App and requirement files

app.py

@@ -0,0 +1,57 @@
+import logging
+import os
+
+import gradio as gr
+import keras_cv
+import numpy as np
+import tensorflow as tf
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__file__)
+prompt_token = os.environ.get("TOKEN", "<token>")
+text_encoder_path = os.environ.get(
+    "TEXT_ENCODER", "./models/example/text_encoder/keras"
+)
+
+logger.info(f'Inversed token used: "{prompt_token}"')
+logger.info(f'Loading text encoder from: "{text_encoder_path}"')
+
+stable_diffusion = keras_cv.models.StableDiffusion()
+stable_diffusion.tokenizer.add_tokens(prompt_token)
+loaded_text_encoder_ = tf.keras.models.load_model(text_encoder_path)
+stable_diffusion._text_encoder = loaded_text_encoder_
+stable_diffusion._text_encoder.compile(jit_compile=True)
+
+
+def generate_fn(input_prompt: str) -> np.ndarray:
+    """Generates images from a text prompt
+
+    Args:
+        input_prompt (str): Text input prompt
+
+    Returns:
+        np.ndarray: Generated image
+    """
+    generated = stable_diffusion.text_to_image(
+        prompt=input_prompt, batch_size=1, num_steps=1
+    )
+    return generated[0]
+
+
+iface = gr.Interface(
+    fn=generate_fn,
+    title="Textual Inversion",
+    description="Textual Inversion Demo",
+    article="Note: Keras-cv uses lazy intialization, so the first use will be slower while the model is initialized.",
+    inputs=gr.Textbox(
+        label="Prompt",
+        show_label=False,
+        max_lines=2,
+        placeholder="Enter your prompt",
+        elem_id="input-prompt",
+    ),
+    outputs=gr.Image(),
+)
+
+if __name__ == "__main__":
+    app, local_url, share_url = iface.launch()

requirements.txt

@@ -0,0 +1,5 @@
+numpy
+gradio
+keras-cv==0.4.2
+tensorflow-datasets
+tensorflow==2.11.0