Update app.py

app.py

@@ -4,95 +4,97 @@ import base64
 import io
 import soundfile as sf
 
-# 🔥 IMPORTANT: your ngrok API endpoint
+# 🔥 Your ngrok API endpoint — update this each time you restart Colab
 API_URL = "https://bustled-hertha-unprojective.ngrok-free.dev/generate"
 
+# Required to bypass ngrok browser warning page (fixes Invalid JSON error)
+HEADERS = {"ngrok-skip-browser-warning": "true"}
+
 
 def generate(prompt, duration, steps, cfg):
     try:
-        # 🚀 send request to backend
+        print(f"📤 Sending request → prompt='{prompt}' duration={duration}s steps={steps}")
+
         res = requests.post(
             API_URL,
             json={
-                "prompt": prompt,
-                "duration": duration,
-                "steps": steps,
-                "cfg_scale": cfg
+                "prompt":    prompt,
+                "duration":  float(duration),
+                "steps":     int(steps),
+                "cfg_scale": float(cfg),
             },
-            timeout=180  # longer timeout for generation
+            headers=HEADERS,
+            timeout=300,  # 5 min — generation can be slow
         )
 
-        print("STATUS:", res.status_code)
+        print(f"STATUS: {res.status_code}")
 
-        # ❌ backend error
         if res.status_code != 200:
-            print("❌ Backend error:", res.text[:300])
+            print(f"❌ Backend error: {res.text[:500]}")
             return None
 
-        # ❌ empty response
         if not res.text:
-            print("❌ Empty response")
+            print("❌ Empty response from backend")
             return None
 
-        # ❌ invalid JSON
         try:
             data = res.json()
         except Exception:
-            print("❌ Invalid JSON:", res.text[:300])
+            print(f"❌ Invalid JSON (got HTML?): {res.text[:300]}")
             return None
 
-        # ❌ missing audio
         if "audio" not in data:
-            print("❌ No audio key:", data)
+            print(f"❌ No audio key in response: {data}")
             return None
 
-        # ✅ decode base64 → audio
+        # Decode base64 WAV → numpy
         audio_bytes = base64.b64decode(data["audio"])
-        audio, sr = sf.read(io.BytesIO(audio_bytes))
+        audio, sr   = sf.read(io.BytesIO(audio_bytes))
 
+        print(f"✅ Got audio: {len(audio)/sr:.2f}s @ {sr}Hz")
         return sr, audio
 
+    except requests.exceptions.Timeout:
+        print("❌ Request timed out — try fewer steps or shorter duration")
+        return None
     except Exception as e:
-        print("❌ Request failed:", str(e))
+        print(f"❌ Request failed: {e}")
         return None
 
 
-# 🎨 UI
+# ── UI ────────────────────────────────────────────────────────────────────────
 with gr.Blocks(title="AutoMix AI 🎵") as demo:
     gr.Markdown("# 🎵 AutoMix AI Beat Generator")
-    gr.Markdown("Generate AI beats using diffusion model 🚀")
+    gr.Markdown("Generate AI beats using a diffusion model fine-tuned on trap/rap/R&B 🚀")
 
     with gr.Row():
         with gr.Column():
-            prompt = gr.Textbox(
+            prompt_in = gr.Textbox(
                 label="🎧 Prompt",
-                placeholder="Dark trap beat, 140 BPM, heavy 808..."
+                placeholder="A dark trap beat at 140 BPM in C minor, featuring 808 bass and synth bells.",
+                lines=3,
             )
-
-            duration = gr.Slider(
-                1, 10, value=5, step=1,
-                label="⏱ Duration (seconds)"
+            duration_in = gr.Slider(
+                minimum=5, maximum=47, value=30, step=1,
+                label="⏱ Duration (seconds)",
             )
-
-            steps = gr.Slider(
-                20, 80, value=40, step=5,
-                label="⚙️ Diffusion Steps"
+            steps_in = gr.Slider(
+                minimum=20, maximum=200, value=100, step=10,
+                label="⚙️ Diffusion Steps  (more = better quality, slower)",
             )
-
-            cfg = gr.Slider(
-                1, 10, value=7, step=0.5,
-                label="🎛 CFG Scale"
+            cfg_in = gr.Slider(
+                minimum=1.0, maximum=15.0, value=7.0, step=0.5,
+                label="🎛 CFG Scale",
             )
-
             btn = gr.Button("🚀 Generate Beat", variant="primary")
 
         with gr.Column():
-            output = gr.Audio(label="🎵 Generated Beat")
+            output = gr.Audio(label="🎵 Generated Beat", type="numpy")
 
     btn.click(
         fn=generate,
-        inputs=[prompt, duration, steps, cfg],
-        outputs=output
+        inputs=[prompt_in, duration_in, steps_in, cfg_in],
+        outputs=output,
     )
 
 demo.launch()