Elizabeth Olsen (HUNYUAN+FLUX+SDXL)

HUNYUAN v1.0 (Olsen in 2012) :

Trained on Hunyuan Video fp8 with 512x512 px with 58 photos of Elizabeth Olsen in 2012 with detailed GPT-4 captions. Tested on Hunyuan Video fp8 and Fast Hunyuan Video fp8 ! No keywords needed. Use around Lora strength=1.1.embedded_guidance_scale= 5.0-8.0 and flow_shift=7.0-12.0:

Positive : {Short summary of the scene e.g. Professional video of a blonde woman giving and interview on the red carpet}, {more detailed scene and background description}, {lighting description}, {camera direction e.g. panning in, panning out, zoom in etc.},<lora:eliolsen_2012_hunyuan_epoch60:1.1>

FLUX v.2 (Olsen in 2017) :

Please Donate Buzz for FLUX Lora Training !

Trained on FLUX.1 [dev] with 80 photos of Elizabeth Olsen in 2017 with detailed GPT-4 captions. Tested on FLUX 1.dev (full) and FLUX fp8 and FLUX nf4 ! No keywords needed. Use around strength 0.8-1.0. Distilled CFG 3.5 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:eliolsen_2017_local_164_merger_20v1_8v2_34v2_03_03_04:1>

FLUX v.1 (Olsen in 2012) :

Trained on FLUX.1 [dev] with 70 photos of Elizabeth Olsen in 2012 with detailed GPT-4 captions. Tested on FLUX 1.dev (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.8-1.0. Distilled CFG 3.5 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:Elizabeth_Olsen_2012_FLUX_epoch_46:1>

SDXL v8.0 2012:

Last versions probably. I wasn't satisfied with the likeness of the other models so I restricted all images to around 2012. Trained on Juggernaut X with 220 photos of Elizabeth Olsen in 2012. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3 and AlbedoBase 2.0! Use with keyword : "elxolsn" . Use around strength 1.0-1.1. CFG 5.0-7.0. Clipskip 1. 10-40 steps. Can be used for example as follows:

Positive : {Artstyle}, {Character and scene description}, elxolsn, <lora:eliolsen_2012_juggerX_xl_1_wocap-elxolsn-000120:1.05>

Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child

SDXL v7.0 :

Trained on Juggernaut X with 347 photos of Elizabeth Olsen. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3! Use with keyword : "elixolsen" . Use around strength 0.85-1.0. CFG 5.0-7.0. Clipskip 1. Can be used for example as follows:

Positive : {Artstyle}, {Character and scene description}, elixolsen, <lora:eliolsen_juggerX_xl_8_standard_wocap_merger_lastmodel_120_04_06-elixolsen:0.95>

Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child

SDXL v6.0 Juggernaut X :

Trained for Juggernaut X with 320 photos of Elizabeth Olsen. Use with Juggernaut X, Juggernaut v7 or RealismEngine 2. Works best with Juggernaut X. Use with keyword : "elixolsen". Can be used for example as follows:

Positive : {Artstyle}, {Character and scene description}, elixolsen, <lora:eliolsen_juggerX_v7_standard_wocap_merger_27_74_98_02_03_05-elixolsen:1>

Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child

SDXL v5.0 160mb :

Trained on 150 even more consistent (2015-2017) photos of Elizabeth Olsen with a smaller network dimension (32 instead of 64) to allow for smaller model size by half. Use with keyword : "elixolsen"!

SDXL v4.0:

Trained on 347 high quality photos of Elizabeth Olsen between 2015-2019 to ensure better facial consistency. Slight likeness improvement.

SDXL v3.0:

Lora trained on 199 images of Elizabeth Olsen and captioned by GPT-4 Vision. Lora strength should be between 0.8-1.2.

SDXL v2.0:

Lora retrained on 100 images of Elizabeth Olsen with SDXL 1.0 base and an improved Lora architecture to increase flexibility and likeness. The recommended strength on ComfyUi is 0.9-1.1.

SDXL v1.0:

Lora trained on 100 images of Elizabeth Olsen with SDXL 1.0 base. The recommended strength on ComfyUi is 0.9-1.1.