Type | |
Stats | 477 2,675 |
Reviews | (58) |
Published | Aug 8, 2023 |
Base Model | |
Training | Steps: 3,870 Epochs: 15 |
Trigger Words | claws |
Hash | AutoV2 CB17F29861 |
This is the first time I am trying to train an object with which we should interact. I spent many days experimenting on this one. I couldn't find enough people discussing about how to properly train such object(it is especially a nightmare to deal with hands, so much harder than faces). Using my normal training settings, the model just cannot target what I want to train and result in bloody mess.
I have generated some right claws in txt2img, but I imagine this will be more useful in Inpainting. This Lora is far from perfect, in my testing, about one out of five times it will work properly(in Inpainting). I have retrained many times to get this model done (to the extent of at least useful in some cases). If it is useful to you, I would be more than happy to hear that.😊
If someone can suggest me some good tips or tutorial to train a specific body part or a small object that we can interact with, please leave me in the comment below. 🙏
This model is trained on Photon.
When you are using Inpainting, some tips may be useful.
Use ADetailer to make sure hands are ok
Sometimes it would be more natural if you mask out entire forearm, but it depends on what you want to change.
Leave some rooms for claws to grow.
Use slightly lower CFG Scale and DDIM sampler.(You can experiment with the settings)
You may want to put "muscle, arm hair" in the negative prompt.
Entirely white background is not good for this lora.
One hand at a time is highly suggested.
If you guys generated some good images, please share them below, they can help me improve the model. And if they are in high resolution, can even help me build a SDXL version.🙏
If you like this model, please consider leaving a like. Or you can buy me my next meal here. Thanks.❤️