Видео

Yeah, far easier

@@gammastudio3126 thank you bro. How did you find this video?

@@ammarzammam2255 be sure that my channel is the one you are looking for 😁. In the gguf video which you can find in the channel I completely explained which model is suitable for which system. For you I think Q4 is the best

@@999hamstein you're welcome bro ✨

@@armauploads1034 thank you so much my friend 😎🤘🏻

@@ДжониФэйл where did you put them in your system?

@@Jockerai I do exactly as you have the video and put it where you have shown in the video. Thank you so much for your work. You are the best in this field that I have seen. My main problem is that I have macOS m2 silicon? and you have windows. But I am finding solutions to the problems.

I have a favour to ask of you. I've seen all your videos and where you can train your LoRA. However, I do not have the ability to pay for this service right now. Maybe you can tell me where I can read or see how to make a photo with the face I want. Thank you so much for your hard work)

@@ДжониФэйл very good 👍🏻 you can also test PowerLoraLoader from Rgthree collection

@@ДжониФэйл thank you so much for your kind comment ✨🔥

@@ActorClip-hl1me there is a way to train a lora locally I will make a video for that. Stay tuned

you're welcome. Yes sure

Hi there If you mean for your second group (Sdxl group) you need to save image, You can just double click on the screen and search for save image node . Add it to workflow and connect the VAE Decode node to Save image node that you have just added. And for the 4th group (flux group) it has save image node and it will automatically save your image in output folder in main comfyUi folder

@@Jockerai it works! thank you

for inpainting with flux you can watch this video : ruclips.net/video/U4RLjDKrM3E/видео.htmlsi=YqOEF6iVNAidW2J2

Thank you. there are many ways to have consistent character but I recommend using a trained lora, or if you don't want to train a lora you can use an specific person name in all prompts like a Russian, American, Asian name. It would help ai to somewhat make a consistent character. But the best results would be using trained lora for a character which I made a video for that in my channel

@@Jockerai Thanks a lot for your prompt response and tips.

First time I here that. What upscale model you are using?

This bug is related to using the matrix multiplication (matmul) operation in Metal Performance Shaders (MPS) on macOS, which detects a data type mismatch: FP16 (half-precision floating point) and BF16 (bfloat16). Metal does not support mixed operations with these data types. So it turns out that Flux with custom LoRA will not work on my macOS m2 max? Is it possible to replace this LoRA with another analogue?

Hi! To resolve the issue with loading the Lora, try removing the FluxLoraLoader and replacing it with PowerLoraLoader. But before that, go to Manager, then in the Install Custom Nodes section, install RGthree. After that, restart ComfyUI. This should likely fix your issue.

What does the expanded image area look like? I mean, what exactly is generated in the parts that you're extending? And are you sure that you have downloaded the Controlnet model which I mentioned in the description?

@@Jockerai Yes, I had the models already, but for extra measure I downloaded those again from your links and named them differently to not mix it to previous ones.

Before I went to bed last night, I noticed that it actually does generate something (that looks like what I prompted) on the expanded area, but ONLY if drop the strength in Apply Control net to something like 0.2? If I set it closer to 0.5 or above, KSampler preview image basically shows that the expanded area get noisy, then less noisy, but instead of converging to image, it degenerates back to how it started... I'm wondering what stupid mistake I did lol. I've updated Comfy and all the extensions to latest ones every day, usually more than once.

@@devnull_ what a strange thing🫤 I have just tested it now and with 0.2 and 0.5 and 1.00 , I got good results. what Sdxl checkpoint model you are using? Maybe something is wrong with your model !

@@Jockerai I downloaded the same model which you used, triple checked the values you used.

@@AguimarNeto thank you bro🤘🏻😎

Yes you can. Just be sure to give it images that have body. Whatever you give to ai it will make the lora with them even the shape of fingers and nails and everything.

what do you mean of body only? for color try decreasing FluxGuidance, set it between 2 to 2.5 or increasing it between 4 to 5.5

And for everyone's sake, already start using that new UI, standalone ComfyUI already has that as the default, and repository version too has it, this old one is going to be deprecated and not supported soon anyway :)

@@devnull_ Thanks for sharing this. Yes I know I will use new interface for my next videos. ;)

No you can't. go to their official website blackforestlabs

yes it works, at the end of the video I say that you can use them and how

@@Jockerai Great 👍👍😄

you're welcome ok sure I'll make a better edit for my videos🙏💚

I made the workflow step by step in the video to give a better and more detailed explanation. You can make it to follow the video ✨

it is so funny🤩😁

you can but it takes very long time and in my opinion it's not safe and you hardware might be at risk. For your GPU it may take more than 24h

you're welcome bro

I made the workflow step by step in the video to give a better and more detailed explanation. You can make it to follow the video ✨

@@Jockerai people are used to just downloading the workflows, making them recreate your workflow is really unnecessary

​​@@undoriel Problem with that mentality, is that you won't understand the intracesies that made the workflow in the first place.

I made the workflow step by step in the video to give a better and more detailed explanation. You can make it to follow the video ✨

You're welcome

❤❤❤❤

Thank you ❤

Waiting for a minute?! What do you mean? 🤔

@Jockerai I saw your Reddit post and was having trouble using controlnets with Flux. Thanks again!

@@CoreyJohnson193 which Reddit post😲

@@rick-deckard sometimes you get good results and sometimes not

@@mohsen1208 if you watch the video I used a trick, when you change the prompt and green bar reaches to the sampler custom advanced node , cancel the process and click queue prompt again

@@ehabeltorky88 yes it works and on top menu bar there is section I forget it's name but when you open it you can see options for fp8-e5m2

@@Jockerai I think you mean Diffusion in Low Bits, thanks I'll try it.

@@biggreg100 you're welcome bro ✨