LCM-LoRA
https://arxiv.org/abs/2311.05556
LCM-LoRA: A Universal Stable-Diffusion Accelerati
https://github.com/huggingface/blog/blob/main/lcm_lora.md
SDXL in 4 steps with Latent Consistency LoRAs
LCM-LoRAモデル
https://huggingface.co/latent-consistency/lcm-lora-sdxl
latent-consistency
/
lcm-lora-sdxl
https://huggingface.co/latent-consistency/lcm-lora-sdv1-5
lcm-lora-sdv1-5
https://huggingface.co/latent-consistency/lcm-lora-ssd-1b
lcm-lora-ssd-1b
https://note.com/gcem156/n/nb23aff723431
LCM-LoRAについて by
gcem156
LCM-LoRA適用させるだけで既存のモデルLCM化できるってことぉ?
nomadoor.icon
https://i.gyazo.com/6f6ee9b5570b0444e3942c9e2e222145.gif
LCM-LoRA.json
楽しい
nomadoor.icon
けどDPM++ 2M Karrasの15stepsで0.95s(CFG5で1.05s)、LCMで0.4sなので性能捨てて2倍早いくらいだと微妙か…?
関連
生成速度を2倍にするLoRA
for SDXL
LCM
画像生成の高速化