Add application file

app.py

@@ -0,0 +1,66 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from diffusers import StableVideoDiffusionPipeline
+import torch
+import tempfile
+import os
+
+# Import music generation from separate file
+from music import generate_music
+
+# Load SmolLM-3B model once at startup
+tokenizer = AutoTokenizer.from_pretrained("cognitivecomputations/smolllm-3b")
+model = AutoModelForCausalLM.from_pretrained("cognitivecomputations/smolllm-3b")
+
+# Load Wan2.1 1.3B video model once at startup
+video_pipe = StableVideoDiffusionPipeline.from_pretrained("ByteDance/Wan-2-1-1-3B", torch_dtype=torch.float16).to("cuda")
+video_pipe.enable_model_cpu_offload()
+
+def generate_scenes_with_smol(script, style):
+    prompt = f"Break this {style.lower()} script into cinematic scenes with camera angles, characters, and mood.\nScript: {script}\nScene Breakdown:"
+    inputs = tokenizer(prompt, return_tensors="pt")
+    outputs = model.generate(**inputs, max_new_tokens=512)
+    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    scenes = [{"scene_id": i+1, "description": scene.strip()} for i, scene in enumerate(decoded.split("\n")) if scene.strip()]
+    return scenes
+
+def generate_video_with_wan(prompt):
+    video = video_pipe(prompt, num_frames=24, height=512, width=512).frames[0]  # Assuming single output
+    tmp_dir = tempfile.mkdtemp()
+    output_path = os.path.join(tmp_dir, "scene.mp4")
+    video[0].save(output_path, fps=8)  # Save as GIF or MP4 depending on format support
+    return output_path
+
+def process_script(script, style, want_music):
+    scenes = generate_scenes_with_smol(script, style)
+    video_clips = []
+
+    for scene in scenes:
+        text_prompt = scene['description']
+        video_path = generate_video_with_wan(text_prompt)
+        video_clips.append((scene['description'], video_path))
+
+    music_path = generate_music(script) if want_music else None
+    return video_clips, music_path
+
+with gr.Blocks() as app:
+    gr.Markdown("# 🎮 Vividly MVP – AI Video Creator")
+
+    with gr.Row():
+        script_input = gr.Textbox(label="Video Script", lines=6)
+        style_input = gr.Dropdown(["Cinematic", "Vlog", "Explainer"], value="Cinematic", label="Video Style")
+        music_toggle = gr.Checkbox(label="Generate background music")
+
+    submit_btn = gr.Button("Generate")
+
+    video_outputs = gr.Video(label="Scene-wise Video Clips", interactive=False, visible=False)
+    music_player = gr.Audio(label="Generated Music", visible=False)
+
+    def wrap(script, style, music):
+        scenes, music_path = process_script(script, style, music)
+        first_video = scenes[0][1] if scenes else None
+        return gr.update(value=first_video, visible=True), gr.update(value=music_path, visible=music is True)
+
+    submit_btn.click(wrap, inputs=[script_input, style_input, music_toggle], outputs=[video_outputs, music_player])
+
+app.launch()