Right. Very cool experiment. I've been interested in these ping pong dialogues, curious how other people approach them.
@genekogan
-
AI System Transparency: Hidden Instructions and Model Visibility
By
–
Can you clarify what you mean by this? Both AIs see the whole conversation, don't they? I would think only the instructions to each model are hidden from each other in each's system prompt.
-
Optimizing LoRA Training: Rank and Step Count Adjustments
By
–
you can also try adjusting the training settings with a different lora rank or step count. it's a tricky balance depending on the training set.
-
Reducing Concept Strength to Prevent Overfitting
By
–
Lowering concept strength to something like 0.2 when generating tends to reduce overfitting
-
Video Generation Maturity and AI’s Next Frontiers
By
–
Like Imagenet classification, there’ll come a time when video generation is basically solved and you can’t really improve anymore. And then on to the next frontier, whatever that may be…
-
Runway gen3 Overcomes Strong Visual Priors in Generation
By
–
Impressed that Runway gen3 succeeds here in making the banana eat the person! Historically challenging to prompt image generators to reverse strong priors, e.g. "yellow strawberry next to red banana" pic.twitter.com/BSimyUzOwf
— Gene Kogan (@genekogan) 1 juillet 2024Impressed that Runway gen3 succeeds here in making the banana eat the person! Historically challenging to prompt image generators to reverse strong priors, e.g. "yellow strawberry next to red banana"
-
Image-to-Video AI Reconstructs Original Images with Excellent Precision
By
–
The image-to-video is excellent. It's almost a perfect reconstruction of the original images.
— Gene Kogan (@genekogan) 12 juin 2024
Works amazing with @va2rosa Little Martians! pic.twitter.com/mXYNCwhvasThe image-to-video is excellent. It's almost a perfect reconstruction of the original images. Works amazing with @va2rosa Little Martians!
-
Luma Dream Machine Text-to-Video Model Impresses with Quality Output
By
–
Been playing around since yesterday with @LumaLabsAI new text-to-video #LumaDreamMachine model. It's pretty amazing.
— Gene Kogan (@genekogan) 12 juin 2024
Mostly trying out science fiction tropes. Great quality and prompt adherence. It even does scene cuts effectively. pic.twitter.com/c3K5rBCeTlBeen playing around since yesterday with @LumaLabsAI new text-to-video #LumaDreamMachine model. It's pretty amazing. Mostly trying out science fiction tropes. Great quality and prompt adherence. It even does scene cuts effectively.
-
Teaching GANs: From Evangelist to ML4A Practice
By
–
For the next two years, I became something of an evangelist for GANs, incorporating them into my teaching practice called @ml4a_ and showing people how to train their own GANs using Jupyter notebooks. I wrote about that experience here: https://
medium.com/@genekogan/mac
hine-relearning-8cd57ba242
… Until NVIDIA -
A Book from the Sky: AI Art Homage to Xu Bing
By
–
Francis also introduced me to the original "A Book from the Sky" by Chinese artist Xu Bing, who composed thousands of fictitious glyphs that resembled Chinese characters. Inspired, I named my project A Book from the Sky as an homage.