Is AuraFlow The Open Source Model Of The Future? 

Yeah it’s not great at it at the moment, although you could do a refiner workflow using sd1.5 or sdxl but will be much better when it’s fully trained.

Indeed very promising! The fully trained model should be great.

Honestly for this kind of stuff a 3090 is all you need

Not sure if it does directly but you can import the workflow into comfy and use swarm front end.

@@MonzonMedia thank you so much for the answer and the content. i think i'm gonna stick to comfyui since its the most updated one 🙏

Thank you, appreciate you stopping by!

Yes indeed! Have those lined up to record soon. Been trying out both all week. Have to tried them yet?

@@MonzonMedia yes , i tried kolors and it's actually pretty good, maybe not so good again with photorrealism and skin,but apparently You have better results if You promp in traditional chinese, i coul not try 900m since i only use this models in places like tensor art, and it requires a Lot of stuff to make The model work so i guess i Will wait until someone make it work on There

@@danielc121 The 900M model and generally Sigma can be run pretty easily in ComfyUI. There is even an easy way to do it in SwarmUI which I've been meaning to cover in a video. As for AuraFlow and chinese prompts, I did a quick test just using google translate and it does give it a different look.

They have their own upscaler github.com/fal-ai/aura-sr

@@MonzonMediaty ok interesting! And sth like img2img hires fix would be possible too ?

@@fixelheimer3726 It should but unfortunately because it's still not optimized it's way to slow and vram hungry at the moment for me to test out other workflows. I hear though that once the official model is released they plan on doing their own controlnets and other conditioning methods.

hahaha yeah I hear ya, nothing wrong with that at all. In fact data sets containing NSFW help a lot with body composition.

😊 Yeah, I mean it's great to see the progress but not realistic to use regularly at the moment.

@@MonzonMedia Well curiosity got the better of me and I gave it a whirl. Terrible results with human bodies....misaligned, twisted, appears to be good prompt adherence tho! Definitely needs more cooking as still a bit too underdone.... Thanks for sharing tho appreciated!!!

@@yiluwididreaming6732 Couldn't resist huh? hehehe, yeah it's similar to SD3 medium in that regard but lot's of potential. Hopefully the full model will be released soon. Rumor has it could be late Aug or Sep but we'll see!

LMAO! Yeah big freakin file so best to wait when the optimized model is out. The next video I'm covering is Kolors. Runs way better on ComfyUI, optimized for my system and it's fully trained. So far, very impressed.