[mathi0750]

Hey folks, With FLUX.1 Kontext [dev] dropping yesterday, we're comparing prompting it vs a fine-tuned FLUX.1 [dev] and PixArt on generating consistent characters. Besides the comparison, we'll do a deep dive into how Flux works and how to fine-tune it. What we'll go over: * Which models performs best on custom character gen. * Flux's architecture (which is not specified in the Flux paper) * Generating synthetic data for fine-tuning examples (how many examples you'll need as well) * Evaluating the model before and after the fine-tuning * Relevant papers and models that have influenced Flux * How to set up LoRA effectively This is part of a new series called Fine-Tune Fridays where we show you how to fine-tune open-source small models and compare them to other fine-tuned models or SOTA foundation models. Hope you can join us later today at 10 AM PST! https://lu.ma/fine-tuning-friday-3