Version 1.0
Trained on fluffusion
Tested on FluffyRock, Fluffusion, Furtastic, and BB95 furry mix, but should work on any model that uses e621 tags
Use the tag penis milking machine to activate
Textual inversions/models/LoRA used: FluffyRock, Fluffusion, BB95 Furry Mix, Boring_e621, Useful Quality Embeddings,
Description
FAQ
Comments (13)
So... I'm a dumbass but I want to ask this stupid question anyway.
"Why you only make this furry based?"
I mean flat human futanari could be a good idea.
Because I only really use furry models, so that's what I trained it for. If i had trained on an anime or realistic model it wouldn't work very well on the models i use
@MewtwoBlueShoes Understandable. Sadly it's the only lora out there who does provide this setting :(
@lost_moon i have now made an anime version since a lot of people wanted one
@MewtwoBlueShoes fantastic. Thanks a lot.
I LOVE THIS!!!
Any advice for a poor, unfortunate soul who wants to try to train something like this for non-anthro content?
While i have now uploaded an anime version, making your own LoRA can be relatively simple. If you don't have a powerful enough computer (you'll need at least a 3060 12gb to train locally) google colab works surprisingly well, i personally recommend hollowstrawberry's LoRA training colab, as it's the most up-to-date and easy to use. For gathering images i recommend using Hydrus, it can be a bit advanced but it's by far the most feature-filled and streamlined option. For the rest i recommend reading "The other lora training rentry" since it delivers good in-depth explanations of everything (although i wouldn't trust it's advice on furry models)
You dont need 12GB VRAM, 8GB is fine aswell (some even train on 6 but pretty limited), but yea, recommended 12 GB, you can use holowstrawberry's colab if you want and i recommend his guide if you plan to train anime (tho it mostly works for photorealism too). As for the base model for training use animefull-final-pruned or base nai, since its compatible with most anime checkpoints and can even work in photo realism depending on the concept and how well is trained the LoRa
@Jelosus1 having tried training on a 3060 8 gb while it certainly works training with large-ish datasets can take days which makes any kind of experimentation really difficult. It is certainly possible though
@MewtwoBlueShoes Have you tried to use gradient checkpointing? It takes so much less VRAM, yes it makes training a bit slow but definetely worth it
@Jelosus1 I did a little bit but the speed penalty was too much. I've largely switched to using colab myself since i need to be able to use my computer and can't leave it to train for more than a few hours
@MewtwoBlueShoes Yes, that's fine
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.