{"id":1276,"date":"2023-07-08T17:07:51","date_gmt":"2023-07-08T21:07:51","guid":{"rendered":"https:\/\/home.woodchuckhunters.com\/blog\/?p=1276"},"modified":"2023-07-11T12:03:00","modified_gmt":"2023-07-11T16:03:00","slug":"my-ai-photo-generating-process","status":"publish","type":"post","link":"https:\/\/home.woodchuckhunters.com\/blog\/2023\/07\/08\/my-ai-photo-generating-process\/","title":{"rendered":"My AI photo generating process"},"content":{"rendered":"\n

I’m trying for photorealism, but the only card I can make Stable Diffusion work on is an ‘ancient’ AMD R9 200 series video card with 3GB of VRAM so the resolution is incredibly limited. Still, there’s something magical (and oddly addicting) about being able to render MySpace-like images of fake or imaginary women. It’s as if I\u2019m able to suck a photo off someone else\u2019s phone through an inter-dimensional rift. It\u2019s definitely hitting a dopamine trigger in my reptile brain.<\/p>\n\n\n\n

Regardless, sometimes when I post images on Reddit or elsewhere, people ask what my process is and I wanted to get something written down for posterity. So here’s the basic thing that works for me…<\/p>\n\n\n\n

\n
\n
    \n
  1. First I use an ad-sponsored Stable Diffusion app on my phone that quickly generates tons of “Anime” images and I use terms like “highly detailed” and “realistic” which tends that model toward images that are closer to photorealism.<\/li>\n\n\n\n
  2. I probably generate 10s of images with prompts like “woman in pink dress” to get even a single image that pops out and demands some more detail…<\/li>\n<\/ol>\n<\/div>\n\n\n\n
    \n
    \"\"
    anime model rendering of “woman in pink dress”<\/figcaption><\/figure>\n<\/div><\/div>\n<\/div>\n\n\n\n
    \n
    \n
      \n
    1. Next I put that image into the img2img section of SD Automatic1111<\/a> on my workstation and try to bring that person to life with a model like the “Edge of Realism”.\n
        \n
      • Approximate settings on my AMD R9 3GB VRAM:\n
          \n
        • 0.5 MP resolution around 512×904 <\/li>\n\n\n\n
        • CFG between 10 and 20<\/li>\n\n\n\n
        • Denoise between 0.45 and 0.55<\/li>\n\n\n\n
        • Enable Tiled VAE; encoder size 480, decoder size 48<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n\n\n\n
        • Before I click “Generate”, I write a prompt from scratch with details that stand out.
          This image had a prompt like,
          “brunette woman in satin metallic dress leaning back against apartment wall”<\/li>\n<\/ol>\n<\/div>\n\n\n\n
          \n
          \"\"
          brunette woman in satin metallic dress leaning back against apartment wall<\/figcaption><\/figure>\n<\/div><\/div>\n<\/div>\n\n\n\n