I've been using FLUX. It is quite an upgrade from previous Stable Diffusion models. Being able to do text and fingers is really nice. Still not perfect, you get decent fingers 40-50% of the time, which is better than the 5% from previous models.
1. Yes, I think flux is awesome, I tried Stable diffusion on Mimicpc, and of course this product also includes popular tools for AI such as RVC, Fooocus, and others. I think it handles detail quite well too, I can't get away from detailing images in my profession and this fulfills exactly what I need for my career.
I finally got it working on my AMD RX 7900 XTX graphics card. It took a while as I had to use ZLUDA, but I’m very happy with the outcome and the speed. Flux is amazing!
Thanks for the tutorial, i was able to get it working. I mostly use A1111 for 1.5 and sdxl and will probably stick with that for awhile because it takes my 3080 2 to 3 minutes to generate a single image with flux with just the default settings. I am impressed with the quality of the images though!
Hello Mohammed, thank you so much for this Video and the one who you explain how to install Comfyui. I looked so many videos how to run Flux on Comfyui. No one works...until i saw your video and now... everthing works fine and smooth on my PC. You have exact those videos i was looking for a long time. Thanks again. Have a great day. 😃
@MDMZ, My friend, the quality, precision, and skill you consistently deliver are nothing short of amazing. You always provide us with impressive and amazing work, and I can only imagine the effort that goes on behind the scenes. Thank you, and please never stop. Just a quick question, if you don't mind: do you live in Malaysia?
Yes, I think what you said is great. I am using mimicpc which can also achieve such effect. You can try it for free. In comparison, I think the use process of mimicpc is more streamlined and friendly.
Installed Comfy UI, downloaded the .safetensor, vae and clip files, the workflow and did exactly like in the video. On my 3080 ti I get got prompt Using pytorch attention in VAE Using pytorch attention in VAE C:\Users\shutd\ComfyUI_windows_portable>pause Press any key to continue . . . and then the prompt window closes
Good video. Please make another video on how to integrate "img to img" using flux. This will help us take existing image and write text on top of it instead of generating an image with text from scrach.
@@RANJEET3939 to be honest, I rarely make instant tutorials about new tools, cause that means I will be covering something unstable that's gonna be improved over time.
Thanks bro! edit: Actually i got just black image... i did everything as you did. Googled it and it said i probably have the wrong vae, but i have the exact same vae as you. Do you know where I can get help troubleshoot this? edit edit: actually, it was the clip loader problem. you didn't quite explain how that clip loader worked and somehow i had two of the same clip chosen. You have to have different clips. in any case, thanks again for this otherwise great tutorial.
Hey buddy @MDMZ I loved your work super easy to understand and very crisp video. Please can you suggest what to do if we want same persons face in all different prompts?
I don't know why, but my images are just solid noise... (I tried to add a link to a picture, but my comment deletes🥲). I downloaded all the dependencies (model, 3 encoders, VAE, and the workflow you mentioned), but it's still a mess. I have a MacBook Pro M3 Pro, so I also updated PyTorch to the nightly version. What am I doing wrong? I tried changing schedulers, samplers, steps, and even the size 😃 but nothing helps...
When using the new low-mem (4GB) model Q4KS - Do I need to install the other files you mention beside the main FLUX Q4KS under unet? or because it's a low-ram different model, other files are needed in stead? Thanks ahead 🙏
I am currently using ComfyUi and as a checkpoint Dreamshaper 2.1 Turbo the combination works well together with my AMD system. My questions, does Flux run stable on an AMD system? I have a 7600 XT with 16GB VRAM, 32GB RAM and AMD Ryzen 5 7500F
Hi. Thanks for the good guide. I have a question: why my pc crash with this setup? I have 16 RAM and a RTX 4070. It's strange, because even if I start Queue process and immediately close the app, after few seconds my pc totally freeze
wanted to try this but when I draged the workflow json file instead of "load diffuse model" node I had "UNETLoader" node without weight, so an error occured... have no idea why there's no "load diffuse model" node
Great video, thanks! How do we create additional images that are based on the first image? For example, once you create the cover for LUNA, how do you create all subsequent images using the same character and style?
I am on AMD GPU RX 7600XT. For me dreamshaper_8 works very well in comfyui using checkpoint. But I tried so much to get flux_1 working, it just doesnt work. I followed your instructions, as well as other tutorials. I even updated comfyui to latest version. But everytime I queue prompt, it gets stuck at first node 'Load Checkpoint' for a minute and then server crashes. No error in CMD, just 'got prompt' message. I even tried checkpoint version of flux from their official documentation, still the same issue. I wonder if I am doing something wrong, or flux is not yet supported on AMD GPU.
i have the same problem except mine says: got prompt model_type FLOW model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16 and then it hangs forever (ive waited hours) RTX3060 Ti
How much time per image would it take on 12 gb vram, for making 2048x2048 resolution images? Have people recorded the time for that kind of resolution? Please share all kinds of timings on high resolution. I don't like low resolution generations. Upscaling landscapes seems to ruin everything.
Milion tutorials how to use FLUX it in Comfy UI and very few how to set everything to work in normal UI for ordinary people and none how to use Controlnet with it without Comfy UI involved.
2:51 => "Welcome hell". LOL. Also, you want to hyphenate "smiley-face" otherwise you'll get a token drift. 3:08 => There's no such thing as "hyper-realistic". Something is either real or it is not. Also, "portrait" is a painting, so you're asking the machine for both a realistic image and a painting. You kind of screwed yourself there. You can't ask for "detailed skin textures" because it's not a 3D model. You have to render at a higher resolution than 512x or 768x. 5:24 => How can a network of machines know what "fitting perfectly with the theme" means? It's not another human. You're watering down your prompts, which leads to worse results.
hey friend, you're telling a human what he told a machine that was taught by a human that he's wrong. A better understanding of tokenization might help you get why certain tags (tokens) "work" because they are specific to the dataset the model was trained on. Example: using danbooru tags as opposed to natural language can give you more ideal results based on the tags used to train the model.
@@MDMZ the free version is generating crap, I saw the option for the pro version for better generations but I gotta pay, which site are you using to generate with flux??
i think i need a 4090,my 4060 laptop is too slow to use flux😢,it takes about 5 mintues to produce a 1024*1024 picture with flux dev fp8,oh god,who can buy me a 4090😥
FLUX is truly amazing, but what you're doing is just subpar. It's a little bit of this and a little bit of that... What's the point of making these videos if you don't take the time to explain things properly?