Guide Overall basic guide to using Stable Diffusion WebUI

  • Welcome to the Fakes / AI / Deepfakes category
    When posting AI generated content please include prompts used where possible.
Catflix.su banner

dkdoc

Superfan
Mar 11, 2022
63
519
639
0fya082315al84db03fa9bf467e3.png
How to make huge images in SD 1.5: I'll be converting 1 image to another using a lora of HeyImBee

  • Using txt2img
    • Find an image to reference, load it into 2 ControlNets
      • This generally helps the prompt out a little and helps to avoid mutations but comes at the expense of generating a completely original image
    • Load the image into Canny/Openpose, enable pixel perfect
      • Canny if you want to just copy the image
      • Openpose if you want to copy just the pose but replace everything else
    • Load the image into Depth, enable pixel perfect
      • self-explanatory, helps add depth to the image
  • Create your prompts and set the image size
    • Use 512 for the lowest dimension
  • Set the sampler to DMP++ 2M Karras
    • Set samples to 25
  • Enable Highres.fix
    • Upscaler: 4x_NMKD-Superscale-SP_178000_G
    • Highres Steps, 10
    • Denoising Strength, 0.4
    • Upscale by 2
The settings for Step 1:

10e6b415411669ce0.jpg


The resulting image:

29e6bddb72738e9a4.jpg


Looks okay, but it's only 1024x1824 and the teeth look bad. So let's fix it

  • Below the generated image, click "Send to inpaint"
    • Your prompts will be copied over. Copy the positive prompts and delete it, then replace it with what you need to fix. In this example, "mouth, teeth"
  • Set the sampling method to DPM++ 2M Karras
    • steps to 25
  • Leave Denoising strength to 0.75 (for now)
  • Change inpaint area from Whole picture to Only Masked
  • Paint the area that you need to fix
  • Generate an image
Before
3afd2109aa825a010.jpg

After
4bbe3718f45712eab.jpg

  • When you're happy after a few generations click "Send to img2img" under the preview generation
  • Change the Denoising Strength to 0.3 (do not change any other parameters)
  • in the txt2img tab, copy your positive prompt and paste it back in to the img2img tab
  • Scroll down to the bottom of the page
    • in the "Script" dropdown menu select "SD upscale"
      • Set the SD upscale Upscaler to 4x_NMKD-Superscale-SP_178000_G
      • Do not change any other parameters
  • Generate an image
From the original:
57cc2a4f9ef1ea0c8.jpg


tx2img with ControlNet Openpose + Depth and Highres.fix 512x912 to 1024x1824:
00017-214619222959a0eaeb3d27ddaa.png


To the final with img2img SD upscale finale size of 2048x3648:
00017-1325820688078172ae6cc0ed76.png


FAQ:
  • How do I install ControlNet / 4x_NMKD upscaler / SD upscale
    • Google is your friend
  • I'm getting Out of Memory errors in the txt2img / img2img steps!
    • Get a better GPU
  • Can I do this on my phone/tablet?
    • lol
 
  • Like
Reactions: LittleJ and surfing
Catflix.su banner