Please Donate Buzz for FLUX Lora Training !
FLUX v2.0 :
Trained on FLUX dev with 90 photos (more carefully selected) of Margot Robbie (2015-2016) with detailed GPT-4o captions. Tested on FLUX 1.dev (full) and FLUX fp8 and FLUX nf4 ! Use around strength 1.0-1.1. 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:Margot_Robbie_FLUX_v2-000034:1>
FLUX v1.0 :
Trained on FLUX dev with 80 photos of Margot Robbie with detailed GPT-4 captions. Tested on FLUX 1.dev (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.8-1.0. 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:Margot_Robbie_2016_FLUX_v1-000037:1>
SDXL v6.0 :
Trained on Juggernaut X with 200 photos of Margot Robbie in 2016 with improved likeness due to consistency. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3, AlbedoBaseV2! Use with keyword : "mxrbbie" . Use around strength 1.0. CFG 5.0-7.0. Clipskip 1. Can be used for example as follows:
Positive : {Artstyle}, {Character and scene description}, mxrbbie, <lora:marobbie_2016_juggerX_xl_2_st_wocap-mxrbbie-000074:1>
Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child
SDXL v5.0 :
Trained on Juggernaut X with 165 photos of Margot Robbie with an improved Lora training method for better likeness and flexibility. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3, AlbedoBaseV2 and SDXXXL 3! Use with keyword : "mxrobbie" . Use around strength 1.0. CFG 5.0-7.0. Clipskip 1. Can be used for example as follows:
Positive : {Artstyle}, {Character and scene description}, mxrobbie, <lora:marobbie_juggerX_standard_xl_11-mxrobbie-000058:1>
Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child
SDXL v4.0 160mb :
Trained on 150 photos of Margot Robbie with a smaller network dimension (32 instead of 64) to allow for smaller model size by half while having slightly better likeness. Use with keyword : "maxrobbie"! Works well for Lora strength of around 1.0-1.1. General patter can be something like this:
positive prompt : {art style}, {action of the subject and characteristics},(maxrobbie), <lora:marobbie_32_small_xl_5_standard_wo_cap-merger-25_41_58_02_05_03-maxrobbie:1.05>
negative prompt : child, ugly, cleavage, blurry, etc...
SDXL v3.0 :
Trained on 270 more carefully selected images, captioned by GPT-4 vision and trained with an improved Lora model. The strength should be around 0.8-1.1.
SDXL v2.1 :
Trained an additional 4050 steps on the same 390 images
SDXL v2.0 :
Trained on 390 images of Margot Robbie. The ComfyUi weight should be kept at 0.95-1.15.
SDXL v1.0 :
Lora of Margot Robbie trained on 41 images of her with the SDXL base 1.0 model. The recommended strength for ComfyUi is 0.85-1.1.
