RUBY RELEASE
Ruby is a more detailed and realistic version of Amethyst. It can handle more easily with your prompt with no need to use detail LoRas (well u can if u want to). Please, if you like the model don't forget to review it. It helps me to improve more and more. U can leave suggestions too :). It can yet generate NSFW stuff you are wondering. All versions are different, not necessary an upgrade from the older. All versions comes from the original one (V1). Feel free to review it.
(PAST VERSIONS)
Jade
- Anime(ish) version of V1
Amethyst
- More NSFW support and more realistic
V1 - Initial Release
- The model can generate some nsfw (and sfw) material (female mainly)
Tips for usage:
- Use Adetailer for faces and hands. If you are making NSFW content, this will help you too.
- Don't need to use much of negative embeddings.
- The model works better with 9:16 image resolution
- Use 4x-UltraSharp for upscaling
- The best samplers for this model is DPM++ 2M SDE Karras and DPM++ 2M Karras (sometimes Euler a)
Description
More NSFW support and style changes
FAQ
Comments (4)
For the benefit of newbies like me, can we always mention what is the difference between the 4GB and the 2GB files?
Im not the greatest person to explain this but there we go:
The 4GB file is the full version of the model. You usually uses it to make your own model merges. You can generate images too if u want tho.
The 2GB file is the pruned version of the model, and it is what end-users want to use, because its more liteweight and faster to generate images.
These are almost (if not) identical when generating images, so if u just want to use the model for txt2img/img2img/inpaint etc, just download the pruned version.
The 4GB version still contains information from it's "construction", making it easier, if you wanted to train the model further or merge it with another one.
If it was a freshly constructed house, it would still have scaffolding around it and all the blueprints still lying around - nice to have, if you plan to extend/change the house, but unnecessary if you just work from inside the house.
The 2GB has that removed, making it hard to train the model further and merging models quite finicky, but leaving your GPU with more room to breath and the results should 99% be the same.
BTW, neither of the above. 2GB means it's been pruned (the duplicate EMA information removed) and the precision has been changed from floating point 32 to 16. You can still merge the 2GB models however you want with any other model, it will just lose the additional precision that 99% of people will not care about, because you generate images using fp16 in A1111 or any other codebase
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.