FLUX is the Best Free AI Model period.

[@ayandangcobo1755]

RIP Midjourney

[@hotyrona]

RIP SD

[@ayandangcobo1755]

@@hotyrona yeah... unless they come up with something really good- SD3 is done for

[@ragemax8852]

I stopped using Midjourney and SD a while ago, FLUX is pretty much the future.

[@ayandangcobo1755]

@@ragemax8852 yea Flux is the future... I still use all three but i downgraded my Midjourney to the cheapest plan. Because it's still amazing in some aspects...

[@WantMoar]

It’s definitely better and more flexible especially the dev version. However Midjourney will always have a customer base that aren’t technical enough to use a model like flux.

[@AI_Made_Simple.]

Forgot to mention , you will need to update Comfy UI for this to work. * Tip: You can set the weight_dtype above to one of the fp8 types if you have memory issues. Also If you get an error in any of the nodes above make sure the files are in the correct directories.

All model types worked on my RTX 3060 (12GB). Happy times!

[@AI_Made_Simple.]

yay

[@blackg3gaming]

How long does image generation take when you used the full model (the 23gb version)? I have a 12gb vram also and I’m just wondering how slow it’s gonna be 😂

[@AI_Made_Simple.]

@@blackg3gaming FP8 should be fine. ComfyUI supports loading model weights directly in FP8 format (12 gigabytes), and automatically detects available VRAM to swap loading method. (Do you have less than 12 gigs of VRAM? Worry not, Comfy's core model handler will detect this and automatically offload VRAM partially to system RAM, allowing the model to run on smaller GPUs at the cost of slightly slower running times.) Source: blog.comfy.org/august-2024-flux-support-new-frontend-for-loops-and-more/

[@KamauNwabudike]

So there isn't just one thing to download and use? I have to install a gazillion different things?

Its really good at certain things like text, but other things are gone like knowing who Trudeau is (SDXL knows most celebs, but not Flux).

[@xbon1]

It's against license agreement to create and distribute copyright material anyways, so dw about it. What it can do it does better than any other model.

[@aegisgfx]

Who cares, celebs can all EFF off

[@MilesBellas]

LORAs and SD workflows ?

[@ragemax8852]

Doesn't even really need LORAs to be honest.

[@froilen13]

Any compatibility with controlnet?

[@pb2806]

Flux Dev wins when there is text to display, otherwise it's often worse than Dalle3 from an artistic viewpoint. IMO

[@dave47663]

Non-commercial license is a non-starter. This is just another great paper weight.

[@franciscobutte]

Accesible for 1% of the planet though because it's requirements

[@JBereza]

Once it will be 99%... I am looking forward to inherit money from a "Nigerian Prince" ;-)

[@drendelous]

1:46 im not sure about the paws though

[@AI_Made_Simple.]

Yeah paws 🐾 were a bit of hit or miss. Some came out as human fingers

[@YoottanaTiyaworanan]

can I use with my AMD GPU?

[@AI_Made_Simple.]

yeah, Start ComfyUI with the --directml flag

[@Dreamingofyou317]

text good. image 2022. be real man.

[@AkshayKumar-vd5wn]

Bruh the requirements to run it.....

[@context_eidolon_music]

No it isn't, period.

[@AI_Made_Simple.]

🤣

[@DivergentDroid]

AI might be o.k. at creating images and videos but it's still always going to be a dumb programmed mimic.

[@jooptablet1727]

needs controlnet and inpainting. THEN it'll easily be the best.

[@creepybeat]

mhmm not sure, i still think midjourney is crazy good.

[@ayandangcobo1755]

@@creepybeat been testing out and yeah Midjourney is still amazing. Flux is superior- especially when it comes to hands... There are places where Midjourney still reins supreme though.

[@Leamaunier14]

​@@ayandangcobo1755It's still surprising because, having tried both, Midjourney is still superior in terms of quality and realism. The only real point where FLUX is superior is the lack of censorship, which allows for more variety. But in everything else, Midjourney is still better, whether it's in faces, hands, skin, or even animals in general. I'm talking about realism, just to be clear.

[@ayandangcobo1755]

@@Leamaunier14 i have the opposite experience... in my case, Flux is better for realism and Midjourney is amazing for fantasy images... It's good that Midjourney now has competition, it means they have to improve it even more. In the end we are the winners.

[@itsyaboidaniel2919]

Definitely a cool new AI art tool. Not that I'll use it anytime soon, but it's cool to see the progress being made with these models. 1:25 Eh, it looks a lot like a fish, but it has the wrong number of bottom fins, though it is consistent with the other pink fish. Also, sort of minor, but some of the cat whiskers come out of nowhere instead of the normal whisker holes. Sort of like there's a correlation with cat whiskers and the general area, but lacking with the understanding of where they come from.

[@markogrubisic]

Nope...image, especially photo-realistic still not there...maby one day...

[@Dave-rd6sp]

The fine-tunable version of FLUX is non-commercial, making it effectively DOA.

[@EvanLefavor]

They don’t have any way of telling if you used Flux to generate an image.

[@ayandangcobo1755]

@@EvanLefavor this 🤣... how are they going to prove it?

[@Dave-rd6sp]

@@EvanLefavor Non-commercial as in it's not open source and can't be used in other projects like an open source model can, nor will it attract the same level of community engagement and thus tooling.

[@DepthDutchy]

It's good but it's really really slow to run it locally unless you have at least 24gb vram

[@ragemax8852]

I agree it's really slow to run even on my 48gb vram machine, it's definitely pretty great, but I need speed times to improve.

[@VintageForYou]

WOW, it looks amazing.🙋🏼‍♂

[@AI_Made_Simple.]

i thought so too, and its pretty fast :D Great to have so many amazing models that are being released as open source :)

[@flynifty3951]

I love it ❤

[@devnull_]

...non-commercial model though.

[@devnull_]

Sorry if I missed it, but I think you didn't mention this at all.

[@livb4139]

what times are you guys getting? it takeslike forever on my 4090

[@kitsunegamedev4526]

I have exactly this in ComftyUI portable but at the time of gerenation it says "Reconnecting" and the server window doesn't show any errors :S Triple checked everything is in the right folders and as for the UI everything seems in place >.

[@AI_Made_Simple.]

I forgot to mention you may need to update Comfy UI. Hopefully you can get it working 💪

[@ncouvin]

same here with nvidia 3090 :(

[@kitsunegamedev4526]

@@ncouvin Found the solution Its the windows page file. Set virtual memory to at least 32 GBs and it will work.

[@ncouvin]

@@kitsunegamedev4526 no luck here 🙁

[@asaejapan7143]

I was wondering if the not-quite realistic madame tussauds wax museum texture was intentional.

[@mice3d]

@3:21 the hand has an extra nubin :P

[@cardboardgod]

It's completely useless for what I need. It's far from 'the best'.

[@ragemax8852]

It really is the best, but needs some improvement. I'd recommend sticking with SD 1.5 for now, but play around with this one, it's too slow to generate images right now.

[@oggy3D]

can this model work with fooocus ai ?