prompt: erotic from below pov like, nipple showing, sweating redhead 30yo babe freckles (slut) smile/smirk riding bike at the gym, cleavage, (crowd spin class), Instagram photo|cinematic scene

negative: penis, sex, 3d, rendering, watermark, signature, logo, ugly hands, mouth open, motion

  • ColdCreasent@lemmy.caEnglish
    1·
    10 months ago

    Thanks for including the prompt. Would you be willing to share what model you are using?

    • Lumie@lemmynsfw.comOPEnglish
      9·
      10 months ago

      Mine is a custom merge, but you can get a similar result using SDXXXL(probably even better than mine); I just like to merge models. https://civitai.com/models/167764/sdxxxl?modelVersionId=253250

      Here are the other settings I used: Steps: 30, Sampler: DPM++ 2M SDE Heun Karras, CFG scale: 4.5, Seed: 2711655974, Face restoration: CodeFormer, Size: 1024x1280, Model hash: b190397c8a, VAE hash: 235745af8d, Version: v1.6.0-2-g4afaaf8a

      SDXXXL generates the following image.

      • MrGG@lemmy.caEnglish
        1·
        10 months ago

        This stuff is fascinating to me, but I’ve yet to dabble in it. What kind of hardware do you have / need?

        • Skyhighatrist@lemmy.caEnglish
          2·
          10 months ago

          The most important is your GPU. If you have a modern GPU you should be able to run stable diffusion locally. I’ve got an older GeForce 980 Ti, and I can run it it, but I did have to tweak the settings to lower the VRAM usage so that it didn’t crash all the time. Still crashes sometimes with larger models and higher steps and such. I also have to limit my images to 512x512 to keep it within my VRAM budget, but I’ve been able to get larger images by upscaling.

          I guess what I’m saying is that if you’ve got a better GPU than I do, you should get better results than I do.

        • Lumie@lemmynsfw.comOPEnglish
          2·
          10 months ago

          Check the pinned notes under https://lemmynsfw.com/c/aigen https://lemmynsfw.com/post/330

          I use a Ryzen 9 and Nvidia 4060 TI 16gb. Since I like to merge models I invested heavy in CPU/RAM/M.2 SSD with an “okay” GPU.

          If you are just going to download SDXL models and generate images. Then focus on GPU VRAM 8gigs+ more the better and make sure you have decent hard drive space. Each SDXL model is 6.7-13.5gigs in size.

          One other thing. Nvidia has the best graphics card support in the StableDiffusion space. AMD Radeon or Intel Arc will work but you might run into some headaches.