V2.5: F-r95 trained on lyreil 150 steps for 300 images merged whit merginmergesofmerge
0.45(95lyreil150) + 0.55(z_merginmergesofmerge_1).ckpt
V1: Is amixedd mix, of mixed mix models, (((Fr95+furryVixens_v10)+FR95)+FR95)+FR95)+FR95),
wait for the next updates, there will be an intermediate version before the next V3.5
Description
F-r95 trained on lyreil 150 steps for 300 images merged whit merginmergesofmerge
0.45(95lyreil150) + 0.55(z_merginmergesofmerge_1).ckpt
FAQ
Comments (9)
did you actually just make a mix, then mix the result with one of the original ingredients over and over and over? lol
I mean if it works out...you can see it as iterative improvements
With this model (V1.0) I always got better FR95-style images than with the model called FR95 Style
I'm curious about the new version now
@iGudeWie the idea is that the desired details are stronger, the Lyriel model is more realistic/anime has interesting colors and details, but both the mergin and the F95 suffered from overfit, I improved the database and tags, the next training it will be up the Flufusion merged with the Lyriel and then merged with the previous
Did you mean to publish this without safetensors?
i usually do everything in ckpt
Okay, carry on. Obviously there are those of us who try to observe some small amount of security hygiene and therefore won't use non-safetensors models, but I understand if that's not your concern.
@Harmil I intend to release the ly95 version based on the last version, separately this one was saved in safeternsors it has the same dataset as the f-95 lora if you want a version like this that much
what is HIo955 lora? i can't get that any of sites
it's a previous version of F-r95SFW lora