Photo restoration using Photoshop and AI SDXL model in Automatic 1111

แชร์
ฝัง
  • เผยแพร่เมื่อ 9 มิ.ย. 2024
  • The photo was taken in 1935 or 1936 and it was the only copy of my grandfather and grandmother most probably on their wedding day. The photo can be seen in the following link
    pics.kamilari.org.gr/picture....
    Other tutorials and maybe better ones, about image to image photo restorations, can be seen in another video • Using SDXL Turbo based...
    00:00 - 00:07 Intro
    00:08 - 03:12 Scratches removal, missing parts restoration
    03:13 - 05:46 Face enhancement
    05:47 - 07:43 Dodging and Burning
    07:44 - 10:01 First rejected SDXL Image to Image generation
    10:02 - 11:27 Back to Photoshop for a second try
    11:28 - 12:00 First SDXL Image to Image generation
    12:01 - 13:22 Back to Photoshop for more cleaning and photo enhancement
    13:23 - 13:37 Second SDXL Image to Image generation
    13:38 - 14:39 Final touches and save in Photoshop
    A photo restoration that is done mostly in Photoshop, but it also combines generation from the SDXL model in AUTOMATIC 1111 web ui and the image to image mode. This generation is used to help with cleaning up and possibly generate more details for missing parts of the photo.
  • วิทยาศาสตร์และเทคโนโลยี

ความคิดเห็น • 14

  • @user-frond-end_dev
    @user-frond-end_dev 6 หลายเดือนก่อน +1

    Hello, please leave the sources of photos from your video lessons in the description under the video

  • @andy-jg1vr
    @andy-jg1vr 6 หลายเดือนก่อน

    Awesome!!! Thanks!!!👍👍👍 And how about colorizing?

    • @nikolakakism
      @nikolakakism  6 หลายเดือนก่อน +1

      Hello, thank you! I personally avoid colorizing since it is a form or re-imagining the photo so I do not have lot of experience with it. Photoshop neural filters have introduced a Colorize plugin which can automate a color layer from where to start and make some adjustments. However the results are not the best, most of the times.
      Another workflow is to use the Stable Diffusion model in combination with the Canny controlnet model to generate alternative color images that do look like the input one. From these similar ones you can get color and details information.

    • @andy-jg1vr
      @andy-jg1vr 6 หลายเดือนก่อน

      @@nikolakakism Thanks a lot!!!👍👏👏

    • @nikolakakism
      @nikolakakism  6 หลายเดือนก่อน

      Hello there, I made one video that refers to a colorizing technique with the recolor model of the Controlnet th-cam.com/video/gCgOnOQdcqU/w-d-xo.html @@andy-jg1vr

  • @Joe_I_Hadley
    @Joe_I_Hadley 4 หลายเดือนก่อน

    Holly Molly

  • @MindEversion
    @MindEversion หลายเดือนก่อน

    super! but next time please without music. please.

    • @nikolakakism
      @nikolakakism  15 วันที่ผ่านมา

      There won't be a next time for this type of content anyway. Thank you though!

  • @kyawaung1232
    @kyawaung1232 7 หลายเดือนก่อน

    how to stable diffusion easy install tutorial sir

    • @nikolakakism
      @nikolakakism  7 หลายเดือนก่อน +1

      Hello there, you can follow along this tutorial for the installation. There are more videos on how to install it however th-cam.com/video/rtUpIY9Opjs/w-d-xo.html

  • @rokos3701
    @rokos3701 6 หลายเดือนก่อน

    *promosm* 💯

  • @skitkatis4666
    @skitkatis4666 5 หลายเดือนก่อน

    Very cool,only women legs doesn’t look nice.

    • @nikolakakism
      @nikolakakism  4 หลายเดือนก่อน +1

      Indeed, the whole photo in the state shown in the video needs more attention. I already have addressed some issues like the legs that you mention, the floor that it is like they are floating on a grey background etc. In general, photos that are in bad shape, it is better to generate more ai images with low denoising strength to improve it step by step in a better and more controlled way.

    • @skitkatis4666
      @skitkatis4666 4 หลายเดือนก่อน

      @@nikolakakism thank you for your advice ,yes exactly, I had few old photos with many missing of bodies and agree with you