From the announcement, Flux.3 Klein seems like both more capable and smaller model (~5b versus Flux.2 Klein at 9b).
In my experience, Klein, even though it does technically support text-to-image, is principally focused around editing. I had noticed significant text-to-image issues with Flux2 Klein 9B, particularly around anatomy with hands occasionally turning into lobster-like Cthulhu nightmare fuel - so I'll be interested to see if Txt2Img has improved.
With respect to editing, this new one appears to offer some really nice fine-grained control including the ability to notate bounding boxes around an object in the scene, then another bounding box for the target location, and prompt the model to move the object from the source rectangle to the destination.
From the announcement, Flux.3 Klein seems like both more capable and smaller model (~5b versus Flux.2 Klein at 9b).
In my experience, Klein, even though it does technically support text-to-image, is principally focused around editing. I had noticed significant text-to-image issues with Flux2 Klein 9B, particularly around anatomy with hands occasionally turning into lobster-like Cthulhu nightmare fuel - so I'll be interested to see if Txt2Img has improved.
With respect to editing, this new one appears to offer some really nice fine-grained control including the ability to notate bounding boxes around an object in the scene, then another bounding box for the target location, and prompt the model to move the object from the source rectangle to the destination.