Type | |
Stats | 250 673 32 |
Reviews | (26) |
Published | Jul 4, 2024 |
Base Model | |
Training | Steps: 6,000 Epochs: 3 |
Usage Tips | Clip Skip: 2 |
Trigger Words | high agle shot of a woman lying on the grass |
Hash | AutoV2 DDA25CD885 |
If you enjoy my work, please give me a "like" rating; it means a lot to me!
Check out other creations here
I don't speak English, but I wrote the explanation very carefully. It won't take more than two minutes to read. The following translation is from ChatGPT.
《LyingBack_v1.0》
SD3 can't do it, I'll do it [joke] and even SD1.5!
As we all know, the SD3 model has cut corners in body training, leading to severe body collapse in certain specific compositions.
As shown in the picture, SD3, a woman lying on the lawn in a top view.
This is my LoRa, it produces four images at once, with almost no breakdown, significantly increasing the probability of successful image generation!
This model is trained with over 300 high-quality images of people lying down in a top view, which not only reduces the chances of image breakdown but also improves the quality of the output images!
Now you can lie down however you like! The Ge You slouch is not just a dream! Let's be the lying flat youth together!
Use it in conjunction with my large model for even better results, click to go jian'tou→→→
Recommended parameters:
LoRa weight:0.1~0.9
Steps: 30
Sampler: Euler a
Schedule type: Automatic
CFG scale: 6
Hires. fix model:4x_NMKD-Siax_200k
Easter egg:
By the way, I also trained some pets to lie down...
Alright, this is just a joke. The body expression of SD1.5 was originally better than SD3, SD3's body is just terrible!
But this LoRa did make some optimizations. As for what optimizations were made, it depends on your feedback with images![joke]