The Versions are not sorted by quality or whatever. It's simply the order in which I originally created the LoRAs the first time.
The LoRAs did not learn any faces
Body List:
https://mega.nz/file/Jf0xAbQD#Z9XN3eVGYSxrCdGNUhC5h5y1FrQrURaby8b-pmt6DbY
AI-Toolkit Job Config used for Version A
https://mega.nz/file/oHFEwYgI#jmkGule-6UF2ClA_mPrLLi7lf_wU4_IYYjGaXL9GYMw
Description
FAQ
Comments (5)
The way you trained version P is better than anything you've ever done before. Possible to do the same quality without the grain for future versions?
not sure what you mean, used the same settings for all versions, the only variable is the dataset. When you mean the grain from the showcase it's because of the prompt.
I love the concept of this, please keep doing what you're doing.
I watch too much porn that I can tell which porn stars most of these bodies belong too.
Thank you for your effort, this is truly a masterpiece. Can you make a version for Qwen Edit? Qwen's image editing is a bit clearer.I just want to use type M.
Details
Files
The_Body_Version_P_Flux.2.klein.9B.r16_AdamW8Bit_Weighted_woman_000005000.safetensors
Mirrors
The_Body_Version_P_Flux.2.klein.9B.r16_AdamW8Bit_Weighted_woman_000005000.safetensors
The_Body_Version_P_Flux.2.klein.9B.r16_AdamW8Bit_Weighted_woman_000005000.safetensors
The_Body_Version_P_Flux.2.klein.9B.r16_AdamW8Bit_Weighted_woman_000005000.safetensors
The_Body_Version_P_Flux.2.klein.9B.r16_AdamW8Bit_Weighted_woman_000005000.safetensors