Rukisan - LyCORIS LoCon - PonyDiffusionXLV6

LoCon trained on Rukisan's artstyle for ponydiff: https://civitai.com/models/257749

Artist: https://www.pixiv.net/en/users/94208188

Because of my lack of experience and time, the version number may be based on changes that have made certain aspects worse. As such a previous version may perform better than a later version.

I'm still learning the ins and outs of the model and LoRA training in general, please leave reproducible feedback or tips so that I may improve over time.

Check the "About this version" section for any quirks or pitfalls I'm aware of or have experienced.

Some Q's:

• What's your workflow?

  • Sampler is used with restarts with segments [3,2,0.06,0.30]

  • The scheduler used is Karras

  • Initial resolution is selected from the SDXL whitepaper appendix I: https://arxiv.org/pdf/2307.01952

  • Upscaler is usually 4x_fatal_Anime_500000_G or 4x_NMKD-Siax_200k, with this one in particular upscaling with lanczos also yields acceptable results

  • Second sample pass is the same as the first (params are on civit's meta interface) but at ~.35-.45 denoise

  • A facedetailer is used

  • I use ComfyUI, sorry

• What strength should I be using?

  • Base Pony: From my limited testing ~.8 - 1

  • Autismmix: ~.6 - 1

  • Other Pony merges: Depends on merge lora compat, but around autismmix levels should work

• Sample images Model?

  • All images are Autismmix_confetti

• What does your filename structure mean?

  • It's split into 4 parts of [XX]-MAnon-[YYYY]-[ZZZZZZ]

    • [XX] - Artist initials

    • MAnon - My moniker

    • [YYYY] - SD version and Model version (Ex. XLV4.1 is on SDXL [Not 1.5] and it's version 4.1 of the model)

    • [ZZZZZZ] - The epochs, only useful to me

Explanation of Issues:

• Western cartoon influence

  • The artist tends to draw all characters with big heads and expressive faces, even including ones that should really not. As a result this training tends to lean strongly towards big heads. You can partially remedy this by using "chibi" in the negs at 1+ weight but it's pretty baked in and my attempts to caption it out haven't been successful.

• Jagged, crusty linework

  • For the most part the linework is satisfactory, but at points it can be uneven, lumpy, or have inconsistent coloring. I always recommend upscaling, but for this one in particular I recommend upscaling. Tiled upscaling also smooths it out more than usual so it may be something to consider. I think it's the result of downscaling the training images but there's only so much I can do about that with the free time I have.

• Issues with angles

  • If you're doing from above/from below/high-angle/low-angle weigh them higher than usual, even then it'll probably fight you

• Facial linework has a slight white outline

  • Sometimes the lines around the eyes/mouth I've noticed will have a white tinge around them, which is exacerbated by upscaling beyond 1.5x scale. They're usually easily fixable in paint/ps with a stamp tool but other than than it's something you'll have to live with.