<h2 id="please-donate-buzz-for-flux-lora-training-!-lgh974fxb">Please Donate Buzz for FLUX Lora Training !</h2><h3 id="flux-v.3-(2000-2004)-:-xq9gwqx0l">FLUX v.3 (2000-2004) :</h3>Trained on FLUX.1 [dev] with 85 photos of Natalie Portman in 2000-2004 with detailed GPT-4o captions and square 1024px ratios. Tested on FLUX <a target="_blank" rel="ugc" href="http://1.dev">1.dev</a> (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.9-1.2. 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},&lt;lora:Natalie_Portman_Squared_FLUX_v3_merger_31_52_61_02_05_03:1.1&gt;<h3 id="flux-v.2-:-qovppdweu">FLUX v.2 :</h3>Trained on FLUX.1 [dev] with 84 photos of Natalie Portman with GPT-4 captions. Tested on FLUX <a target="_blank" rel="ugc" href="http://1.dev">1.dev</a> (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.8-1.2. 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}, &lt;lora:natportman_2004_local_164_v2-000005:1.0&gt;<h3 id="flux-v.1-(portman-in-2004)-:-unsw5z2dh">FLUX v.1 (Portman in 2004) :</h3>Trained on FLUX.1 [dev] with 100 photos of Natalie Portman in 2004 with GPT-4 captions. Tested on FLUX <a target="_blank" rel="ugc" href="http://1.dev">1.dev</a> (full) and FLUX fp8 and FLUX nf4 ! Use around strength 0.8-1.2. 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}, &lt;lora:Natalie_Portman_2004_FLUX_epoch_16:1.1&gt;<h3 id="sdxl-v6.0-2004:-qjy33t3n8">SDXL v6.0 2004:</h3>Trained on Juggernaut X with 230 photos of Natalie Portman from 2003-2004. Tested on Juggernaut X, Juggernaut v7, RealismEngine 2, RealVisXL3 and AlbedoBase 2.0! Use with keyword : "ntxprtman" . 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, &lt;lora:natportman_2003_juggerX_xl_1_wocap-natxprtmn-000149:1.0&gt;Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child<h3 id="sdxl-v5.0-juggernaut-x-:-nnvpzag1n">SDXL v5.0 Juggernaut X :</h3>Trained for Juggernaut X with 270 photos of Natalie Portman. Use with Juggernaut X, Juggernaut v7 or RealismEngine 2. Works best with Juggernaut X. Use with keyword : "natxportman". Can be used for example as follows:Positive : {Artstyle}, {Character and scene description}, natxportman, &lt;lora:natportman_gpt4_juggernautX_2_wocap-merger_21_65_83_04_02_04-natxportman:1.0&gt;Negative : ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child<h3 id="-82ng37kna"></h3><h3 id="sdxl-v4.0-80mb-:-vcxx9nieq">SDXL v4.0 80mb :</h3>Trained on 137 photos of Natalie Portman with an improved Lora model which has a much reduced size while maintaining likeness. Put the keyword : "natxportman" at the beginning of prompts! Use Lora with strength around 1.1.<h3 id="sdxl-v3.1:-l4o7s8o71">SDXL v3.1:</h3>Trained on a subset of the 200 images dataset from v3.0 but that is better balanced and has better captioning (more refined GPT-4 Vision prompt). Lora strength between 0.8-1.2.<h3 id="sdxl-v3.0:-iiypog86a">SDXL v3.0:</h3>Retrained on 200 images but with an improved Lora training method and model to yield higher flexibility while preserving likeness and GPT-4 Vision captioning. Keep weight at 1.05-1.1.<h3 id="sdxl-v2.0:-0fty5r4yf">SDXL v2.0:</h3>Lora trained on 270 images of Natalie Portman with SDXL 1.0 base. The lora is also tested on Juggernaut XL 3.0. Most of the sample images were created with Juggernaut other than the "...2004..." generated ones.<h3 id="sdxl-v1.0:-mfliigeu7">SDXL v1.0:</h3>LoRa trained on 45 images of Natalie Portman with SDXL 1.0 base, The recommended strength for ComfyUi is 1.1.

Natalie Portman (SDXL + FLUX)

Slight improvement with following prompts better. Trained on a subset of the original dataset that is better balanced. Lora strength between 0.8-1.2.

natportman_2_xl_5_standard-merger_46_51_58_90_02_03_03_02.safetensors

v3.1

FLUX v3.0 (2000-2004)

FLUX v.2

FLUX v1 (Portman in 2004)

v6.0 (Portman in 2004)

v5.0 Juggernaut X

v4.0 80mb

v3.0

v2.0

v1.0

Files with SHA256: 387329b20600c4e029b43ac061ee78bd9ff9eea405fada1dfc59ad1f38fb7ea7