(👁 ͜ʖ👁)

ᕕ( ᐛ )ᕗ

[(👁 ͜ʖ👁) @Uconeni]

SlangPy is seriously cool — it lets you run gradient descent inside your shader.

Texture reconstruction, parameter fitting - all with auto-diff in shaders, Python and PyTorch.

https://t.co/PZWxbHXh5Z

#ShaderDev #PyTorch #SlangPy #MachineLearning #ProceduralArt

[(👁 ͜ʖ👁) @Uconeni]

Exploring 🔥 #SlangPy + #PyTorch for auto-tuning shader parameters via gradient descent!
I wrote a simple brightness shader in Slang and learned the RGB offset to match any reference image. The code: 👉
https://t.co/zd0zCBMOIs
Example is simple, but the potential is great!

[Aleksa Gordić (水平问���) @gordic_aleksa]

Humans have failed the AI art Turing test?

Main takeaways from this blog post: https://t.co/i65dJkZ1yj

The test was taken by 11,000 people and they had to classify 50 pictures as either human (some of the greatest human artists) or AI-generated (cherry-picked as well by the author of the test).

Takeaways:

1. Most People Had A Hard Time Identifying AI Art - random guess would yield 50% accuracy, both mean & median were ~60%!

2: Most People Couldn’t Help Judging Art By Its Style - predictions were skewed towards AI if the image was digital and towards humans if image was an oil painting - humans are heavily biased and use style as a superficial shortcut/heuristic to come up with a conclusion.

3: Most People Slightly Preferred AI Art To Human Art - 2 most liked pictures were both by AI, and 60% of the top ten are also by AI!

4: Even Many People Who Thought They Hated AI Art Preferred It - "The 1278 people who said they utterly loathed AI art (score of 1 on a 1-5 Likert scale) still preferred AI paintings to humans when they didn't know which were which (the # 1 and # 2 paintings most often selected as their favorite were still AI, as were 50% of their top ten)." <- lol!

5: But Others Might Genuinely Be On A Higher Plane Than The Rest Of Us - some people have an amazing attention to detail, and professional artists definitely score significantly better than general population.

Here are accuracies by groups:
* avg participant - 60%
* hates AI - 64%
* professional - 66%
* both professional & hates AI - 68%

So all of these groups still struggle big time to distinguish 2024 AI art from some of the greatest human artists. :) What will be the result in 2030?

The highest score was 49/50 which 5 out of 11,000 people achieved, truly impressive and would be cool to interview these people!

Did they spend an ungodly amount of time analyzing the pictures, did they follow all the rules (don't zoom in, etc.), did they use some AI classifiers to aid their decision making process? I don't know the answer to that, but would be extremely valuable to find out.

I'll finish with this paragraph from the article:
"But it also seems very human to venerate sophisticated prestigious people, and to pooh-pooh anything that feels too new or low-status or too easy for ordinary people to access - without either impulse connecting with the actual content of the painting in front of you."

What a time to be alive, eh?

[(👁 ͜ʖ👁) @Uconeni]

Tried to train GPT-2 to win simple Connect Four game:

https://t.co/rDUvFf8o4H

[(👁 ͜ʖ👁) @Uconeni]

@xorqs @cubiq How did you make the deal and how they broke it? I thought, that they just do not allow the commercial use.
By the way, @cubiq - Great job, thanks for the update!

[(👁 ͜ʖ👁) @Uconeni]

🚀 Just did a side-by-side of https://t.co/xVE5FbymMF’s img-to-3D models, pre and post-update: ✅ Noticeable enhancement in initial image projection quality—though it’s not always a perfect fit. 🔄 No significant changes on the flip side; some aspects even look worse. #3D

[(👁 ͜ʖ👁) @Uconeni]

AnimateDiff FTW

[(👁 ͜ʖ👁) @Uconeni]

I experimented with generation of 3D variations of a 2D reference with control over features
Also, started testing SV3D

https://t.co/axYEOF66kV

#SV3D #stabilityai #GenerativeAI #3D #MachineLearning #3dmodeling #StableZero123

[(👁 ͜ʖ👁) @Uconeni]

Tripo AI and Stability AI released a demo of their model, so I did a quick comparison to my recent tests:
https://t.co/ZUlmQ0aEWP

Pros: It is extremely fast! It also works with shapes that made Stable Zero 123 crash
Cons: Textures are low res. Some objects are deformed

[(👁 ͜ʖ👁) @Uconeni]

HuggingFace space with Stable-Zero123
https://t.co/tKXHssbpSY

#stabilityai #GenerativeAI #3D #MachineLearning #3dmodeling #StableZero123

[(👁 ͜ʖ👁) @Uconeni]

I wrote a post on How to prepare assets for 3D object generation with Stable123, solve issues, and debug problems:
https://t.co/Kehlhfw6D0

#stabilityai #GenerativeAI #3D #MachineLearning #3dmodeling

[Stability AI @StabilityAI]

Announcing Stable Diffusion 3, our most capable text-to-image model, utilizing a diffusion transformer architecture for greatly improved performance in multi-subject prompts, image quality, and spelling abilities.

Today, we are opening the waitlist for early preview. This phase is crucial for gathering insights to improve its performance and safety ahead of open release.

You can sign up to join the waitlist and learn more here: https://t.co/lX8z1UTc8s #stablediffusion3

Prompt: Epic anime artwork of a wizard atop a mountain at night casting a cosmic spell into the dark sky that says "Stable Diffusion 3" made out of colorful energy

[(👁 ͜ʖ👁) @Uconeni]

Checking results of text → Concept art → 3D assets pipeline with SDXL and Stable zero123. In some cases, it works very well! In others, it totally fails...
More: https://t.co/Tu0nF2nomu

#stabilityai #GenerativeAI #3D #MachineLearning #3dmodeling

[(👁 ͜ʖ👁) @Uconeni]

Checking out mesh generation with threestudio and stable-zero123
https://t.co/BR5qjJ8Yhc

[(👁 ͜ʖ👁) @Uconeni]

I'm still checking out Stable-zero123 3d generation - it is awesome!
https://t.co/haAuD4PeZN
#stabilityai #GenerativeAI #3D #MachineLearning #3dmodeling

[(👁 ͜ʖ👁) @Uconeni]

Time to take a look at some Image-To-3D methods: https://t.co/KCZvcUEK2v

[(👁 ͜ʖ👁) @Uconeni]

Checking out some of text-to-3D methods:
https://t.co/Qz7kpQ3qWD

[(👁 ͜ʖ👁) @Uconeni]

Translated part of my thesis on water surface simulation in UE4 with Neural Networks:
https://t.co/j1rXVDimHu

The complete thesis is available as well - in Polish:
https://t.co/abYnyTj0ol

#MachineLearning #mlmodel #AIArtworks #unreal #UnrealEngine #Simulation

[(👁 ͜ʖ👁) @Uconeni]

GSGEN: Boeing 737 plane after 140000 iterations, ~1.5h