AboutI proudly introduce you to Famira, my first Flux Lora that is based on a set of merged faces, which turns out to be a very average Asian woman. Flux.1 Dev is really astonishing, but it still contains render flaws. I am not using the best Workflow setup, but hey... maybe you guys may do better.TrainingFirst time I rendered a Lora was on <a target="_blank" rel="ugc" href="http://Fal.ai">Fal.ai</a>, but due to limited parameters, the results was really nothing like the Civitai Lora trainer. At <a target="_blank" rel="ugc" href="http://Fal.ai">Fal.ai</a> my model was probably also overtrained. It seems like for Lora faces it requires about 2000-ish steps.Famira is trained with 45 (square 1024) with (5 repeats, 4 batches), which means... 45 x 5 repeats / 4 batches x 36 epochs = 2025 steps.About rendering with Flux.1 DevWell, what can I say, they really hate cigarets. I tried to render Famira smoking, but still the hands are an issue. Probably requires different Lora. For the rest, I really hate the fact that it is missing negative prompts. Maybe I am doing something wrong, in order to fix that, I sometimes have to use false positive prompt.If you like my work, don't hesitate to leave a like or a review. Please share your renders so I gain knowledge what other potentials this model can be used for.Last but not least, I drink a lot of <a target="_blank" rel="ugc" href="https://ko-fi.com/zaka">Ko-Fi</a> :)

Files with SHA256: 99d502fa3c757fb42bab186ffa55fc530067ec828a967dbd07e8fac013e83ecb