Core ML Converted Model:
- This model was converted to Core ML for use on Apple Silicon devices. Conversion instructions can be found here.
- Provide the model to an app such as Mochi Diffusion Github - Discord to generate images.
split_einsum version is compatible with all compute unit options including Neural Engine.
original version is only compatible with CPU & GPU option.
- Custom resolution versions are tagged accordingly.
vae tagged files have a vae embedded into the model.
- Descriptions are posted as-is from original model source. Not all features and/or results may be available in CoreML format.
- This model was converted with
vae-encoder for i2i.
Liberty:
PLEASE READ DESCRIPTION
Extremely NSFW biased model!!! But awesome at SFW too. Use it at your own risk of getting unprompted explicit images.
This is not either an easy to prompt model. Not the best one for begginers.
According to one user: "If you have the model-keyword extension make sure you uncheck it. It appends all the triggers words and results in complete nonsense results. Was getting super frustrated that even copy/pasting prompts gave me trash".
MODEL (there are two possibilities):
Liberty-Main: You should be using this model. It might not be the easiest one to prompt depending on your style. But you'll get used to it soon and is the most powerful one.
Liberty-BadClip: This version uses a broken CLIP model in the same style aEros CLIP model was broken, so outputs are very different from main version. If you really know what you are doing, or you really don't want to change your prompting style from aEros and it performs bad in main version, or are getting general bad results with main one, etc... you could try this one. Keep in mind that some UIs have problems with broken CLIPS. Also that I won't continue or support this version of the model.
VERSIONS (both models have 3 versions):
Standart: Intended for general use. It has VAE vae-ft-mse-840000-ema-pruuned.vae.pt baked in. There's no need to use any additional file.
Inpainting: Intended to use in the inpainting section of the img2img tab, it is mixed with original SD 1.5 inpainting model and is much more coherent whith masking areas. Only select it for that purpose. There's no need to use any additional file.
Training: Intended to use in the Train tab, for embedding creation. It uses SD 1.5 original VAE baked in, so it doesn't deepfries embeddings. Only select it for that purpose. There's no need to use any additional file.
All of the files are provided in ckpt and safetensors format for your convenience ;)
ABOUT:
Liberty is freedom. A merge with over 23 other models with a methodical, careful and genuine approach. Check 'CREDITS' section for the full list.
Freedom of prompting art or photo or both, landscapes or backgrounds or interiors, people or entities or scenes, stiff poses or movement or even mouth and facial emotions, SFW or nudes or even hardcore sex. I tried to make it as versatile as possible and merged it with half CivitAI to get the most out of any free model out there.
And it is free to use for any open source purpose, commercial or not. All the models I used were licenseless when I grabbed them, although some have changed licenses afterwards. However it is also much more modular in it's development process, which means that if any problem arised, I could rebuild it much quickier to avoid getting chained to a license or other type of problems.
HOW TO USE:
This model does not need to use any of it's trigger words. They are a tool, and the knowledge of those trigger words is deeper known and better accesed directly. So 'a photo of a cyborg woman' will work better than 'cyborgdiffusion, photo of a woman', but you can also try 'cyborgdiffusion, photo of a cyborg woman' or even 'cyborgdiffusion, photo of a knight woman' for special and unique effects. If you want to get an idea of what they can help to achieve visit the 'CREDITS' section to check the models they come from.
I mainly try natural language for my tests. You have an example prompt screenshot avaible to build from there if you want to. There's a more advanced prompting guide at my Patreon:
https://www.patreon.com/posts/aines-prompting-77121972It is merged with UnstablePhotoreal (but with much less weight than aEros), and it reacts well to their prompting language, you don't need to use it as comma separated tags:
https://docs.google.com/document/d/1-DDIHVbsYfynTp_rsKLu4b2tSQgxtO5F6pNsNla12k0/editAs with my other models I have tested most intensively how good it is at making erotic and artistic nudes. But this time I have tried many other types of prompts too, some of them borrowed. But keep in mind that some uses are edge case. For example it is not so good at hardcore sex scenes as far as I know, you'll get more consistent and effective sex scenes at other specialised models instead of Liberty. And it won't make the best anime art out there either.
It may not be the easiest model to use, give it some time to get used to it :)
CREDITS (Many thanks in no particular order to all of the authors of this awesome models, part of Liberty, without them this wouldn't exist):
SynthwavePunk V2 (do not mistake it with the licensed one):
https://civitai.com/models/1102/synthwavepunkLet's all have fun and play together!
