Type | |
Stats | 378 |
Reviews | (55) |
Published | May 23, 2024 |
Base Model | |
Usage Tips | Clip Skip: 1 |
Hash | AutoV2 4413017AD7 |
Please share and post your results. I am looking forward to seeing your creations! :)
This is the end result of mixing many models to get the style I was looking for. Specifically a versatile, highly detailed, people/portrait focused model. It understands lighting direction to a good degree, along with camera techniques and cues such as blurry background, depth of field, golden hour etc. It can produce SFW and NSFW and doesn't lean hard on NSFW unless prompted. the model also produces detailed backgrounds and quality proportions. This is intended to be the Semi Realistic version of My Personafied Model.
After about 100 different merge and mixes It was created as a mixture of 3 models. My own personal model and 1 realistic and 1 semi realistic model. I have begun experimenting with using generated images to train lora.
What's this model
"Persona Fied Mix"is a PERSONAification of Shigenori Soejima's art style and an Aesthetic I wasn't able to get with any single model. I think it is the right blend to produce some captivating images with depth, detail, and style.
The starting point for this model was my Persona Mix Hard. From there, I then added a few of the models which gave me a level of quality and a close style to the Persona one I was looking for. I tried many of the persona characters with different weights like .75, .8, 1.0 all with great results.
Using this model
Clipskip 1 or 2, with Vae-ft-mse-840000-ema-pruned.pt
Patreon is pending. Support is appreciated via
TIP:
I like to put the username in Parenthesis (LJA) when I download Lora's and models, etc. to be more organized and to find the source content again. I advise this as best practice.
and it gets confusing otherwise.