Ostris(@ostrisai):Added an option to train @krea_ai Krea2 edit training with isolated reference-token attention. Training time is identical, but inference is 2-3x faster due to cached KV for references. LoRAs trained without it will still work, but will work better trained with it. 🧵

Ostris

@ostrisai

AI / ML researcher and developer. Creator of AI Toolkit - Support my work -

Joined August 2023

330 Following 14.7K Followers

Ostris@ostrisai

2026.07.09 18:31

Added an option to train @krea_ai Krea2 edit training with isolated reference-token attention. Training time is identical, but inference is 2-3x faster due to cached KV for references. LoRAs trained without it will still work, but will work better trained with it. 🧵