AI Art Generation |OT| Midjourney and beyond

Thread: AI Art Generation |OT| Midjourney and beyond
Official Thread
Reader mode
FLUX NF4 Install
Animated GIF


For those who aren't keen on the Spaghetti nature of ComfyUI and might lack the Stiff GPU requirements to run the FLUX full Dev model, a new solution has come to the fore over the weekend.

lllyasviel who developed Web-Forge (A1111 variant) carried out a significant update to Web-Forge that utilizes some hitherto untapped Nivia Tech called NF4 and essentially released a new version of the FLUX F8.Dev model that runs pretty fast, even on machines with as little as 6GB of GPU RAM.

On my rather aged 3070Ti with 8GB of RAM I've found I'm able to run and get a generation of around 1280x720 in less than a minute, which isn't too shabby at all.

Sebastian Kamph has a video about it here



Details about how to get and install everything are below


There has been an update since he posted that Video and on top of Upscaling and Inpainting now being supported, LORA support has been added though I haven't tested that yet.

Anyway some Forge-generated pics, some with 1.5 upscaling (click on to enbiggen).

PYH3rqs.png
Rrgq5mM.png

n285uBt.png
PzmL3hw.png

qFUFGrQ.png
eaxZQER.png

RO0vB7D.png
6yjgFQ7.png
 
  • Love
  • Shocked
  • Like
Reactions: regawdless, Toecutter and Dreamlord
Looks like X.com's Grok is starting to cook now as well. Seen some impressive generations on the site and it seemingly has no filter. For example, someone generated the prophet Muhammad cuddling with Putin on a couch 😂
 
  • Funny
Reactions: Kadayi and Mirabilis
Been playing round a bit more with FLUX using the NF4 through Web Forge Really enjoying the quality of the outputs but goddamn it really need a better GPU . Generation speed is decent, but trying to upcale with high rez fix is time consuming to say the least.

I will do a proper deep dive into ComfyUI as I suspect for larger outputs that might be a better way to go, although I prefer webforge for the UI aspects.

Was very impressed that I managed to get the one of the girl floating in Zero-G. That's surprisingly hard to do with older models I've found, AI has a tendency to default to floating meaning the subject is in water, and although some people have created Zero-G lORA they're a bit hit and miss. Blondie with the short hair isn't an intentional homage to the Turkish Hitman, rather pure happenstance, but I dig the vibe.



wcNJLsq.png
0RncYJI.png
m5fHYnL.png
nahOcY5.png
Z72jzL6.png
UEMmri6.png
iux6Xux.png
qOLfHXx.png
 
  • Like
Reactions: kiNeMs and Toecutter
Because it's Sunday and I felt like a masochist I thought I'd give my GPU and CPU a serious workout generating some full 21:9 widescreen outputs, just to see how well FLUX handles things and to see how the rig responds. Some of these were just straight outputs and a couple I did some minor (1.5x) upscaling on. Anything past 1.7 and my system is in full-on

No No No GIF

​
Mode. Considering I'm on a 3070Ti with 8GB of RAM and 32GB of system RAM, pretty happy wth the results for the most part. I do think FLUX has some flaws. However as a base model it performs admirably and you can get some great outputs out of it. I think with things like text, although it generally is good with adherence, more than likely it isn't going to get things right off the bat, however you can always just inpaint that stuff and it should be fine.

With faces, it really loves cleft chins, and sharp jawlines, which albeit are a common beauty standard, aren't that common overall (1 in 5 for females) and quite rare in certain cultures. It feels like Black Forst Labs female data set was comprised solely of German Teutonic Supermodels or something.

Repetitive elements is another thing, as you can see from the shots below, where it's often replicating forms in an exact manner, and often there are disappearing into infinity, esp when it comes to corridors . Again, something that could be resolved with inpainting, but hopefully as people produces finetunes of the base model a lot of these aspects will get ironed out. Still good stuff.

I2hcqR3.png

at1ySc8.png

PQMZMtq.png

bFCs9ku.png

4AI5eHO.png

G9bPXw4.png

WfMEMLa.png

lU6xHDn.png


Last two are esp for @Grinchy ;)
 
  • Shocked
  • Like
Reactions: Toecutter and Grinchy
Dude those are amazing. It's blowing my mind what is possible right now with AI.

The untrained eye (like mine) wouldn't have even picked up on the repeated elements.
 
  • Strength
Reactions: Kadayi
Grinchy said:
Dude those are amazing. It's blowing my mind what is possible right now with AI.

The untrained eye (like mine) wouldn't have even picked up on the repeated elements.
Click to expand...

It is pretty nuts. Anyway. As I figured I'd have some fun I've done a few more Syds for you, though this time going for a Blade Runner vibe, and again going for the film format 21:9 ratio . Can't upload this sort of thing at Civitai unfortunately as Celeb likenesses need to be prim and proper but figured I might as well share them here.

9UFzLdU.png

ajlGccP.png

f6WAe9d.png

aHQxt6K.png
 
  • Bacon
  • Like
  • Coffee
Reactions: teezzy, Toecutter and Grinchy
You could get around the celeb likeness thing by doing what everyone should do in the first place when dealing with Syd's face - ask the AI to add a paper bag :ROFLMAO:
 
  • Funny
Reactions: Kadayi
Grinchy said:
You could get around the celeb likeness thing by doing what everyone should do in the first place when dealing with Syd's face - ask the AI to add a paper bag :ROFLMAO:
Click to expand...

wKTT3OY.png

JpO7n7L.png

uIu68mt.png

5Cl2jx7.png



Captain America GIF

​
Gotta say, that last one is fucking perfection tbh, with the double reflection.
 
  • Like
Reactions: Toecutter
Holy hell. That is amazing and pretty hot tbh

Is it possible that it reaches a point where you could give an AI a screenplay you wrote and have it spit out a 90 minute animated movie at that fidelity?
 
  • Brain
Reactions: Kadayi
Grinchy said:
Holy hell. That is amazing and pretty hot tbh

Is it possible that it reaches a point where you could give an AI a screenplay you wrote and have it spit out a 90 minute animated movie at that fidelity?
Click to expand...

I wouldn't necessarily expect anything coherent from AI for a while, but it would certainly look good that is for sure
 
King of Foxes said:
Click to expand...


There's still no consistency, no real coherence in what's being presented, very little in the way of deliberate intention or artistic presentation.

By it's very Nate, AI generated art is always incredibly generic and forgettable. It's very impressive and can often look nice, but it's always trapped in the uncanny valley of the familiar but unreal.

It also seems to have plateaued in quality. There were massive jumps in quality month to month at first, but that video is really no better than stuff I was seeing 6 months ago now, just with maybe slightly smoother animation.

Honestly, I think we're close to the peak of what this stuff is capable of now. It could well end up producing some neat work for those will to put a fuckton of effort in, but it's not the end of the true creatives it was claimed to be.
 
  • Brain
Reactions: Kadayi
Stilton Disco said:
There's still no consistency, no real coherence in what's being presented, very little in the way of deliberate intention or artistic presentation.

By it's very Nate, AI generated art is always incredibly generic and forgettable. It's very impressive and can often look nice, but it's always trapped in the uncanny valley of the familiar but unreal.

It also seems to have plateaued in quality. There were massive jumps in quality month to month at first, but that video is really no better than stuff I was seeing 6 months ago now, just with maybe slightly smoother animation.

Honestly, I think we're close to the peak of what this stuff is capable of now. It could well end up producing some neat work for those will to put a fuckton of effort in, but it's not the end of the true creatives it was claimed to be.
Click to expand...

It does ultimately come down to the effort you put in. Some of the dudes I hang out with are putting out next-level stuff, in large part because they are utilizing cutting-edge tools and not just operating with the default.
 
Was playing around with Grok for the boys as they love Wolverine and SpiderGwen, was surprised how good these ended up

2ZMHW2p.jpg
2ZMJ1A7.jpg


And then i spent far too much time trying to get a cross eyed Wolverine, this was my best attempt

2ZMdx0x.jpg
 
  • Strength
  • Like
  • Funny
Reactions: teezzy, quickwhips and Kadayi
Reader mode
You must log in or register to reply here.
Top Bottom
Back