from huggingface_hub import from_pretrained_keras import keras_cv import gradio as gr from tensorflow import keras keras.mixed_precision.set_global_policy("mixed_float16") resolution = 512 dreambooth_model = keras_cv.models.StableDiffusion( img_width=resolution, img_height=resolution, jit_compile=True, ) loaded_diffusion_model = from_pretrained_keras("melanit/dreambooth_voyager_v2") dreambooth_model._diffusion_model = loaded_diffusion_model def generate_images(prompt: str, negative_prompt:str, batch_size: int, num_steps: int): """ This function will infer the trained dreambooth (stable diffusion) model Args: prompt (str): The input text batch_size (int): The number of images to be generated num_steps (int): The number of denoising steps Returns: outputs (List): List of images that were generated using the model """ outputs = dreambooth_model.text_to_image( prompt, negative_prompt=negative_prompt, batch_size=batch_size, num_steps=num_steps, ) return outputs with gr.Blocks() as demo: gr.HTML("

Keras Dreambooth - Voyager Demo

") with gr.Row(): with gr.Column(): prompt = gr.Textbox(lines=1, value="a photo of voyager spaceship", label="Prompt") negative_prompt = gr.Textbox(lines=1, value="", label="Negative Prompt") samples = gr.Slider(minimum=1, maximum=10, value=1, step=1, label="Number of Images") num_steps = gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Denoising Steps") run = gr.Button(value="Run") with gr.Column(): gallery = gr.Gallery(label="Outputs").style(grid=(1,2)) run.click(generate_images, inputs=[prompt,negative_prompt, samples, num_steps], outputs=gallery) gr.Examples([["photo of voyager spaceship in space, high quality, blender, 3d, trending on artstation, 8k","bad, ugly, malformed, deformed, out of frame, blurry", 1, 50]], [prompt,negative_prompt, samples,num_steps], gallery, generate_images) gr.Markdown('Demo created by [Lily Berkow](https://huggingface.co/melanit/)') demo.launch()