import torch import gradio as gr from functools import partial from diffusers_patch import OMSPipeline def create_sdxl_lcm_lora_pipe(sd_pipe_name_or_path, oms_name_or_path, lora_name_or_path): from diffusers import StableDiffusionXLPipeline, LCMScheduler sd_pipe = StableDiffusionXLPipeline.from_pretrained(sd_pipe_name_or_path, torch_dtype=torch.float16, variant="fp16", add_watermarker=False).to('cuda') print('successfully load pipe') sd_scheduler = LCMScheduler.from_config(sd_pipe.scheduler.config) sd_pipe.load_lora_weights(lora_name_or_path, variant="fp16") pipe = OMSPipeline.from_pretrained(oms_name_or_path, sd_pipeline = sd_pipe, torch_dtype=torch.float16, variant="fp16", trust_remote_code=True, sd_scheduler=sd_scheduler) pipe.to('cuda') return pipe class GradioDemo: def __init__( self, sd_pipe_name_or_path = "stabilityai/stable-diffusion-xl-base-1.0", oms_name_or_path = 'h1t/oms_b_openclip_xl', lora_name_or_path = 'latent-consistency/lcm-lora-sdxl' ): self.pipe = create_sdxl_lcm_lora_pipe(sd_pipe_name_or_path, oms_name_or_path, lora_name_or_path) def _inference( self, prompt = None, oms_prompt = None, oms_guidance_scale = 1.0, num_inference_steps = 4, sd_pipe_guidance_scale = 1.0, seed = 1024, oms_prompt_flag=True, ): pipe_kwargs = dict( prompt = prompt, num_inference_steps = num_inference_steps, guidance_scale = sd_pipe_guidance_scale, ) generator = torch.Generator(device=self.pipe.device).manual_seed(seed) pipe_kwargs.update(oms_flag=False) print(f'raw kwargs: {pipe_kwargs}') image_raw = self.pipe( **pipe_kwargs, generator=generator )['images'][0] generator = torch.Generator(device=self.pipe.device).manual_seed(seed) pipe_kwargs.update(oms_flag=True, oms_prompt=prompt, oms_guidance_scale=1.0) print(f'w/ oms wo/ cfg (consistent) kwargs: {pipe_kwargs}') image_oms_cp = self.pipe( **pipe_kwargs, generator=generator )['images'][0] if oms_prompt_flag: generator = torch.Generator(device=self.pipe.device).manual_seed(seed) pipe_kwargs.update(oms_prompt=oms_prompt) print(f'w/ oms wo/ cfg (inconsistent) kwargs: {pipe_kwargs}') image_oms_icp = self.pipe( **pipe_kwargs, generator=generator )['images'][0] else: image_oms_icp = None oms_guidance_flag = oms_guidance_scale != 1.0 if oms_guidance_flag: generator = torch.Generator(device=self.pipe.device).manual_seed(seed) pipe_kwargs.update(oms_guidance_scale=oms_guidance_scale) print(f'w/ oms +cfg (inconsistent) kwargs: {pipe_kwargs}') image_oms_cfg = self.pipe( **pipe_kwargs, generator=generator )['images'][0] else: image_oms_cfg = None return image_raw, image_oms_cp, image_oms_icp, image_oms_cfg, gr.update(visible=oms_prompt_flag), gr.update(visible=oms_guidance_flag) def mainloop(self): with gr.Blocks() as demo: gr.Markdown("# One More Step for SDXL w/ LCM-LoRA") with gr.Group() as inputs: prompt = gr.Textbox(label="Prompt", value="a cat against orange ground, studio") with gr.Accordion('OMS Prompt'): oms_prompt_checkbox = gr.Checkbox(info="Inconsistent OMS prompt allows the additional control of low freq info, default is the same as Prompt. \n Tips:When there is a conflict between the OMS prompt and the base prompt in describing the same object, the model will respect the base prompt.", label="Adding OMS Prompt", value=True) oms_prompt = gr.Textbox(label="OMS Prompt", value="a black cat", info='try "a black cat" and "a black room" for diverse control.') with gr.Accordion('OMS Guidance'): oms_cfg_scale_checkbox = gr.Checkbox(info="OMS Guidance will enhance the OMS prompt, specially focus on color and brightness.", label="Adding OMS Guidance", value=True) oms_guidance_scale = gr.Slider(label="OMS Guidance Scale", minimum=1.0, maximum=5.0, value=2., step=0.1) run_button = gr.Button(value="Generate images") with gr.Accordion("Advanced options", open=False): num_steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=4, step=1) sd_guidance_scale = gr.Slider(label="SD Pipe Guidance Scale", minimum=1, maximum=3, value=1.0, step=0.1) seed = gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, randomize=False, value=1024) with gr.Row(): output_raw = gr.Image(label="SDXL w/ LCM-LoRA ") output_oms_cp = gr.Image(label="w/ OMS (consistent) w/o OMS CFG") output_oms_icp = gr.Image(label="w/ OMS (inconsistent) w/o OMS CFG") output_oms_cfg = gr.Image(label="w/ OMS w/ OMS CFG") oms_prompt_checkbox.input( fn=lambda oms_prompt_flag, prompt, oms_prompt: (oms_prompt if oms_prompt_flag else prompt, gr.update(interactive=oms_prompt_flag)), inputs=[oms_prompt_checkbox, prompt, oms_prompt], outputs=[oms_prompt, oms_prompt] ) oms_cfg_scale_checkbox.input( fn=lambda oms_cfg_scale_flag: (1.5 if oms_cfg_scale_flag else 1.0, gr.update(interactive=oms_cfg_scale_flag)), inputs=[oms_cfg_scale_checkbox], outputs=[oms_guidance_scale, oms_guidance_scale] ) ips = [prompt, oms_prompt, oms_guidance_scale, num_steps, sd_guidance_scale, seed, oms_prompt_checkbox] run_button.click(fn=self._inference, inputs=ips, outputs=[output_raw, output_oms_cp, output_oms_icp, output_oms_cfg, output_oms_icp, output_oms_cfg]) demo.queue(max_size=20) demo.launch() if __name__ == "__main__": gradio_demo = GradioDemo() gradio_demo.mainloop()