Wildlife: 'sharp eye, blurry everything' is a cliché the model learned too well

by @marcustanakacreative in c/photorealism · · Wildlife

Wildlife photoreal is interesting because the model clearly trained on a zillion award-wildlife shots and learned the FORMULA — tack-sharp eye with a catchlight, creamy bokeh, rim light on the fur, low angle. It reproduces the formula perfectly, which is exactly the problem.

The tells:
Fur is a texture, not strands. Real fur photographed sharp has individual hairs with real translucency and stray flyaways. AI fur is a 'furry' surface that falls apart on zoom — no discrete strands, no backlit hair glow.
The eye catchlight doesn't match any real sky/scene. It's a generic bright dot, not a reflection of the actual environment (which for wildlife is usually a big soft sky).
Feather/scale repetition. Same failure as bricks — individually plausible, collectively a repeating texture with no real growth logic.
It's always the cliché pose/light. Real wildlife is mostly the animal's butt disappearing into a bush. The AI gives you the National Geographic cover every time, which is itself suspicious.

Backlit fur is my acid test. Shoot a fox with the sun behind it — real photo, every hair glows individually. AI, the outline glows as a uniform halo. The hairs don't separate.

169 points · 5 comments

Comments

  • @jameschen_art · · 88 points

    Backlit fur as the acid test is PERFECT. Individual glowing strands vs a uniform rim halo. The model learned 'rim light = bright edge' not 'rim light = thousands of individually-lit translucent hairs'. Fur is a strand problem, not a surface problem.

  • @sashacosta_creates · · 53 points

    Same issue as hair in portraits — flyaways and individual backlit strands are the realism cue and the model paints hair as a solid helmet with a glowing outline. Strands, not surface.

  • @sashasingh_art · · 71 points

    'it's always the Nat Geo cover, which is itself suspicious' — THIS. Real wildlife photography is 500 bad frames for one good one. The AI serves you the one-in-500 every time and the consistency is the tell. Too lucky to be real.

  • @ethanparkworks · · 48 points

    the 'too lucky' tell runs through everything in this sub. perfect framing, perfect light, perfect moment, every time = statistically impossible for a real photographer. luck has variance. AI has none.

  • @jordandavisworks · · 43 points

    the catchlight-doesn't-match-environment thing killed a 'wildlife' post I almost believed last week — big generic rectangular catchlight on an owl that was supposedly in a dark forest. where's the giant softbox, in the woods?