app.py

import gradio as gr
import tempfile
import imageio
import torch
from transformers import pipeline
from diffusers import DiffusionPipeline

# ---------- Configuration ----------
AVAILABLE_MODELS = {
    "GPT-2 (small, fast)": "gpt2",
    "Falcon (TII UAE)": "tiiuae/falcon-7b-instruct",
    "Mistral (OpenAccess)": "mistralai/Mistral-7B-v0.1"
}

device = "cuda" if torch.cuda.is_available() else "cpu"
text_model_cache = {}
chat_memory = {}

# ---------- Load Image Generator ----------
try:
    image_generator = DiffusionPipeline.from_pretrained(
        "runwayml/stable-diffusion-v1-5",
        safety_checker=None,
        torch_dtype=torch.float16 if device == "cuda" else torch.float32
    )
    image_generator.to(device)
    image_enabled = True
except Exception as e:
    print(f"[Image Model Load Error]: {e}")
    image_generator = None
    image_enabled = False

# ---------- Load Video Generator ----------
try:
    video_pipeline = DiffusionPipeline.from_pretrained(
        "damo-vilab/text-to-video-ms-1.7b",
        safety_checker=None,
        torch_dtype=torch.float16 if device == "cuda" else torch.float32
    )
    video_pipeline.to(device)
    video_enabled = True
except Exception as e:
    print(f"[Video Model Load Error]: {e}")
    video_pipeline = None
    video_enabled = False

# ---------- Streamed Response Generator ----------
def codette_terminal(prompt, model_name, generate_image, generate_video, session_id, batch_size, video_steps, fps):
    if session_id not in chat_memory:
        chat_memory[session_id] = []

    if prompt.lower() in ["exit", "quit"]:
        chat_memory[session_id] = []
        yield "🧠 Codette signing off... Session reset.", None, None
        return

    # Load text model if not already loaded
    if model_name not in text_model_cache:
        try:
            text_model_cache[model_name] = pipeline(
                "text-generation",
                model=AVAILABLE_MODELS[model_name],
                device=0 if device == "cuda" else -1
            )
        except Exception as e:
            yield f"[Text model error]: {e}", None, None
            return

    generator = text_model_cache[model_name]

    # Generate response
    try:
        output = generator(prompt, max_length=100, do_sample=True, num_return_sequences=1)[0]['generated_text'].strip()
    except Exception as e:
        yield f"[Text generation error]: {e}", None, None
        return

    # Stream the output character by character
    response_so_far = ""
    for char in output:
        response_so_far += char
        temp_log = chat_memory[session_id][:]
        temp_log.append(f"🖋️ You > {prompt}")
        temp_log.append(f"🧠 Codette > {response_so_far}")
        yield "\n".join(temp_log[-10:]), None, None
        import time
        time.sleep(0.01)

    # Finalize chat memory
    chat_memory[session_id].append(f"🖋️ You > {prompt}")
    chat_memory[session_id].append(f"🧠 Codette > {output}")

    # Image Generation
    imgs = None
    if generate_image and image_enabled:
        try:
            result = image_generator(prompt, num_images_per_prompt=batch_size)
            imgs = result.images
        except Exception as e:
            response_so_far += f"\n[Image error]: {e}"

    # Video Generation
    vid = None
    if generate_video and video_enabled:
        try:
            result = video_pipeline(prompt, num_inference_steps=video_steps)
            frames = result.frames
            temp_video_path = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False).name
            imageio.mimsave(temp_video_path, frames, fps=fps)
            vid = temp_video_path
        except Exception as e:
            response_so_far += f"\n[Video error]: {e}"

    yield "\n".join(chat_memory[session_id][-10:]), imgs, vid

# ---------- Gradio UI ----------
with gr.Blocks(title="🧬 Codette Terminal – Streamed AI Chat") as demo:
    gr.Markdown("## 🧬 Codette Terminal (Chat + Image + Video + Batch + NSFW OK)")
    gr.Markdown("Type a prompt, select your model, and configure generation options. Type `'exit'` to reset.")

    session_id = gr.Textbox(value="session_default", visible=False)
    model_dropdown = gr.Dropdown(choices=list(AVAILABLE_MODELS.keys()), value="GPT-2 (small, fast)", label="Language Model")
    generate_image_toggle = gr.Checkbox(label="Generate Image(s)?", value=False, interactive=image_enabled)
    generate_video_toggle = gr.Checkbox(label="Generate Video?", value=False, interactive=video_enabled)
    batch_size_slider = gr.Slider(label="Number of Images", minimum=1, maximum=4, step=1, value=1)
    video_steps_slider = gr.Slider(label="Video Inference Steps", minimum=10, maximum=100, step=10, value=50)
    fps_slider = gr.Slider(label="Video FPS", minimum=4, maximum=24, step=2, value=8)
    user_input = gr.Textbox(label="Your Prompt", placeholder="e.g. A robot dreaming on Mars", lines=1)
    output_text = gr.Textbox(label="Codette Output", lines=15, interactive=False)
    output_image = gr.Gallery(label="Generated Image(s)").style(grid=2)
    output_video = gr.Video(label="Generated Video")

    user_input.submit(
        codette_terminal,
        inputs=[user_input, model_dropdown, generate_image_toggle, generate_video_toggle, session_id, batch_size_slider, video_steps_slider, fps_slider],
        outputs=[output_text, output_image, output_video],
        concurrency_limit=1,
        queue=True,
        show_progress=True
    )

# ---------- Launch ----------
if __name__ == "__main__":
    demo.launch()