An Image from Start to Finish using Photoshop and ComfyUI

แชร์
ฝัง
  • เผยแพร่เมื่อ 2 ก.พ. 2024
  • I thought I would do an AI video with all the bits people usually edit out... where it all goes wrong and show how to change tack and sort out unexpected problems. It's not often mentioned that making AI images which fit your intention is slightly like herding cats.

ความคิดเห็น • 11

  • @jaimeortega8939
    @jaimeortega8939 5 หลายเดือนก่อน

    Love this video because it demystifies that notion that one gets what one looks for just by one clever prompt

    • @robadams2451
      @robadams2451  5 หลายเดือนก่อน +2

      The problem is AI can produce a nice-looking image from any set of words. So people type in some daft prompt with "masterpiece" and "trending on Artstation" in it and a great image appears. They then think "I did that!" but they didn't, they only did typing which was mostly ignored.

  • @Mehdi0montahw
    @Mehdi0montahw 5 หลายเดือนก่อน +2

  • @mick7727
    @mick7727 5 หลายเดือนก่อน

    Curious, but did you try creating the AI image first without adding all those little nitty gritty details just to see what happens? Reason I ask is, I'm not very creative on that spectrum of detail so I'm wondering if this method would be worth the effort. Cheers! Oh, I really enjoyed the video about the red-dressed girl and the effects of adding styles and artists.

    • @robadams2451
      @robadams2451  5 หลายเดือนก่อน

      Not in this case, occasionally I get a prompted image I want to take further but they are mostly very generic and I don't feel any connection to them. The art style thing is quite amusing at present because the images used to train are very hit or miss and the captioning poor. Lots of internet scraped imagery has unconnected words associated with the image. I believe there are advances in the pipe line as ChatGPT 4 and others can caption more precisely.

  • @user-dj3rd4my5k
    @user-dj3rd4my5k 5 หลายเดือนก่อน

    There is a photoshop to comfy ui node... seems like it would make great for your use cases.

    • @user-dj3rd4my5k
      @user-dj3rd4my5k 5 หลายเดือนก่อน

      It allows for you to use photoshop as a front end.

    • @robadams2451
      @robadams2451  5 หลายเดือนก่อน +1

      Yes, I've looked at it, I don't have the hardware unfortunately. It's main use seems to be that Comfy iterates as you edit. @@user-dj3rd4my5k

  • @maximood-tired
    @maximood-tired 5 หลายเดือนก่อน

    very interesting. maybe i missed the info, but do you run SD on a different machine?

    • @robadams2451
      @robadams2451  5 หลายเดือนก่อน

      No I quit out of Photoshop as it makes ComfyUI slow due to taking a chunk of GPU band width. There is a plugin where you can automatically feed information to nodes from Photoshop but it won't work on my set up at present.

    • @maximood-tired
      @maximood-tired 5 หลายเดือนก่อน

      does SD run on macos? as far as I knew it only runs on cuda hardware efficiently@@robadams2451