{ "id": 127862, "modelId": 52023, "name": "Erect Horse Penis v5", "createdAt": "2023-07-28T13:44:26.250Z", "updatedAt": "2023-07-28T14:49:15.781Z", "trainedWords": [ "eqcok", "Read Description!" ], "baseModel": "SD 1.5", "earlyAccessTimeFrame": 0, "description": "
Through some time Erect Horse Penis LoRA got many improvements, and this version is not an exception.
Whats better?
Greatly expanded dataset and adjusted training settings
Less style impact
Use (simple background:1.2)
in negatives for better result, if you're not going to do simple background images
Better penis shape and quality
Less artifacts and mistakes like \"double testicles\"
Better \"From Side\" view
Better (at least, working for 10% of images) \"From Behind\" view
Easier tip shape control
Blunt and heart/apple-shaped by defaut and with flared tip
in neagtives
Flared with flared tip
in positives
Less \"out of frame\" penis
Also, now penis not trying to take all the image space, except some horizontal images, because of lack in dataset, i guess.
Less \"piercing the sky\" penis, pointing straight upwards
Whats worse?
Temporaly removed \"autofellatio\" and \"autopaizuri\" concepts from LoRA due to many mistakes and problems overall. Need time to refresh it.
I've figured out, that overcomplicating with a lot of concept with penis (like tip shape, length, girth etc) would lead to the same effect, as character LoRAs. It would be harder to generated penis w/o those tags, and of course, it will lose it's consistency. So, i've decided to turn off some features and train it into basic concept.
Those features are:
Blunt tip for penis, now don't need an activation tag itself
Heart/apple-shaped tip for penis
This one may appear randomly while generations, looks pretty cool, as i think, but it really need more arts with the same shape to work properly.
Lenght, now depends on penis size