import gradio as gr import numpy as np import random import time import spaces from diffusers import DiffusionPipeline import torch from PIL import Image from gradio.themes.base import Base from gradio.themes.utils import colors, fonts, sizes from typing import Iterable class Seafoam(Base): def __init__( self, *, primary_hue: colors.Color | str = colors.emerald, secondary_hue: colors.Color | str = colors.blue, neutral_hue: colors.Color | str = colors.gray, spacing_size: sizes.Size | str = sizes.spacing_md, radius_size: sizes.Size | str = sizes.radius_md, text_size: sizes.Size | str = sizes.text_lg, font: fonts.Font | str | Iterable[fonts.Font | str] = ( fonts.GoogleFont("Quicksand"), "ui-sans-serif", "sans-serif", ), font_mono: fonts.Font | str | Iterable[fonts.Font | str] = ( fonts.GoogleFont("IBM Plex Mono"), "ui-monospace", "monospace", ), ): super().__init__( primary_hue=primary_hue, secondary_hue=secondary_hue, neutral_hue=neutral_hue, spacing_size=spacing_size, radius_size=radius_size, text_size=text_size, font=font, font_mono=font_mono, ) seafoam = Seafoam() # Device and model setup device = "cuda" if torch.cuda.is_available() else "cpu" model_repo_id = "stabilityai/stable-diffusion-3.5-large-turbo" torch_dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32 pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype) pipe = pipe.to(device) pipe.load_lora_weights("prithivMLmods/SD3.5-Turbo-Realism-2.0-LoRA", weight_name="SD3.5-Turbo-Realism-2.0-LoRA.safetensors") trigger_word = "Turbo Realism" pipe.fuse_lora(lora_scale=1.0) # Constants MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 1024 # Define styles style_list = [ { "name": "3840 x 2160", "prompt": "hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic", "negative_prompt": "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly", }, { "name": "2560 x 1440", "prompt": "hyper-realistic 4K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic", "negative_prompt": "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly", }, { "name": "HD+", "prompt": "hyper-realistic 2K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic", "negative_prompt": "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly", }, { "name": "Style Zero", "prompt": "{prompt}", "negative_prompt": "", }, ] STYLE_NAMES = [s["name"] for s in style_list] DEFAULT_STYLE_NAME = STYLE_NAMES[0] # Define grid layouts grid_sizes = { "2x1": (2, 1), "1x2": (1, 2), "2x2": (2, 2), "2x3": (2, 3), "3x2": (3, 2), "1x1": (1, 1), } @spaces.GPU def infer( prompt, negative_prompt="", seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=7.5, num_inference_steps=10, style="Style Zero", grid_size="1x1", progress=gr.Progress(track_tqdm=True), ): if randomize_seed: seed = random.randint(0, MAX_SEED) generator = torch.Generator().manual_seed(seed) selected_style = next(s for s in style_list if s["name"] == style) styled_prompt = selected_style["prompt"].format(prompt=prompt) styled_negative = selected_style["negative_prompt"] or negative_prompt grid_x, grid_y = grid_sizes.get(grid_size, (1, 1)) num_images = grid_x * grid_y output = pipe( prompt=styled_prompt, negative_prompt=styled_negative, width=width, height=height, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps, generator=generator, num_images_per_prompt=num_images, ) grid_img = Image.new('RGB', (width * grid_x, height * grid_y)) for i, img in enumerate(output.images[:num_images]): x = (i % grid_x) * width y = (i // grid_x) * height grid_img.paste(img, (x, y)) return grid_img, seed examples = [ "A tiny, cute astronaut emerging from a cracked eggshell on the surface of the moon, surrounded by cosmic dust and distant planets in the starry sky. The scene is illuminated by soft lunar light, showcasing the texture of the moon's surface. Shot in ultra-detailed 4K resolution, with a sci-fi fantasy atmosphere and planetary background", "A cute, fluffy cat standing upright on its hind legs, holding a hand-drawn sign that says 'Hello World' in bold, playful letters. The background is softly blurred, emphasizing the cat’s detailed fur texture and the colorful sign. Captured in portrait aspect ratio --ar 85:128, using --v 6.0 and --style raw for a semi-realistic, endearing look", "A photorealistic image of a cold coffee beverage in a glass cup, condensation on the surface, sitting on a wooden café table with shallow depth of field. The background features beautiful bokeh lighting, creating a cozy, blurred café ambiance. Shot in portrait mode --ar 85:128, 4K ultra-resolution, using --style raw5 for authentic textures, --v 6.0.", "An anime-style food illustration of a golden-brown wiener schnitzel, perfectly fried and crispy, served on a white ceramic plate. Accompanied by fresh parsley garnish and thin lemon slices, with artistic shading and stylized highlights. Captured in an anime illustration format --style raw5, high color saturation, food-themed aesthetic." ] css = ''' .gradio-container { max-width: 100%; margin: 0 auto; } h1 { text-align: center; } footer { visibility: hidden; } ''' with gr.Blocks(theme=seafoam, css=css) as demo: gr.Markdown("## T2I SD3.5") with gr.Row(): with gr.Column(scale=1): with gr.Row(): prompt = gr.Text( show_label=False, max_lines=1, placeholder="Enter your prompt", container=False, ) run_button = gr.Button("Run", scale=0, variant="primary") result = gr.Image(show_label=False, format="png") grid_size_selection = gr.Dropdown( choices=list(grid_sizes.keys()), value="1x1", label="Grid Size" ) with gr.Accordion("Advanced Settings", open=False): negative_prompt = gr.Text( label="Negative prompt", max_lines=1, placeholder="Enter a negative prompt", value="(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation", ) seed = gr.Slider(0, MAX_SEED, value=0, label="Seed") randomize_seed = gr.Checkbox(label="Randomize seed", value=True) with gr.Row(): width = gr.Slider(512, MAX_IMAGE_SIZE, step=32, value=1024, label="Width") height = gr.Slider(512, MAX_IMAGE_SIZE, step=32, value=1024, label="Height") with gr.Row(): guidance_scale = gr.Slider(0.0, 7.5, step=0.1, value=0.0, label="Guidance scale") num_inference_steps = gr.Slider(1, 50, step=1, value=10, label="Number of inference steps") style_selection = gr.Radio( choices=STYLE_NAMES, value=DEFAULT_STYLE_NAME, label="Quality Style", ) with gr.Column(scale=1): gr.Examples( examples=examples, inputs=[prompt], outputs=[result, seed], fn=infer, cache_examples=False, label="Prompt Examples" ) gr.on( triggers=[run_button.click, prompt.submit], fn=infer, inputs=[ prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, style_selection, grid_size_selection ], outputs=[result, seed], ) if __name__ == "__main__": demo.launch(ssr_mode=False)