Ai getting better and better

[Kwon]

Soon no actors needed

[IroNat]

Awesome.

[obsidian]

Soon no actors needed

Kwon, check out the parts where the AI actors are standing in the rain — there are no raindrops bouncing off their bodies. It looks more like a rain filter or video layer composited over the footage in DaVinci or Premiere, rather than real interaction. I’m still waiting for AI to truly match real-world physics. The lighting, shadows, and reflections all look generic, based on whatever the model was trained on. I wonder if AI models will eventually be able to factor in ray tracing to make it look physically accurate.

[obsidian]

Soon no actors needed

This was done in 1993 and better than that AI Slop.

[obsidian]

[Kwon]

Kwon, check out the parts where the AI actors are standing in the rain — there are no raindrops bouncing off their bodies. It looks more like a rain filter or video layer composited over the footage in DaVinci or Premiere, rather than real interaction. I’m still waiting for AI to truly match real-world physics. The lighting, shadows, and reflections all look generic, based on whatever the model was trained on. I wonder if AI models will eventually be able to factor in ray tracing to make it look physically accurate.

Give it 4-6 months Obsi

[obsidian]

[Kwon]

Compare with ai of one year ago , "Will smith eating Spaghetti".

[obsidian]

Give it 4-6 months Obsi

I just don't know if I can ever trust the shadows, reflections and general lighting.

How would you do this with AI?

[Rambone]

Soon no actors needed

Weird Al did it better. Built different.

[obsidian]

Here's a few images. I used this 360 degree HDRI for the scene:

https://polyhaven.com/a/derelict_airfield_02

I created several cubes of different sizes — 72", 96", and 24", based on my preference. The spheres have specific dimensions as well. Then I added the road bike to the scene. The torus partially sinks into the ground. Two of the cubes use mirror materials, the gold sphere has a blurry reflection, and the chrome sphere a perfect mirror reflection. The 3D “KWON” lettering sits on one of the cubes.

In the animation, the camera starts with a 15 mm wide-angle lens, then zooms in to 30 mm while I slightly rotate the view.

There’s no way current AI generators could create a scene like this — with accurate double reflections, realistic shadows, and consistent 3D spatial relationships. I’m sure they’ll get there eventually, but it’s hard to say when.









The image below has DOF (Depth of Field) enabled — the handlebars are in sharp focus while the background is softly blurred. This is the level of precision I want when creating 3D imagery. I can control the exact focal distance of the DOF to achieve the look I want.



In the image below, Depth of Field (DOF) is disabled, resulting in a fully sharp image where all elements remain in focus.

[Kwon]

Nice bike Obsidian!

[falco]

There were no humvee's in 1975.

[FitnessFrenzy]

Very impressive. It has come far.

[Humble Narcissist]

In 1 or 2 years AI will look more real than real.

[obsidian]

Nice bike Obsidian!

Haha thanks. 

[obsidian]

AI and Raytracing needs to merge. The end goal. Faster renderings for 3D content creators, and physically accurate reflections and shadows for AI content creators.

[obsidian]

[beakdoctor]

Weird Al did it better. Built different.

Ha! UHF was the first thing I thought of when I saw the op post.

[ProudVirgin69]

AI can’t render video longer than 4seconds without artifacts/hallucinations reaching unacceptable levels.  Hardware demands increase quadratically as video length increases, IIRC.  I don’t think AI will be replacing Hollywood anytime soon.

[Humble Narcissist]

AI can’t render video longer than 4seconds without artifacts/hallucinations reaching unacceptable levels.  Hardware demands increase quadratically as video length increases, IIRC.  I don’t think AI will be replacing Hollywood anytime soon.

Is one year soon?

[NaturalWonder83]

Soon no actors needed

TERMINATOR 2 WAS TRYING TO WARN US!!!! SKYNET IS BASED ON AI AND TESLA!!!!!!

[Kwon]

Ai Conan

[obsidian]

I asked Grok and ChatGPT to create:

Can you create a zoomed out wide angle landscape 16:9 image of a Ferrari on a sunny beach, with a 100% reflective cube behind it at a 45 degree angle. There should also be a reflective sphere behind the car. Space the cube and sphere away from each other. A man should stand between the sphere and the cube. Breaking waves should be in the distance. Shadows and reflections should be accurate.

The reflections are jacked up. It will be a while before these become accurate like 3D raytracing engines.

Grok was faster than ChatGPT, but ChatGPT gave more options to tweak the setting before generating it.

ChatGPT Below:



Grok created these two below:

[obsidian]

Here’s a 3D scene rendered with ray tracing. Notice the lighting - the bike’s shadows falling on the first car, and the first car’s shadows cast onto the second. Also take a look at the reflections in the spheres. These are ray-traced reflections and shadows — something AI-generated “reflections” can only fake for now and can’t accurately reproduce anytime soon.

For these renderings, I used an infinite sphere without a ground texture. Because of that, the tire tracks shift between views. It’s possible to render with a finite sphere and a defined ground plane instead, which can then be scaled so the footprints appear correctly sized and remain consistent across views. However, that approach introduces visible distortion where the finite sphere meets the ground plane. To avoid that issue, I rendered this scene using just the infinite sphere.