anonylizard

23 ms·

301.
by anonylizard 4y ago
The best system we can currently observe, is in ML. Look at stable diffusion as an example. Incredible papers, such as dreambooth, LORA, controlnet, are: 1. Published on arxiv before peer review 2. Productionised within 2 weeks of paper rel
302.
by anonylizard 4y ago
You guys need to understand, that the skill ceiling for AI art is very high. Its not some iphone-selfie tier technology. Go to Civitai.com and huggingface, to start comprehending how many AI models there are, and how powerful the latest mod
303.
by anonylizard 4y ago
I think pay-for-verification is going to be dominant on all real-identity social media. Reason is simple: AI 1.You can now trivially create an avatar photo, of any level of attractiveness, of any race, of any age. You can even reproduce the
304.
by anonylizard 4y ago
Is there a GPT-3 disruptor? All the open sourced models are GPT2 improvements, and GPT2 was open sourced by OpenAI. GPT3/4 is simply too expensive for consumer GPUs, any open sourced versions will have to run on A100s in the cloud, so
305.
by anonylizard 4y ago
"There's no money in comics, it's like saying "be a rock star"." There's no money in western comics. Successful manga artists regularly earn millions. Simply put, western comics got left far behind in stor
306.
by anonylizard 4y ago
Yes, far far faster, in full colour too. Traditionally, manga artists took photos and applied filters on them, to serve as the background for the manga panels. Its just that time consuming to draw things in detail, that they didn't bot
307.
by anonylizard 4y ago
I've tried to convince artists to transition to the AI-augmented future, since October. There's almost no successes. I've seen artist communities on multiple websites, in multiple languages. At first they tried to laugh at th
308.
by anonylizard 4y ago
Draw comics, or really manga in particular. The only real challenge in AI art, that won't be solved any time soon, is storytelling: chaining multiple panels together for coherent stories. Even ChatGPT cannot tell a coherent story longe
309.
by anonylizard 4y ago
This entire article misses critical developments in the AI art space. Controlnet was just released last week. It allows for precise fine-tuning of the image with skeletal positions, depth maps, outline sketches, etc. It has been exceptional
310.
by anonylizard 4y ago
In case you don't know, AI art is not an AI pumping out art by itself. A human prompter has to prompt the AI. The prompter can perfectly understand 'police brutality'/'shelled hometown', as they are human, just
311.
by anonylizard 4y ago
Because COBOL is not difficult to learn, in fact it was designed so that 'businesspeople' could easily learn it (Which didn't happen). COBOL is merely undesirable from a long term career perspective. The aggregate demand is
312.
EDiffi, Nvidia's new State-of-the-Art Diffusion model(deepimagination.cc)
5 points by anonylizard 4y ago|3 comments
313.
by anonylizard 4y ago
Highlights 1.Beats Imagen in quality 2.Incredible img+text2img capabilities
314.
by anonylizard 4y ago
Not exactly, tracing over the 3d is called 'rotoscoping', and has been used in animation for a few years now, basically hand-drawing animation over a 3d video. There are two problems with rotoscoping. 1. The cost, its not much che
315.
by anonylizard 4y ago
https://www.bilibili.com/video/BV1B14y1575G/ It uses diffusion to post-process a 3d video frame-by-frame. You can see the massive improvements in facial expressions, from bug-eye alien face to something quite plea
316.
by anonylizard 6y ago
From basic evolutionary biology, there is no universall "fitness", only situational fitness. In the middle ages, in a world without contraception, with extreme pressures to marriage, without pornography or any other non-intercours