#!/usr/bin/env python from __future__ import annotations import gradio as gr import numpy as np import spaces import random from diffusers import AutoencoderKL, DiffusionPipeline import torch import os import PIL.Image MARKDOWN = """ The demo is based on OpenDalle V1.1 by @dataautogpt3 The demo is based on the fusion of different models to provide better performance, comparatively. You can try out the prompts and check for yourself. **Parts of codes are adopted from [@hysts's SD-XL demo](https://huggingface.co/spaces/hysts/SD-XL) running on A10G GPU ** You can check out more of my spaces. Demo by [Sunder Ali Khowaja](https://sander-ali.github.io) - [Github](https://github.com/sander-ali) """ if not torch.cuda.is_available(): MARKDOWN += "\n

Running on CPU 🥶 This demo does not work on CPU.

" MAX_SEED = np.iinfo(np.int32).max CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "0") == "1" MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "1024")) USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE") == "1" ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD") == "1" ENABLE_REFINER = os.getenv("ENABLE_REFINER", "0") == "1" device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") if torch.cuda.is_available(): vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16) pipe = DiffusionPipeline.from_pretrained("dataautogpt3/OpenDalleV1.1", vae=vae, torch_dtype=torch.float16, variant="fp16", use_safetensors=True) pipe.enable_xformers_memory_efficient_attention() if ENABLE_REFINER: refiner = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", vae=vae, torch_dtype=torch.float16, variant="fp16", use_safetensors=True) if ENABLE_CPU_OFFLOAD: pipe.enable_model_cpu_offload() if ENABLE_REFINER: refiner.enable_model_cpu_offload() else: pipe.to(device) if ENABLE_REFINER: refiner.to(device) if USE_TORCH_COMPILE: pipe.unet = torch.compile(pipe.unet, mode='reduce-overhead', fullgraph=True) if ENABLE_REFINER: refiner.unet = torch.compile(refiner.unet, mode="reduce_overhead", fullgraph=True) def randomize_seed_fn(seed: int, randomize_seed: bool) -> int: if randomize_seed: seed = random.randint(0, MAX_SEED) return seed @spaces.GPU(enable_queue=True) def infer( prompt: str, seed: int = 0, width: int = 1024, height: int = 1024, guidance_scale_base: float = 5.0, guidance_scale_refiner: float = 5.0, num_inference_steps_base: int = 25, num_inference_steps_refiner: int = 25, apply_refiner: bool = False, negative_prompt: str = "", prompt_2: str = "", negative_prompt_2: str = "", use_negative_prompt: bool = False, use_prompt_2: bool = False, use_negative_prompt_2: bool = False, progress=gr.Progress(track_tqdm=True), ) -> PIL.Image.Image: print(f"** Generating image for: \"{prompt}\" **") generator = torch.Generator().manual_seed(seed) if not use_negative_prompt: negative_prompt = None # type: ignore if not use_prompt_2: prompt_2 = None # type: ignore if not use_negative_prompt_2: negative_prompt_2 = None # type: ignore if not apply_refiner: return pipe( prompt=prompt, negative_prompt=negative_prompt, prompt_2=prompt_2, negative_prompt_2=negative_prompt_2, width=width, height=height, guidance_scale=guidance_scale_base, num_inference_steps=num_inference_steps_base, generator=generator, output_type="pil", ).images[0] else: latents = pipe( prompt=prompt, negative_prompt=negative_prompt, prompt_2=prompt_2, negative_prompt_2=negative_prompt_2, width=width, height=height, guidance_scale=guidance_scale_base, num_inference_steps=num_inference_steps_base, generator=generator, output_type="latent", ).images image = refiner( prompt=prompt, negative_prompt=negative_prompt, prompt_2=prompt_2, negative_prompt_2=negative_prompt_2, guidance_scale=guidance_scale_refiner, num_inference_steps=num_inference_steps_refiner, image=latents, generator=generator, ).images[0] return image examples = [ "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", "An astronaut riding a green horse", "A delicious ceviche cheesecake slice", ] css=""" #col-container { margin: 0 auto; max-width: 520px; } """ # if torch.cuda.is_available(): # power_device = "GPU" # else: # power_device = "CPU" theme = gr.themes.Glass( primary_hue="blue", secondary_hue="blue", neutral_hue="gray", text_size="md", spacing_size="md", radius_size="md", font=[gr.themes.GoogleFont('Source Sans Pro'), 'ui-sans-serif', 'system-ui', 'sans-serif'], ).set( body_background_fill_dark='*background_fill_primary', background_fill_primary_dark='*neutral_950', background_fill_secondary='*neutral_50', background_fill_secondary_dark='*neutral_900', border_color_primary_dark='*neutral_700', block_background_fill='*background_fill_primary', block_background_fill_dark='*neutral_800', block_border_width='1px', block_label_background_fill='*background_fill_primary', block_label_background_fill_dark='*background_fill_secondary', block_label_text_color='*neutral_500', block_label_text_size='*text_sm', block_label_text_weight='400', block_shadow='none', block_shadow_dark='none', block_title_text_color='*neutral_500', block_title_text_weight='400', panel_border_width='0', panel_border_width_dark='0', checkbox_background_color_dark='*neutral_800', checkbox_border_width='*input_border_width', checkbox_label_border_width='*input_border_width', input_background_fill='*neutral_100', input_background_fill_dark='*neutral_700', input_border_color_focus_dark='*neutral_700', input_border_width='0px', input_border_width_dark='0px', slider_color='#2563eb', slider_color_dark='#2563eb', table_even_background_fill_dark='*neutral_950', table_odd_background_fill_dark='*neutral_900', button_border_width='*input_border_width', button_shadow_active='none', button_primary_background_fill='*primary_200', button_primary_background_fill_dark='*primary_700', button_primary_background_fill_hover='*button_primary_background_fill', button_primary_background_fill_hover_dark='*button_primary_background_fill', button_secondary_background_fill='*neutral_200', button_secondary_background_fill_dark='*neutral_600', button_secondary_background_fill_hover='*button_secondary_background_fill', button_secondary_background_fill_hover_dark='*button_secondary_background_fill', button_cancel_background_fill='*button_secondary_background_fill', button_cancel_background_fill_dark='*button_secondary_background_fill', button_cancel_background_fill_hover='*button_cancel_background_fill', button_cancel_background_fill_hover_dark='*button_cancel_background_fill' ) with gr.Blocks(css="footer{display:none !important}", theme=theme) as demo: gr.Markdown(MARKDOWN) gr.DuplicateButton() with gr.Group(): prompt = gr.Text( label="Prompt", show_label=False, max_lines=1, container=False, placeholder="Enter your prompt", ) run_button = gr.Button("Generate") result = gr.Image(label="Result", show_label=False) with gr.Accordion("Advanced options", open=False): with gr.Row(): use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=False) use_prompt_2 = gr.Checkbox(label="Use prompt 2", value=False) use_negative_prompt_2 = gr.Checkbox(label="Use negative prompt 2", value=False) negative_prompt = gr.Text( label="Negative prompt", max_lines=1, placeholder="Enter a negative prompt", visible=False, ) prompt_2 = gr.Text( label="Prompt 2", max_lines=1, placeholder="Enter your prompt", visible=False, ) negative_prompt_2 = gr.Text( label="Negative prompt 2", max_lines=1, placeholder="Enter a negative prompt", visible=False, ) seed = gr.Slider( label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, ) randomize_seed = gr.Checkbox(label="Randomize seed", value=True) with gr.Row(): width = gr.Slider( label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024, ) height = gr.Slider( label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024, ) apply_refiner = gr.Checkbox(label="Apply refiner", value=False, visible=ENABLE_REFINER) with gr.Row(): guidance_scale_base = gr.Slider( label="Guidance scale for base", minimum=1, maximum=20, step=0.1, value=5.0, ) num_inference_steps_base = gr.Slider( label="Number of inference steps for base", minimum=10, maximum=100, step=1, value=25, ) with gr.Row(visible=False) as refiner_params: guidance_scale_refiner = gr.Slider( label="Guidance scale for refiner", minimum=1, maximum=20, step=0.1, value=5.0, ) num_inference_steps_refiner = gr.Slider( label="Number of inference steps for refiner", minimum=10, maximum=100, step=1, value=25, ) gr.Examples( examples=examples, inputs=prompt, outputs=result, fn=infer, cache_examples=CACHE_EXAMPLES, ) use_negative_prompt.change( fn=lambda x: gr.update(visible=x), inputs=use_negative_prompt, outputs=negative_prompt, queue=False, api_name=False, ) use_prompt_2.change( fn=lambda x: gr.update(visible=x), inputs=use_prompt_2, outputs=prompt_2, queue=False, api_name=False, ) use_negative_prompt_2.change( fn=lambda x: gr.update(visible=x), inputs=use_negative_prompt_2, outputs=negative_prompt_2, queue=False, api_name=False, ) apply_refiner.change( fn=lambda x: gr.update(visible=x), inputs=apply_refiner, outputs=refiner_params, queue=False, api_name=False, ) gr.on( triggers=[ prompt.submit, negative_prompt.submit, prompt_2.submit, negative_prompt_2.submit, run_button.click, ], fn=randomize_seed_fn, inputs=[seed, randomize_seed], outputs=seed, queue=False, api_name=False, ).then( fn=infer, inputs=[ prompt, negative_prompt, prompt_2, negative_prompt_2, use_negative_prompt, use_prompt_2, use_negative_prompt_2, seed, width, height, guidance_scale_base, guidance_scale_refiner, num_inference_steps_base, num_inference_steps_refiner, apply_refiner, ], outputs=result, api_name="run", ) if __name__ == "__main__": demo.queue(max_size=20, api_open=False).launch(show_api=False, share=True)