Opensource, Uncensored, Unbothered. - Flux.1 Image Gen

YESSIR! We called it. Video is coming. And by this time next year, all of this will run on 4 year old laptops. THERE IS NO MOAT and consumers have no loyalty to the giants. Gonna be a wild decade :)

The flux Dev version allows for commercial usage of images made by flux dev, what it doesn't allow is selling Flux dev or any finetunes of it as a service like a image generation service

7:30 "we have Kirby at home" 💀

As far as I know only the Flux "schnell" and "dev" model are open source, but not the best "pro" version. But still probably the best open source models we currently have.

12:51 "let's get friogay" 💀

That's not luigi, thats green mario

Was really excited about this video, but you are using Flux Pro which is in fact not open source.

Flux is awesome. No disputing that. But there are some caveats. While it may not be (strictly speaking) censored, my understanding is that it isn’t trained on anything nsfw and so it’s naturally not really capable of such generations out-of-the-box. It’s also not going to be an easy model to work with if you want to fine-tune it.

I've been here ever since your dall-e 3 prompt testing videos. I like how your videos keep progressively getting higher in quality! Keep it up!

Gemma 2 2b is the biggest development in the text gen space IMO. It finally brings a good fast chatbot that you can run just about anywhere.

Amazing model. Did my standard "an old man sitting on a tugboat eating a foot long". Looked incredible.

Hi. Just checked a couple of guys from the website - they are former employees of Stability AI

12:51 Alex Jones *Heavy Breathing*

Thanks for mentioning us! Just gave you the PRO plan if you want to try us out

I got Flux dev working on my home computer late in the day when it was released. The image quality is great so far and I was shocked at how easy it was to make a tiled upscaling workflow -- it just works with barely any tile inconsistency, even without any controlnets. The downside is it is slow -- on a computer where I could generate a 4k+ final image with SDXL in about 12 minutes, Flux takes about an hour. I haven't had a chance to really try to optimize the step number, though, so it's possible I could get it better. It also doesn't really use CFG but its own custom guidance scheme, which means that it does not support negative prompts. Depending on what you need to do, this could make life difficult for you (e.g. the "fried rice with no peas" quest). And it's not really something I care about, but I've seen people complaining about not being able to do styles well with Flux. It is definitely the case that Flux can kind of do its own style choices unprompted, such as a photographic image with a cartoony element within it.

It's getting pretty good. I look forward to running some static generations through Kling on my animations channel.

You could also mention how diverse the output is for the same prompt! Many other models look some over fine-tuned, outputting almost the same image for a constant prompt again and again.

6:05 Obviously Kirby absorbed a hostile leader and became a cartoon caricature of a president.

Midjourney wins in my book. midjourney’s getting a 6.2 update next month too 😏

I've been using flux.1 on my 24GB GPU for 3 days already 😁