I’ve merged FLUX.1-dev with fal.ai's Turbo LoRA and converted the weights into a fast int8 format.
flux2devturbo_int8_convrot.safetensors
flux2DevTurboInt8_v10.safetensors