Hi everyone,
I’m completely new to the AI world, so please bear with me if I use the wrong terms! I’m a commercial product photographer by trade, and after 35 years in the studio, I’ve realized I’m sitting on a pretty large library of images that might be useful for what you all do here.
I’ve seen people talking about “datasets” and “LoRAs” (I think that’s the term for teaching an AI a specific style/look?), and I’m wondering if my archive is enough to build something meaningful.
What I have:
-
25,000+ unique scenes: Mostly high-end products like jewelry, watches, fragrances, and fashion accessories.
-
The Tech: A lot of this was shot on old Phase One H20, H25, and P45 digital backs. I’m told the “CCD” sensors in those cameras see light and color differently than modern ones.
-
The Files: These are a combination of 16-bit RAW files (IIQ/TIFF) and layered PSD files. They have a considerably large dynamic range.
-
The Masks: For almost every shot, I have the original layered PSD. These have hand-drawn masks that separate the product, the shadows, and the highlights. My thinking was that this information would be useful as a guide to train the original “how” if that makes any sense.
I also have “brackets” (multiple exposures of the same shot) and slight angle shifts for most of these.
My questions for the everyone:
-
Is 25,000 images “big enough” to actually teach an AI how to render things like gold or diamonds correctly?
-
Does the fact that I have manual masks and 16-bit files actually help, or is that overkill?
-
I’ve heard people say AI images aren’t quite right because they are training on other AI images—do these older files act as a 'clean’ baseline or beginning?
I’m really just curious if this is a resource that could help the community or if I’m overthinking the value of old files. Thanks for any guidance!