I have a folder on my desktop called “failures.” I open it once a month, scroll through old composites that never worked, and try to figure out why. The honest answer, in probably half of those files, is that I picked the wrong stock images before I ever made a single adjustment. Bad perspective, mismatched lens focal lengths, conflicting light color temperatures. The composite was doomed at the selection stage. I just didn’t know it yet.

After fifteen years of building images for movie posters, book covers, and album art, I’ve come to believe that stock selection is the highest-leverage skill in compositing. Better than masking. Better than color grading. If your source images aren’t compatible at a physical and optical level, no amount of Photoshop wizardry will save you.

Focal Length Is the Silent Killer

Most artists think about lighting when they’re hunting for stock. Lighting matters, but focal length mismatch is what actually makes composites read as fake to viewers who can’t articulate why.

Every lens compresses or expands the apparent distance between the subject and the background differently. A portrait shot at 85mm has a very different relationship between subject depth and background than one shot at 35mm. When you take a subject photographed at 85mm and drop them into an environment shot at 24mm, the background feels unnaturally large, or the subject looks like a cardboard cutout floating in space.

When I’m browsing Adobe Stock or Shutterstock, I always check the EXIF data before purchasing. On Adobe Stock, you can see this in the file details panel once you download a comp. On Shutterstock, you sometimes have to infer it from the visual distortion in the image. I’m looking for focal lengths within about 15mm of each other for anything closer than mid-distance compositing. For full-scene environmental work, I try to stay within 10mm. It sounds finicky. It is finicky. It’s also the reason the work holds together.

Color Temperature Is Not Just About the Grading Panel

Here’s what I see constantly in compositing tutorials: the instructor selects obviously mismatched stock images and then “fixes” them with a Color Balance layer and a Hue/Saturation adjustment. That approach is fighting the image rather than working with it.

The problem isn’t hue, it’s physics. A subject photographed under overcast daylight at roughly 6500K has soft, diffuse shadows and relatively even highlight rolloff. Drop that subject into a golden-hour background sitting around 3200K and no amount of orange grading will reconcile the shadow direction or the quality of light hitting the skin. The specular highlights will be in the wrong place. The shadow edge transitions won’t match.

What I look for before buying: I want color temperatures within 1000K of each other for close-up character work. Wider environmental work gives you maybe 1500K of tolerance if the subject isn’t front-lit. I use the eyedropper in Camera Raw to sample the highlight of the main light source in both images. If one reads as 5800K and the other as 4200K, I pass on the combination. Finding compatible images at the search stage saves me two to three hours of remedial work per project.

The Resolution Math People Get Wrong

For print work, which is the majority of what I do, I need my final deliverable at 300 DPI at the intended print dimensions. A movie poster at 27 by 40 inches at 300 DPI is 8100 by 12000 pixels. That means I need my hero subject stock image to be at least that resolution, or I need to plan the composition so the subject never exceeds roughly 80 percent of the frame height.

Adobe Stock’s Extended License images often top out at around 5000 pixels on the long edge. That gives me a subject at roughly 17 inches tall at 300 DPI. For a full-bleed poster, that subject is going to be upscaled significantly. I use Topaz Gigapixel AI to handle upscaling before compositing, usually running the image through at 2x if I need to reach poster dimensions. Processing time on my M2 Mac Studio runs about four minutes for a 50MB file. I factor that into my project timeline.

For digital-only work, this math relaxes considerably. Web and social media comps can work fine with stock at 3000 pixels on the long edge, which opens up a much wider library of affordable options.

When the Best Image Is the Wrong Image

I spent six months studying how light behaves on water for an album cover project. I mean actually studying it: photographs, physics papers, going down to the East River in the morning and taking my own reference shots. When it came time to choose the background plate, I found an absolutely beautiful image of a lake at dusk, the kind of shot that makes you stop scrolling. The composition was perfect for the brief.

I didn’t buy it. The light was coming from camera-left at a low angle, and my subject image had overhead light with no directional quality at all. Making those two images coexist convincingly would have meant rebuilding the lighting on the subject from scratch using dodge and burn and painted light layers, probably 12 to 15 hours of work on top of the standard composite build. The less beautiful image with compatible overhead lighting got the job done in a third of the time and looked more cohesive in the final print.

The right stock image is not the most beautiful stock image. It’s the one that is physically compatible with the rest of your scene.

Before I open a browser to search for stock, I sketch the composite on paper. Not a masterpiece, just enough to note the light source direction, the approximate camera angle, the rough focal distance of the main elements. That sketch becomes my filter. I’m not browsing for inspiration; I’m searching for very specific physical parameters. It sounds like it would slow the search down, but it cuts my stock selection time roughly in half because I stop buying images I want to use and start buying images I can use.

The single most important thing I can tell you about stock selection is this: your composite lives or dies based on decisions you make before the software is open, and every hour you spend choosing compatible source material returns three hours on the back end.