Type | |
Stats | 423 592 3.4k |
Reviews | (78) |
Published | Aug 17, 2024 |
Base Model | |
Training | Steps: 294 Epochs: 5 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Hash | AutoV2 3BE3213FE5 |
This is my first LORA model I've ever made, but I think it turned out pretty well. Since I have no idea what I'm doing, I'll explain how I use it to achieve such effects. The model was trained on a collection of dozens of my own works in such a way that I could present the Cyber-Pet concept from different sides in the horizontal plane using descriptions such as "side view", "side angled", etc. The model was trained on different Cyber-Pet colors and different degrees of cyberretization with different heads from fully human to completely artificial android masks. Describing it in natural language, these effects are easily achievable.
The model was also trained on examples where the Cyber-Pet is led on a leash or is accompanied by a caregiver, who for the purposes of training the model was in three variants "owner", "mistress" and "maid". However, the option with a leash does not always work because the model tends to confuse it with a tail. The model was also trained with a fairly accurate arrangement of limbs in the variants "bent" and "straightened" "walking" etc. and with various hands, paws, claws, so it should respond to such prompts.
Generally, the best effects come with long, extensive prompts describing many details because all input images were described quite accurately and meticulously, which cost me quite a lot of work.
The almost mandatory prompts for this model are "on all fours", "cyber-pet", "pows", etc.