Type | |
Stats | 924 1,766 |
Reviews | (175) |
Published | Aug 31, 2024 |
Base Model | |
Training | Steps: 749 Epochs: 16 |
Usage Tips | Clip Skip: 2 Strength: 0.8 |
Trigger Words | Pr1nc3ssC@rry, fr0ntC@rry, 1boy, first person, lying on arms, facing viewer, dutch angle, 1girl, |
Hash | AutoV2 51F6A83C65 |
Another LoRA trained on a synthetic dataset straight out of the GTX 1650 oven. This one aims to introduce the concept of carrying your waifus in POV perspective. This pose is pretty much impossible to generate with prompts, and even the dataset for this is excessively rare.
I will start off by letting everyone know that this one can get wonky sometimes with the hands and the pose, and the LoRA in itself is a bit hard to use for perfect results. Follow the prompting technique given below for some consistency in your results, or just experiment yourself. I will test it more extensively later and update this. In general, it does not seem to perform well with some character LoRAs.
Also, prompting any tags that describe the behind of the girl can ruin the pose most likely. Try only including frontal upper body features.
The resolutions that work best with this are: 1344 x 768 (most consistent), 1216 x 832 (sometimes gets the face in the middle), and other landscape aspect ratios. Make sure that you don't use portrait resolutions (NOT tested). If you have any doubts, feel free to discuss in the comments.
Start with strength 0.75, go upto 0.9. Going higher can change the style, going lower is not tested. Set clip skip to 2 (or don't), as this was trained with clip skip 2. If character face gets in the middle, reduce the weight of the character LoRA by 0.5.
Add "blur, smudged" to your negative prompt. (optional)
Start your prompt with the LoRA trigger words: "Pr1nc3ssC@rry, fr0ntC@rry, 1boy, first person, lying on arms, looking at viewer, dutch angle, 1girl,"
Then, include your character trigger words (example): megumin, embarrased smile,
Finally, add the background descriptors: outdoors, near lake, sunset, cloudly skies, golden hour
This template is purely for reference and you can always come up with a better way than mine.
If you are using A1111/Forge, and the LoRA does not show up, you can force it to appear by enabling the "Always show all networks on the Lora page" option on the settings page.
If you want to support more models like this, or you just like what you see: ko-fi (｡♥‿♥｡)
Also, I don't plan on charging buzz for my LoRAs. So, I'd really appreciate if by any chance you could tip the model.