Painting over photographs using 3D models

I’ve been working on a series of photo illustrations using my not-so-good photos from Mono Lake tufas. They are the photos that aren’t strong enough for a portfolio, but have something about the composition that I wanted to rescue, and that had a sense of scale and form that reminds me of ancient ruins.

I’m not done with the series yet, but I want to share my process. I use a mix of photo collage (replacing skies with other photos I’ve taken, for example), painting by hand, and 3D renders. Here’s how I do it:

1. I start with a photo that has potential, but is not quite there for me. In this case, I liked the shapes, they reminded me of obelisks, so I went with an Assyrian theme. I also thought that snow for a desert theme would be an interesting contrast. I develop the image in Lightroom as usual, just minimal basic touches.

2. In Maya, I place and scale cubes in a general idea of what the scene looks like, making sure I’m using depth properly too. I make the cubes semi-transparent.

3. I match my camera focal length in Maya to my original photo. Then I tumble the camera around until the angle roughly matches my photo. The vertical tilt is the most important, matching the convergence to the up axis really makes a difference.

4. I add an image plane to the camera with the original photo, set to 50% opacity, so that I can line up all the blocks very clearly. Once The alignment is close enough, I make sure to lock the camera so that I won’t touch it ever again by mistake.

5. I download a lot of free 3D scans from sites such as myminifactory.com, thingiverse.com and cgtrader.com. I download a ton, everything I can because I’m not sure what I will want to use or not yet, but I make sure they match the theme and time period I’m looking for.

6. I import these (usually STLs) into Maya and do some cleanup. Normal process is: rotate/scale, merge vertices, mesh cleanup, reduce (they are way too dense usually), delete parts I don’t care about, and apply a 3D noise material that looks like rock.

7. The fun begins! I love finding shapes that remind me of other shapes, and puzzling everything together. I try to think about the space, what it would’ve been, what I want to feature and how I can make it fit in the context of the original photo. Sometimes I need something to balance the composition, such as making the pillar on the right side much larger, so I scale them but try not to overdo it. For this series I want to keep the original compositions, otherwise I’d go nuts and add a lot more elements.

8. I add some lights that roughly match the original photo, mostly directional lights, and some point lights with which I take some liberties to highlight certain areas that otherwise would lack contrast. I do a quick hardware render at full size (6400x4267) with antialias, ambient occlusion, and alpha channel. I found that a fancy render is not needed, as these will not need much detail later, they will pick up the detail from the underlying layer of rocks and snow. Then I bring the renders into Photoshop.

9. I position the renders in Photoshop carefully, and I rough them up a bit (a bit of blur, break some chunks off, etc). Then I make a selection from that render layer (Cmnd+Click on layer thumbnail) and I start using the clone stamp tool over the original photo, within the new selection, to add the chunks of rock that have changed the silhouette. Then I invert selection and do the same when I want to paint sky in the opposite side. I also clean up weird holes and shapes that are distracting here. I removed all the plants from this one since they gave away the real scale of the area, and I wanted it to feel much bigger.

10. I change the blending mode of the renders to Hard Light, and use Levels to control how bright and dark things are. By reducing contrast, I reduce the effect. I prefer using levels to opacity since I can control lights and darks and midtones separately. I also mask parts out, blur some things, use dodge/burn, whatever is needed to integrate the parts better. I tend to do these in parts, only one major structure at a time so I can focus on it, then later merge them all back into one layer when done.

11. Since I have all these layers split already, I can easily make a selection from them and I use that to paint a black and white depth map. I use that as a mask for a layer with the sky color, making the distant object recede a lot more. Haze is one of the main tools I have to cheat the scale, so I abuse it.

12. I add a lot of light effects, glows and highlights (usually layers in Linear Dodge/Add mode), sometimes I add cast shadows (layers in Multiply mode), and all sorts of noisy effects like snow and dust that hide my mistakes. To get the color, I usually create a LUT (Layers > New Adjustment Layer > Color Lookup), try out all the presets, and then try changing its opacity and blending mode from Normal to Overlay or anything else, just to see the effect. Once I find the mood, I add more LUTs, mix them up, find a combination that pushes what I like, and then tone it down by painting a mask for the LUTs with soft brushes. I also add Levels and Curves layers to increase contrast where I want.

And that’s it! It’s a complicated process, but I enjoy mixing 2D painting, 3D, and photography all in one. I’ll keep working on a few more images for this series.