app.py

import torch
import gradio as gr
import numpy as np
import random

from diffusers import (
    StableDiffusionPipeline,
    StableDiffusionInstructPix2PixPipeline,
    StableVideoDiffusionPipeline,
    WanPipeline,
)
from diffusers.utils import export_to_video, load_image

# Force CPU mode
device = "cpu"
dtype = torch.float32
MAX_SEED = np.iinfo(np.int32).max

# Global pipeline holders
TXT2IMG_PIPE = None
IMG2IMG_PIPE = None
TXT2VID_PIPE = None
IMG2VID_PIPE = None

# Helper to load models
def make_pipe(cls, model_id, **kwargs):
    pipe = cls.from_pretrained(model_id, torch_dtype=dtype, **kwargs)
    pipe.to(device)
    return pipe

# Text → Image
def generate_image_from_text(prompt, seed, randomize_seed):
    global TXT2IMG_PIPE
    if TXT2IMG_PIPE is None:
        TXT2IMG_PIPE = make_pipe(StableDiffusionPipeline, "stabilityai/stable-diffusion-2-1-base")
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.manual_seed(seed)
    image = TXT2IMG_PIPE(prompt=prompt, num_inference_steps=20, generator=generator).images[0]
    return image, seed

# Image → Image
def generate_image_from_image_and_prompt(image, prompt, seed, randomize_seed):
    global IMG2IMG_PIPE
    if IMG2IMG_PIPE is None:
        IMG2IMG_PIPE = make_pipe(StableDiffusionInstructPix2PixPipeline, "timbrooks/instruct-pix2pix")
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.manual_seed(seed)
    out = IMG2IMG_PIPE(prompt=prompt, image=image, num_inference_steps=8, generator=generator)
    return out.images[0], seed

# Text → Video
def generate_video_from_text(prompt, seed, randomize_seed):
    global TXT2VID_PIPE
    if TXT2VID_PIPE is None:
        TXT2VID_PIPE = make_pipe(WanPipeline, "Wan-AI/Wan2.1-T2V-1.3B-Diffusers")
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.manual_seed(seed)
    frames = TXT2VID_PIPE(prompt=prompt, num_frames=12, generator=generator).frames[0]
    return export_to_video(frames, "/tmp/wan_video.mp4", fps=8), seed

# Image → Video
def generate_video_from_image(image, seed, randomize_seed):
    global IMG2VID_PIPE
    if IMG2VID_PIPE is None:
        IMG2VID_PIPE = make_pipe(
            StableVideoDiffusionPipeline,
            "stabilityai/stable-video-diffusion-img2vid-xt"
        )
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.manual_seed(seed)
    image = load_image(image).resize((512, 288))
    frames = IMG2VID_PIPE(image=image, num_inference_steps=16, generator=generator).frames[0]
    return export_to_video(frames, "/tmp/svd_video.mp4", fps=8), seed

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("# 🧠 AI Playground – Text & Image → Image & Video")

    with gr.Tabs():
        # Text → Image
        with gr.Tab("Text → Image"):
            prompt_txt = gr.Textbox(label="Prompt")
            btn_txt2img = gr.Button("Generate")
            result_img = gr.Image()
            seed_txt = gr.Slider(0, MAX_SEED, value=42, label="Seed")
            rand_txt = gr.Checkbox(label="Randomize seed", value=True)
            btn_txt2img.click(
                generate_image_from_text,
                inputs=[prompt_txt, seed_txt, rand_txt],
                outputs=[result_img, seed_txt]
            )

        # Image → Image
        with gr.Tab("Image → Image"):
            image_in = gr.Image(label="Input Image")
            prompt_img = gr.Textbox(label="Edit Prompt")
            btn_img2img = gr.Button("Generate")
            result_img2 = gr.Image()
            seed_img = gr.Slider(0, MAX_SEED, value=123, label="Seed")
            rand_img = gr.Checkbox(label="Randomize seed", value=True)
            btn_img2img.click(
                generate_image_from_image_and_prompt,
                inputs=[image_in, prompt_img, seed_img, rand_img],
                outputs=[result_img2, seed_img]
            )

        # Text → Video
        with gr.Tab("Text → Video"):
            prompt_vid = gr.Textbox(label="Prompt")
            btn_txt2vid = gr.Button("Generate")
            result_vid = gr.Video()
            seed_vid = gr.Slider(0, MAX_SEED, value=555, label="Seed")
            rand_vid = gr.Checkbox(label="Randomize seed", value=True)
            btn_txt2vid.click(
                generate_video_from_text,
                inputs=[prompt_vid, seed_vid, rand_vid],
                outputs=[result_vid, seed_vid]
            )

        # Image → Video
        with gr.Tab("Image → Video"):
            image_vid = gr.Image(label="Input Image")
            btn_img2vid = gr.Button("Animate")
            result_vid2 = gr.Video()
            seed_vid2 = gr.Slider(0, MAX_SEED, value=999, label="Seed")
            rand_vid2 = gr.Checkbox(label="Randomize seed", value=True)
            btn_img2vid.click(
                generate_video_from_image,
                inputs=[image_vid, seed_vid2, rand_vid2],
                outputs=[result_vid2, seed_vid2]
            )

demo.queue()
demo.launch(show_error=True)