Model Card for Stably Diffused Models

Various Locon Models trained with lycoris

LuminousShaft v2.01 epoch-10

One of the more perfected runs of training on the anime for Luminous Witches.

Notes

• The model does not like EasyNegative, which generally is a thing for most style loras.
• The general tags for "deformed" (aka the generic negative tags) and such still works.
• Gen with 768 as the minimum resolution for either side.
• Images generated with based66_v30. but it will probably work with any NAI derived model.

Training Details

• Total images: 457
• Image Bucketing: Yes
• Epoch Trained: 15 (Selected, epoch 10)
• Resolution: 768
• network: lycoris (locon)
• Dims: 16
• Alpha: 8.0
• Conv Dim: 8
• Conv A: 0.3
• Steps: 4570
• Base model: NAIModel (That very model everyone knows lol)
• Text LR: 0.0001
• UNet LR: 0.0005

Youkan epoch-05

A model by trained on images drawn by youkan.

Notes

• "Probably can be anime..." is what past me said about the model...
• Tagged with "WD 1.4 Tagger" and "Deepbooru". Though personally I would never try that combo ever again.

Training Details

• Total images: 532
• Image Bucketing: Yes (?)
• Epoch Trained: 5 (?) (Selected, epoch 10)
• Resolution: 768
• network: lycoris (locon)

Note: Other details has been lost to time (It's inside the safetensors. I'm just lazy.)

FAQ

Q: Any Plans for SDXL?
A: Not at the moment. I would also like people to stop asking/harping me to use it. It's getting tiring to keep getting suggested to use it.

Q: Any Plans for SD2.1?
A: Tried it, still not entirely satisifed with the ecosystem for it. I would love people to use it more often though!

Q: CivitAI?
A: Probably it is me, but I personally don't like how CivitAI is all over for "Aesthetic" models.
As such I've decided it would be less worth to upload there and easier to just post in 1 place here. Assume anything not posted by KaraKaraWitch as fake or impersonations.