Please Donate Buzz for FLUX Lora Training ! This one cost over 4000 Buzz to train.
FLUX v1.0 (Late 1990s) :
Trained on FLUX dev with 96 photos of Actress Liv Tyler in the late 1990s with detailed captions (GPT-4o). Tested on FLUX 1.D (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.9-1.2. No keywords needed! Distilled CFG around 1-4 and CFG 1.0 (without negative prompt). Clipskip 1. Can be used for example as follows:
Positive : {Artstyle, Character and scene description in usual FLUX fashion},<lora:Livtyler_FLUX_v1-merger_merger33_73_merger45_73_03_07:1.05>
SDXL v1.0 :
Trained on Juggernaut X with 230 photos of actress Liv Tyler in the late 1990s. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3 and AlbedoBaseV2 ! Use with keyword : "livxtylr". Can be used for example as follows:
Positive : {Artstyle}, {Character and scene description}, livxtylr, <lora:livtyler_juggerX_xl_1_standard_wocap-livxtylr-000069:1>
Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child
Description
FLUX v1.0 (Late 1990s) :
Trained on FLUX dev with 96 photos of Actress Liv Tyler in the late 1990s with detailed captions (GPT-4o). Tested on FLUX 1.D (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.9-1.2. No keywords needed! Distilled CFG around 1-4 and CFG 1.0 (without negative prompt). Clipskip 1. Can be used for example as follows:
Positive : {Artstyle, Character and scene description in usual FLUX fashion},<lora:Livtyler_FLUX_v1-merger_merger33_73_merger45_73_03_07:1.05>
FAQ
Comments (3)
Flux version looks stunning, XL version looks a bit off, maybe training on juggernaut is the issue for XL version ? I was under the impression that you had to train on base SDXL model for best result overall, or SDXL vae fix (one of the base version anyway) maybe I'm wrong but I think it best practice for most consistent results
It's generally much harder to train on SDXl models. The likeness is quite hard to get.
Thank you!