## French Manicure V2 LoRAUpdated version of <a rel="ugc" href="https://civarchive.com/models/2403307/french-manicure-lora">v1</a> trained with a cleaner dataset. Slightly improved performance (you can use heavier weight).A LoRA fine-tune of Flux.1 D trained on a curated dataset of french manicure v2 images.## UsageLoad this LoRA in ComfyUI using LoraLoaderModelOnly (trained on unet only) with strength 0.8u20131.0.## Training Details| Parameter | Value ||-----------|-------|| Base model | Flux.1 D || Network rank (dim) | 16 || Network alpha | 1.0 || Precision | bf16 || Steps | 1500 || Learning rate | 0.0004 || Optimizer | CAME || LR scheduler | constant || Training resolutions | mixed || Bucket training | No u2014 see dataset config || Weighting scheme | logit_normal || Timestep sampling | shift |## Compatibility- Base model: Flux.1 D- Node: LoraLoaderModelOnly (unet-only u2014 no text encoder weights)- Recommended strength: 0.8 u2013 1.0- Sampler: Euler + Simple scheduler, 20 steps- Guidance: FluxGuidance 3.5 - 10 (for Flux)

French Manicure V2 LoRA

<ul><li>Forgot to add the trigger word to captions in the last version. Fixed now.</li><li>Learning rate was also too aggressive. 4e-4 -&gt; 2e-4</li><li>New best Flux Guidance setting: 3.5-4.5.</li></ul>Trained 1500 steps on Flux.1 D. Rank 16, alpha 1.0, bf16 precision. Optimizer: CAME, LR: 0.0002. Trigger: <code>frnchmani</code>.

Files with SHA256: 728d585fa0611a417b2d3727c2ab8fbc5990bdbf0a934f22d47340482c062af2