Flux.1 vs AuraFlow 0.2 - Is Flux The Best EVER?! Free & Local in ComfyUI + Upscaling

Flux.1, AuraFlow 0.2 and AuraSR are some of the latest AI image models you can run at home for free on your own PC. Installing Flux.1 Schnell or Dev for ComfyUI is super easy, as ComfyUI has day 0 native support - simply download the models and run! Upscale Flux images with AuraSR in seconds 😀

AuraFlow & Flux work best with at least 24GB VRAM, but fp8 options are available for Flux which use less than 14GB! Flux has both schnell and dev versions available, with Apache and non-commercial licences respectively. Run for free on your own PC, or use their pro service!

Which is the best? Take a look and find out what each of them can do!

Links -

Want to support the channel?

Patreons get loads of extra Flux.1 goodies too - upscale, inpaint, high res fix and more!

== Start Learning ComfyUI in 3 Easy Steps! ==

[TheDocPixel]

Let’s not forget that Black Forest was the original team that StableDiffusion and Midjourney were built on top of. These guys are geniuses!

[graylife_]

The specs kill me though.. 24gb vram 😅

[wsippel]

Flux was created by 14 members of the original Stable Diffusion team, they certainly know what they're doing. Black Forest Labs, the company behind Flux, is headquartered in the Black Forest (southwest Germany), hence the name.

[muuuuuud]

This is such a thrilling era we live in, I always had too many ideas i had nowhere near the time or energy to produce before and now I can almost produce them in real time :3, its incredibly actualizing creatively. Thanks for the videos Nerdy Rodent! ^-^

[KoolenDasheppi]

The Flux model is very exciting, I've already tried it out a bit and it's currently blowing my mind. Good thing I have a 3090 lmao

[Tystros]

Can you also make a video about the full-quality Flux model? Flux Schnell is basically a Turbo version with worse quality. If it's this good, I want to see what the "real" model can do!

[USBEN.]

Bro they came swinging with FLUX and knocked out all competition. It gets better! They are working on SOTA, that just might force SORA out into the open.

[runebinder]

Hearing the genuine surprise and being lost for words with Flux was very funny, had set up the dev version of Flux earlier so had an idea of what it could do and was interested to hear someone else's perspective, didn't disappoint :)

[jibcot]

I have been making custom AI Birthday cards for over a year now, it's a great use of AI.

[Neko_Dave]

Flux seems the biggest step forward for local image models since SD 1.x. Shame it's so demanding to run

[Ethan_Fel]

sadly Flux license and high gpu requirement will probably stop it from being the new standard.

[Halsu]

Just a note about the schnell version: While it indeed can make decent images at 4 steps, i find that often higher steps will help, for example in cases of missing or extra limbs etc., and for creating a more detailed image overall. I find the most useful range to be between 6 and 25 steps, maybe.

I also did an experiment with higher step counts. Much higher. I stress-tested it on up to 640 steps (yes, 640, no extra zero), and it still produced a valid image, though the results were perhaps a little weird, Flux sort of overdid prompt following and added odd details.

[TheGanntak]

Sitting in the corner into my RTX2070 8 GB

[ickorling]

Looks like the community missed 'Dimba', from 2 months ago. A transformer and mamba image ai, which is good. This flux does look better, but progress will come from blending solutions and rich high quality data. 🎉

[marsandbars]

9:55 "He's got two hands." It's unthinkable that one could possess two hands.

[Firespark]

Sucks its takes so much vram. Needing over 24GB is kinda nuts. Hopefully they get that down in future models.

[robxsiq]

Tried both and Flux...man, Its fantastic. Different level remaking the game. Its what we all wished SD3 would have been...it runs like a truck in mud, but the results are fantastic. it (mostly) knows the human anatomy...to a degree. a bit similar to SDXL base model, and it nails the look overall. This can be the winner if we can tweak it to run a bit better and get some tunes out. It follows prompts better than I ever though could be possible. There are issues still of course. weird feet, some screwy things, but overall. Flux is the king in my book for general artbot prompting. Now, for adult content...yeah, you'll need finetunes for sure as it isn't much on that, but it knows the body. For style...it struggles a bit to get away from realism compared to say, Dall-E, but again, finetunes...

[DanHotz]

i'm using RTX 3060 with 12GB VRAM and it is very fast! Probably faster that stable diffusion, all other things being equal.

[viddarkking]

Honestly I look foward to AuraFlow 1.0.
That number Alone will spark a rush of fine tunes in of it self.

[Patheticbutharmless]

NerdyRodentMan you holding your breath/loosing your so