Flux is a free, open-weight text-to-image model that matches DALL-E 3 and Midjourney in quality. It generates photorealistic humans and renders text in images better than competitors, and you can run it locally or online without paying.
Oh my. What do we have here? This is a new text to image AI system that is absolutely incredible. Perhaps the first one that seems as good as Dall E3 and Midjourney. And I can't believe I'm saying this, but perhaps even better, I'll tell you why. And it has something going for it that those systems don't have. And that is being free. Yes, this quality completely free of charge. And as of the making of this video, you fellow scholars can try it right now. Dear fellow scholars, this is two minute papers with Dr. It is called Flux and can generate incredible images in a variety of topics. It is also great at generating photorealistic humans. And of course I made some images of scholars holding onto their papers. I thought you'd never ask. I am pretty happy with these. Real good. And wait, what is going on here? Are you seeing what I am seeing? That is text after the age of mangled hands created with AI image generators. That one is mostly solved, but generating text is still the bane of these techniques. Is it going better here? Is it finally solved? Well, these are real good. Remember this is an open weight model that you can run for free. And even the paid midjourney cannot do so well with text. However, the text here is not an integral part of the image. In many cases it seems that it just added to the image with which we could do by hand easily too. So now hold on to your papers fellow scholars, and let's try a much harder version of that. Ooo, these are crazy good. Wow. Now I hear you asking OK Caroj, but how much cherry picking does this require? How many images do you need to throw out until you get these good ones? And the answer is zero. Yes, these just came out as you see them. If you try this with the paid mid journey, I asked that the foam forms the words four shots and not a chance. And with Flux, four out of four. Wow. And we haven't even talked about how this will supercharge other techniques that can take a still image and make a video out of it. The whole model is available for all of us free of charge. You can try it on the web through the link in the video description or run it yourself at home. Some of our fellow scholars are already doing that and I get chills when I think that this might run on your phone in your pocket. Just two more papers down the line. What a time to be alive. And what do you think? What would you fellow scholars use this for? Let me know in the comments below. So now let the experiments begin. We need new tools for the era of LLMs and weights and biases. Now has Weave a lightweight toolkit to confidently iterate on LLM applications, use traces to debug how data flows through each step of your app, and use evaluations to measure your progress. It is the best. Try it out now at WNB Me Papers LLM or click the link in the description below.
