test renders or previews. Such folk know exactly what's wrong with the image and can spot things easily at times when a professional cant. Gather some ideas from them, put them together to see what works, and then follow your intuition and creativity on how to execute your ideas.
One more thing I'd like to add is, I always go for more casual shots for any photorealistic scene. I've seen people create more photographic renders of the room with perfectly placed objects, clean tables, a very straight and aligned camera view. It's all good for a nice look but for photorealism, the messier and more casual the render is, the more realistic it looks. I am talking about some simple things such as a crooked (but not too much) camera angle with some lens flares as if the photo is taken with a phone, a messy table or floor with dirt or small objects misplaced. Even something as small as a paper clip or keys. Every single object adds to the realism. Cloth simulations can look as if someone just threw the sheet of cloth casually, you can add clothes that fell down or were laid on the bed or couch as if someone was going out. Make it messy but not too much.