[masona]

The article covers a lot of territory but I can say from experience that the limitations to current CGI techniques mean that automation is still a ways away.

I produced all the shots for a global diaper company that is on shelf today. The pack designs were a tough format: images needed to be extremely horizontal on one side, and vertical on the reverse. But you can't just ask the photographer to shoot wide and crop in, because the new HD Flexo printing is way more hi-res. The approach at the time was to take the baby imagery and 'paint in' the extra background in Photoshop, a very time-consuming process that had to be repeated for every single image for every single region, with wildly inconsistent results. We created a replica of the on-set nursery in CGI, all the way down to matching the lighting in the C4D studio.

Yes it made the process way more flexible, yes we could localize everything with a new render, and yes we could add props or change the decor. But every render required a human eye to match the camera angles/scale of the baby shots. The uncanny valley is real, even with all the little tricks you can do to make it seem more photorealistic.

I do like the idea of AI-generated baby faces - the casting process / ethnicity requirements / rights management challenges are real. And maybe it wouldn't be all that bad if everyone knew that the babies weren't human. But the cost to develop and manage that system require real experts, a huge expense for a company that only needs baby shoots every so often. I imagine a version of 'thisbabydoesntexist' and how that would even slot into the content production workflow - it feels impossible. It's much cheaper to outsource this stuff to production companies that are smart about how they capture all the different kinds of content.

There are definitely companies where CGI makes sense to develop in-house. IKEA's approach comes to mind since they have super modular and global approach to furniture. But most companies have such a big product turnover that photography is still cheaper. Don't even get me started on hard-to-render products that require serious expertise to visualize. We tried to develop a CGI diaper but wow it was insanely complex.

I'm optimistic about the future of the space since there is so much design territory to explore in terms of workflow improvements. But I still love something real - my wife is shooting a story for a national magazine today on a local woodworking artist and I know it could never in a million years be automated. The more that computational photography advances, the more important meatspace photography will become.

[fredophile]

I'm curious why you didn't shoot the surroundings and stitch it together into a panorama. Since you're in a studio you control the lighting. It shouldn't be hard to shoot the pictures for the panorama once and then spend most of your time shooting the baby. Any of the baby images should be able to be stitched into the rest to get a wider or taller image.

[masona]

That would definitely make sense if all the shots were from the same angle.

We designed/built a nursery that could accommodate shooting from all angles so that all the different shots would look like they could be in different places. That was important because there were infants (on the changing table) to crawlers (on a blanket) all the way up to toddlers (bumbling around).

[justjash]

This seems like a very difficult way to do things unless you really needed to mess with items in the background.

I would think there is still plenty of room to crop a single image down from something like a new Canon 1D. Even just making a composite image if really needed sounds much easier.

[masona]

That's what the photographer was shooting and it is definitely not big enough to pull back and crop in later.

The only way to get files big enough is to shoot medium format, but you can't because the files are too large for rapid-fire.