NahidaDiffusion

[NahidaDiffusion @65ythr467]

SwampMachine-V2.0
https://t.co/jgDdC0o1na

[NahidaDiffusion @65ythr467]

SwampMachine-V1.0
https://t.co/5EDSO5X2N8

[NahidaDiffusion @65ythr467]

#AnimagineXL Great base model!
Trained with images of less than 10k, but with excellent results.

[NahidaDiffusion @65ythr467]

An experiment in which each TE is trained with different captions in FT.

1.L:characters tag
G:general tags
2.L:characters tag
G:characters tag+general tags
3.normal training

The method of giving one TE the perfect captions and the other only meta tags may be effective.

[NahidaDiffusion @65ythr467]

NovelAI V3 is kinda quirky. Are they training separate text encoders for general tags and for meta tags and other tags, including copyrights?🤔

[NahidaDiffusion @65ythr467]

a small update
https://t.co/3PjPy4MUtT

[NahidaDiffusion @65ythr467]

Due to some remaining issues, there may be updates to this model at a later date.

[NahidaDiffusion @65ythr467]

CounterfeitXL-V2.0
https://t.co/BLZIUBtCVh

[NahidaDiffusion @65ythr467]

test

[NahidaDiffusion @65ythr467]

In CounterfeitXL-V2.0, you can specify styles using the specialty tag.

[NahidaDiffusion @65ythr467]

New SDXL model.
Training now!

[NahidaDiffusion @65ythr467]

CounterfeitXL-V1.0
https://t.co/3PjPy4MUtT

[NahidaDiffusion @65ythr467]

Improved from beta, but no major changes.
Verson 1.0 is still to be released.

[NahidaDiffusion @65ythr467]

CounterfeitXL-β
https://t.co/3PjPy4MUtT

[NahidaDiffusion @65ythr467]

Once the retraining of the TI is finished up, I'll be releasing the β version.

[NahidaDiffusion @65ythr467]

β🤔

[NahidaDiffusion @65ythr467]

@Hosiokaa Yes, images from Nijijourney are part of the dataset. After a large-scale training, you might need to prepare a smaller dataset. This model was trained exclusively on Unet.

[NahidaDiffusion @65ythr467]

CounterfeitXL
https://t.co/VV6v5bRneh

[NahidaDiffusion @65ythr467]

Merge attempts with WDXL didn't pan out. Maybe extensive training isn't needed, or perhaps using WDXL as a training foundation could yield better results.
#SDXL