Implement Flux 2 for Model Trainer
E
Eamon
1) Flux 2 is superior in every metric and staying outdated is bad.
2) A Flux 2 Model Trainer would allow fine-tuning that the Reference Image options do not allow, and allows things they don't allow.
3) A lack of modern model trainer punishes the traditional method of making AI art and forces people to use a Reference Image option they may not want or find as useful.
A
Ava Bennett
It’s great to see the community pushing for Flux 2 integration. Given how much the original Flux model streamlined the generation process, a second iteration for the model trainer could really be a game-changer for consistency—especially when it comes to maintaining structural integrity in complex renders.
I’ve been following these updates closely because getting high-fidelity results with minimal artifacts is crucial for my current workflow. Whether it’s for architectural visualization or refined digital assets, the precision of the training model dictates the quality of the final output. For those of us focused on high-end results, like my team at Quality Renovation, having a more robust trainer would significantly reduce the post-processing time we currently spend on image correction.
Does anyone know if the dev team has mentioned a specific timeline for the beta, or if they are looking for specific datasets to help refine the Flux 2 training logic? I'd love to see how it handles fine-grain textures compared to the current version.