Get my 3 WORKFLOWS here: www.patreon.com/posts/flux-controlnet-110607421 My Flux Install Guide: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-tXO6SJ-6Eb8.html My IPAdapter Install Guide: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-KXxk_bGDCzc.html Join my Discord: discord.gg/XKAk7GUzAW
Thank you so much, especially for the GGUF, not having enough VRAM is like gambling, sometimes it renders in 2 minutes, sometimes it stucks loading because system is running out of memory
I hated the thought of ComfyUI, but after a couple of days I realized how much more intuitive it is than A1111... If something goes wrong during an A1111 gen, you have to be a "terminal genius" to figure it out... but Comfy shows you what's wrong and actually allows you to literally "connect the dots" better... Pun intended!! Yes, there are still some things missing (easy ADetailer!!) but I'm liking it and I didn't think I would. My 2 cents!! 🤓
I don't understand this sentiment. Comfy is arguably harder to setup, but once you have done so and gathered a repository of workflows, it's easy as cake.
@@vannoo67 What's not to understand, you state the issue yourself - "harder to setup, but once you have it done..." Getting it done is where the problems lie ad the manager only goes so far. Even if you get all the nodes installed it doesn't cover models and errors after that. You've basically said that driving a car is easy once you have it built and running, while passing over that you start with a pile of parts on the floor. I say this as someone that uses Forge and Comfy.
Watching Olivio while I train my first SDXL LoRA :) How do you show the extension name above the node 6:20? I'm all in with ComfyUI, started with A1111 of course because it's easy, but Comfy is way better and more flexible.
@@mirek190 For sure FLUX is establishing itself as the new supreme generator. But it's very demanding on the hardware. Don't diss SDXL yet, it still gives amazing results.
@@KlausMingo yes but how long ... is much harder to train , is different architecture so can not understand prompt as well as flux, Hardware limit is not a big problem (16 GB cards are getting standard for people who are doing something more than watching youtube) people slightly moving on ...
@@KlausMingo I not blame you ;) Probably If you really wanted to play with AI more seriously you would buy some card at least 16 GB. For instance a years ago I changed my RTX 3060 8GB to used RTX 3090 24 GB for 700e ;)
Error : No module named 'controlnet_aux' (ControlNet) (I'm using Flux1-dev-Q4_K_S + comfyUI + Controlnet) There is a error during the node "Canny edge" : "Error occurred when executing CannyEdgePreprocessor : No module named 'controlnet_aux' I re-installed ControlNet auxiliary model manually several times, and updated everything. It didn't change anything.
Silly Question: its been a while since anyone made an up to date update of which Checkpoint Models can render human faces and bodies the very best in august 2024. You have seen and used so many different ones, which one is your goto model to get 100% real Photos of human beings, disregarding all cartoon stuff in this quest /question. ? is it still juggernaut or realistic or pony? Pony seems to be 10% all cartoon anime in it i saw.... what model would You use? i have tried some Loras of a Singer friend of mine with different checkpoint models. but have not found a number one photo realistic one yet... with or without Lora applied to it
i started comfyui with all the recommended setting but still getting failed to allocate (low VRAM), despite have 16GB of VRAM on my 4070ti. Is there another method to resolve this?
I never got ComfyUI to do anything... at all... not even run the demo project... so I just installed FLUX with Pinochio Not a whole lot of options with the web UI that way... like a way to pass it an image to base its generations on. Most of these AI user interfaces and installers... are incredibly... not well thought out, and getting more fractures and spreading out as tree branches to be abandoned... every day. They're all their little fiefdom isolated tools, and not building one big single AI, and then adding talents to it...
I know this may sound dumb, but am still a huge fan of A1111 and SD 1.5... so many options now it gets confusing. Should I scrap A1111 and move to Flux? Will Forge controlnet work as well as with A1111? Just when you get things figured out, something new comes along lol
I am not able to download or install controlnet for stable diffusion, neither on my PC nor on my laptop I also tried it on my friends laptop, and it didn’t work Is it because of antivirus? Have anyone faced similar problem? Need help
ComfyUI for FLUX, everything else Fooocus, especially inpaints. Tried Forge to work with FLUX, for normal use it works like a charm, but whenever lora and others kick in, it starts to mess up with VRAM which leads to being very unrealiable.
I use Forge Daily. It's been recently updated for Flux, It's new NF4 feature Marginally increases SDXL 1024x1024 generation speed. Flux is still kinda new and is why I haven't jumped on it, Too many caveats to Flux's usage such as being a bit dependent on ComfyUI.
Yeah, Forge is great and being actively developed again. But, ControlNet for Flux not totally implemented as yet, but hoping that gets sorted very soon.
I am using Forge. But I have a question: Is Flux really slower than SD1.5 when aiming at higher resolutions? If i want to end up with an 2MP picture, i can render this directly in Flux, SD and SDXL would most likely need to be upscaled or similar techniques. So lets say 1920x1080 render in Flux vs. in SDXL...What is really faster i wonder. By the way no controlnet for Flux in Forge, right?
so far I like A1111 the most, forge is a little faster but it wasn't working with some of the extensions so not really worth a few seconds faster. just now learning comfy ui and boy is it not comfy to learn lol, I can see it having lots of specific uses but not having images for the loras or models is annoying since many are named random things and looking at a picture, list of tags, and having a direct link to its civitai page is faster and more plesent to use in A1111 than trying to rename a bunch of loras to make sense in comfyui
I will have to see about installing the low memory models as I can not any renders to complete as I keep getting the "torch.cuda.OutOfMemoryError: Allocation on device" error. This is with a fresh install on a computer with a nVidia 3080 with 32GB of ram.
Input type (torch.cuda.FloatTensor) and weight type (torch.cuda.HalfTensor) should be the same error from MIdas Depth Map :( dont know how to solve this :(