🌟Get Midjourney to generate your images: 🌟 www.aiarty.com/midjourney-prompts/?ttref=2409-aia-mj-ytb-AN-incl-wxr 🤯Aiarty Image Enhancer, the best AI tool to enhance and upscale images: 🤯 www.aiarty.com/?ttref=2409-aia-mj-ytb-AN-incl-wxr
Thanks for the video! This is already on the limit of best consumer cards people usually buy. At this pace nobody will be able to run next generation of AI image generators on their own computers... I'll accept SDXL is enough for me, at least locally ;-)
SDXL is enough but you need to try this. The challenge "alligator coming out of manhole" was completed by flux in a single click. I was having a very hard time doing in SDXL, it was unbelievable when flux did in a single attempt.
@@Aininja94 "The challenge 'alligator coming out of manhole' was completed by flux in a single click". OMG! Really? I will have to try, I was a bit "scared" about ComfyUI and I'm not very sure about this Frankenstein program, it looks like it's not very polished... Also having to wait some minutes to see the image when with Hyper-SD speed I get acceptable images in some seconds... (I've got a similar setup to yours, but with 12GB VRAM)
12gb should work with the flux I showed here. The project has many bug I checked, but the developer is working daily. Now this too much for a person who may have a job as well. So, can't complaint. Also I am looking at swarmUI for flux which is not much complex. You can check if you are using AI for production purpose.
@@Aininja94 Oh not complaining, I'll just wait for it to be more stable. I'm not using these things for professional matters, it's just curiosity. I've made some wallpapers for my computer and that's it ;-) I prefer simple programs like Easy Diffusion and Fooocus but I understand that for more control you need complex programs. I'm amazed with the the things they're doing with AI these days... and a bit scared about how quickly it's advancing!
Im using FLUX Dev fp8 on a Nvidia RTX 2070 with 8GB VRAM - and it works great. No need to use ComfyUI - just use the UI "Forge" (it´s another Version of A1111), it´s compatible with Flux. Flux has excellent prompt following, which makes creating professional images much easier and less of a radom game. Flux follows prompts so precisely that you can actually determine where (in which corner of the image) which object should be shown. It also generates flawless hands and proportions most of the time. Flux is worth it! ;)
I installed SimpleSDXL on August 20th and did not have the problem with installing the model files, or with missing files in the fooocus_expansion folder. There must have been an update since the video was made.
@@Aininja94 Yes, I saw your videos on Ruined Fooocus. Ruined is not for me. I do not like its divergence from mainline Fooocus, SimpleSDXL is much more consistent which makes it a natural successor to mainline Fooocus. It is worth noting that Ruined Fooocus is 834 commits behind mainline Fooocus (I don't know what version of the mainline that translates to). In contrast, SimpleSDXL is completely up-to-date with Fooocus 2.55. I definitely do not want to go back in time and lose the features I have come to rely on in Fooocus 2.55.
Actually I have not used colab so i do not have knowledge about it. Still learning, I will make collab videos as well but right now, I am learning python to customized such project. Kindly a lot in plate and things are change everymonth.
So as far as I know this "schnell" edition for older GPUs is not as good as Dev. But is it still better than regular Fooocus? What do you think? Maybe a video comparing Fooocus generated images with this? // Btw. does it require an RTX or could it run on GTX 1080?
It's better then regular SDXL bcoz trained on 12 billion parameters. Explains image through text prompt is more easy then SDXL, that's what makes it better. Comparison can be done, the thing is. Image quality and prompt understanding is better in flux. RTX makes the process faster due to xformars and stuff. I am not an expert, still learning.
The file size was a challenge because for some reason my browser was keeping files temporarily in c drive (18gb free space) later moving to other drive with larger space, selected as download path by me. Download got cancelled twice.
I tried this fork yesterday and didn't like it at all. It's too much of everything - a salt is salty... And besides, this fork is constantly slowing down and freezing. I fooled around for a couple of hours and uninstalled it...
Lora with flux is quite a work because there is a difference how fooocus works. The developer is actively working but too much for one person. I have found one more project supporting flux. Checking right now. Will make a video if it's better
you need some serious hardware to run flux dont you? well if you don't want to wait hours for it to generate one picture . i dont think my 3050 will be happy lol
With the GGUF models, you can run FLux in as little as 6GB and get results in 2-3 minutes. For NF4, slightly longer. If you've got a 12GB card, it'll be much faster with the smaller models.
🌟Get Midjourney to generate your images: 🌟
www.aiarty.com/midjourney-prompts/?ttref=2409-aia-mj-ytb-AN-incl-wxr
🤯Aiarty Image Enhancer, the best AI tool to enhance and upscale images: 🤯
www.aiarty.com/?ttref=2409-aia-mj-ytb-AN-incl-wxr
Thank you for creating this video. So far you are the only person to cover SimpleSDXL in English on RUclips, which I find a bit mystifying!
I will go through the updates done by dev, hope there are many things fixed and make a updated video. I saw you had posted errors, that's great 👍
Excellent tutorial, thank you so much!
Your welcome!
Thanks for the video! This is already on the limit of best consumer cards people usually buy. At this pace nobody will be able to run next generation of AI image generators on their own computers...
I'll accept SDXL is enough for me, at least locally ;-)
SDXL is enough but you need to try this. The challenge "alligator coming out of manhole" was completed by flux in a single click. I was having a very hard time doing in SDXL, it was unbelievable when flux did in a single attempt.
@@Aininja94 "The challenge 'alligator coming out of manhole' was completed by flux in a single click". OMG! Really? I will have to try, I was a bit "scared" about ComfyUI and I'm not very sure about this Frankenstein program, it looks like it's not very polished... Also having to wait some minutes to see the image when with Hyper-SD speed I get acceptable images in some seconds... (I've got a similar setup to yours, but with 12GB VRAM)
12gb should work with the flux I showed here. The project has many bug I checked, but the developer is working daily.
Now this too much for a person who may have a job as well. So, can't complaint.
Also I am looking at swarmUI for flux which is not much complex. You can check if you are using AI for production purpose.
@@Aininja94 Oh not complaining, I'll just wait for it to be more stable. I'm not using these things for professional matters, it's just curiosity. I've made some wallpapers for my computer and that's it ;-)
I prefer simple programs like Easy Diffusion and Fooocus but I understand that for more control you need complex programs.
I'm amazed with the the things they're doing with AI these days... and a bit scared about how quickly it's advancing!
Im using FLUX Dev fp8 on a Nvidia RTX 2070 with 8GB VRAM - and it works great. No need to use ComfyUI - just use the UI "Forge" (it´s another Version of A1111), it´s compatible with Flux. Flux has excellent prompt following, which makes creating professional images much easier and less of a radom game. Flux follows prompts so precisely that you can actually determine where (in which corner of the image) which object should be shown. It also generates flawless hands and proportions most of the time. Flux is worth it! ;)
Flux+Fooocus in SimpleSDXL is working via Comfy). Ok, but Ruined Fooocus do it alone?
It has comfy Ui stuff inside.
@@Aininja94 ))) I was beginning to think that
The code will remain the same for the functionality then why not copy and paste it, be it from comfy ui Or any other application.
I installed SimpleSDXL on August 20th and did not have the problem with installing the model files, or with missing files in the fooocus_expansion folder. There must have been an update since the video was made.
I am working on a video, better then simpleSDXL. Will upload in 24hrs, I think you should try that.
@@Aininja94 Yes, I saw your videos on Ruined Fooocus. Ruined is not for me. I do not like its divergence from mainline Fooocus, SimpleSDXL is much more consistent which makes it a natural successor to mainline Fooocus.
It is worth noting that Ruined Fooocus is 834 commits behind mainline Fooocus (I don't know what version of the mainline that translates to). In contrast, SimpleSDXL is completely up-to-date with Fooocus 2.55. I definitely do not want to go back in time and lose the features I have come to rely on in Fooocus 2.55.
Agreed👍
THX, but i need help "WARNING:root:lora key not loaded: lora_unet_output" it doesn't work
Is this the problem in Ruined Fooocus? You posted on a different video
🙌
want to run this on google colab how can we dot this any idea?
Actually I have not used colab so i do not have knowledge about it. Still learning, I will make collab videos as well but right now, I am learning python to customized such project. Kindly a lot in plate and things are change everymonth.
Wow thanks men you are pro
Any way this could be run on colab anyone?
Working on a video, flux on colab. Should be out after 4 days
So as far as I know this "schnell" edition for older GPUs is not as good as Dev. But is it still better than regular Fooocus? What do you think? Maybe a video comparing Fooocus generated images with this? // Btw. does it require an RTX or could it run on GTX 1080?
It's better then regular SDXL bcoz trained on 12 billion parameters.
Explains image through text prompt is more easy then SDXL, that's what makes it better.
Comparison can be done, the thing is. Image quality and prompt understanding is better in flux.
RTX makes the process faster due to xformars and stuff. I am not an expert, still learning.
I think the dev removed some of the files in the page
There have been changes almost daily
Nice video
how to run it on colab?
Sorry, I am not an expert on colab. Still learning and there is a lot to. I will be doing using colab in future.
Is there a way to run this with AMD GPU?
I don't think so, I searched for it after your msg. I did found some documentation for sdxl but not for flux 😕
The files are not 49GB. They are 4+ GB. I certainly would download and install if they were that large.
The file size was a challenge because for some reason my browser was keeping files temporarily in c drive (18gb free space) later moving to other drive with larger space, selected as download path by me.
Download got cancelled twice.
I tried this fork yesterday and didn't like it at all. It's too much of everything - a salt is salty... And besides, this fork is constantly slowing down and freezing. I fooled around for a couple of hours and uninstalled it...
Ohh, this was my first 30min I used. Yes there could be bugs due to which it is freezing.
nice ! except no loras accepted on Flux on this program
Lora with flux is quite a work because there is a difference how fooocus works. The developer is actively working but too much for one person.
I have found one more project supporting flux. Checking right now. Will make a video if it's better
@@Aininja94 really good job anyway
you need some serious hardware to run flux dont you? well if you don't want to wait hours for it to generate one picture . i dont think my 3050 will be happy lol
Is that 6gb variant?
With the GGUF models, you can run FLux in as little as 6GB and get results in 2-3 minutes. For NF4, slightly longer. If you've got a 12GB card, it'll be much faster with the smaller models.
It will run in a MacBook M2?
@@fabiotgarcia2 i think macbook support is patchy at best ( was with foocus anyway) , ideal setup seems to be a windows or linux pc with nvidia gpu
The m2 air 8gb struggles, sometimes stucks and battery goes quick. 16gb should work.