| This is a LORA for stable diffusion 1.5, that improves the generation quality at 4 steps. It uses direct backpropogation and HPSV2 reward scoring. | |
| It can generate decent quality images at only 4 inference steps. | |
|  | |
| Load with LCM https://huggingface.co/latent-consistency/lcm-lora-sdv1-5 | |
| Like this: | |
| ``` | |
| import torch | |
| from diffusers import LCMScheduler, AutoPipelineForText2Image | |
| model_id = "Lykon/dreamshaper-7" | |
| adapter_id = "latent-consistency/lcm-lora-sdv1-5" | |
| pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch.float16, variant="fp16") | |
| pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config) | |
| pipe.to("cuda") | |
| # load and fuse lcm lora | |
| pipe.load_lora_weights(adapter_id) | |
| pipe.fuse_lora() | |
| # load and fuse my lcm-lora-hpsv2 lora | |
| pipe.load_lora_weights("adams-story/lcm-lora-hpsv2-rl") | |
| pipe.fuse_lora() | |
| prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k" | |
| # disable guidance_scale by passing 0 | |
| image = pipe(prompt=prompt, num_inference_steps=4, guidance_scale=0).images[0] | |
| ``` |