Видео

I understand but i can not read white text on black too much, my eyes hurt, i use a lot of interfaces on dark, but for browser my eyes just can not get used to it, i tried but didn't work for me 😂 sorry, i remember i will add a transition between

@pixaroma my suggestion, change the comfyui inferface to light mode 😂, so all consistent.

Many seems to have it, you can install this in different folder and comes with that node github.com/Tavris1/ComfyUI-Easy-Install

I think it works, but didn't test it

@@pixaroma sure I'll test

Is not that easy, i saw something training a Lora with images with floors and then inpainted the image in the area where was the floor with that Lora, not perfect but it might work in a few tries

@@pixaroma Do you have any workflow ? or any link for reference ?

I dont have, i only inpainted some faces with lora, at the end of episode 24

@@pixaroma Okay Thanks

i downloaded and tried out the workflow. You are a saint, an angel from above of workflow heaven. Thank you so much.

also, i modded the workflow a little bit to generate image to image. magnifique.

The interface changed now is on top left bar where is the workflow menu and others, i put a screenshot before and after on discord with all the menu, look for comfyui-faq channel on my discord

I cannot recommend a laptop because i don't use any, only desktop. My recommendation is to get one with nvidiat rtx card and a lot of VRAM, 16gb of vram so you can run flux models, 24gb are quite expensive. Maybe you can do some slower with 12gb of vram, but lower than that only if you use sd1.5 model and some sdxl. But for flux I will go with 16 gb of vram

@@pixaroma hello, thank you very much for your answer, it's fantastic that you found time to answer, in that case a better solution would be to build your own computer, hmm, could you give me the rest of the parameters of such a computer so I could calculate what will be cheaper for me, thank you very much in advance

@@pixaroma or maybe an option would be a macbook m2 and adding a hard drive to it

For Image generation what is important is to have a good Nvidia card with a lot of vram that will speedup the generation, 32 gb of system ram and SSD so it loads your model faster, and a processor that can handle the video card you choose, so as you can see the video card is most important for image generation, more vram and newer the video card is the faster you can generate

Some had problem running on mac some of the nodes or configuration, the best worked on nvidia with windows

Yes 😁

If is the exact same seed the noise is the same and the image will look ugly you can use a fix seed if you want that will work just to be different from the first seed from ksampler, i also i let it random because i upscale multiple times until I get a version without errors, and since first seed is fixed it doesn't regenerate that step so it only upscale from where it left

@pixaroma I see. From the upscaling tests that I did, when using the same seed number as the one used for image generation and changing step to 1, it produces the same image but upscaled, which is what I am looking for.

Not sure it can do much with one step probably is using the model upscaler, so that probably will give you same results without ksampler only using model upscaler

@@pixaroma ah ok. There were just quick tests anyway 😁. You're the teacher and I'm just the student. Thanks for responding to my questions. 🙏

Found it! Crystools custom node is what I needed to download.

Yes that is the one

I have used omnigen in ep 21 also

Happy holidays ☺️

Yeah are many ways to do same thing. I use mostly Photoshop because is fast and I use it for 18 years 😁 so comes easy to me

It needs a lot of vram and that can cause some troubles, but i rarely use it anyway lets hope they improve it

@pixaroma the problems encountered are code based do far. Some very blatant bugs like uninitialized variables, incorrect indentation, etc

I tried.to keep my workflows to minimum

I don't use ipadapter and pulid because it use insightface that causes a lot of problems and also is not for commercial use so i can not use it. I am waiting for better solutions. And i am training lora either on openart but has like only 2 models training per month but is super fast only a few minutes, either on tensor art that is slower but i can train more. I didn't train locally yet, it only cost a few dollars, i am sure there are more sites for training and for me Lora worked ok with flux

Maybe is something with the internet or github, try later maybe

I have an episode for control net with Flux, should work ok except the stacking, using multiple control net didn't work for me, so only one at the time

@@pixaroma I thought that only depth and canny worked and that all the others like openpose or scrible...only worked in sdxl....on the other hand I'd like to take this opportunity to tell you about a doubt I have. In automatic1111 you could easily edit canny, getting very good results to your liking...now in confyui I don't see how to do it. Is it possible? Thank you very much for sharing, your videos are very well organized and explained.

Depends how you edit canny, you can use probably a black and white brush that is thin to edit like add extra stuff, of if you want to edit just the threshold on how much it see you can do that with the comfyui canny node. You can use the Union comtrol net model

@@pixaroma I mean editing with photopea the image created by canny, this is very easy in automatic 1111 but I can't find the way to do it in confuyi, unless it's doing many tasks to get the same result. I don't think you have a chapter dedicated to this, one on this subject would be nice, although I imagine it's a lot to ask. Maybe for Christmas...lol...anyway, thank you very much for everything. Be good

Thanks, happy holidays ☺️🎄

Sorry, for speaking speed i can not adjust since it is an Ai voice and I don't have control on speed,, maybe i leave longer pauses

I am sure it will get better and better each year, The same happened with image generation, we see improvement each month, happy holidays ☺️

You could take the last frame of the video and use it to make another video

that's what I do, but sometimes it goes in the wrong direction.@@pixaroma

Unless someone make a custom node or something or a better model i don't know a fix

Happy holidays Marcel ☺️🎄❄️

Happy holidays 😁🎅

Happy holidays ivo, thanks for all the help ☺️🎄

Thanks, happy holidays 🎄🎅

Unfortunately you cannot extend VRAM, you need to replace it with a better video card 😁 but I will tell Santa and if you have been good he will know if to send it to you or not 😂

I guess we will see, probably the law needs to change and adapt to new technology

Happy holidays Uday ☺️

Merci à toi aussi

I will have a tutorial for something similar on 28 December on the ai2play channel ☺️

Thanks ☺️