parse character position

app.py

@@ -114,8 +114,11 @@ def generate(
     seq_feat = diffuser.net.model.clip_sequential
 
     batch = get_batch(prompt, sample_id, clip_model, dataset, seq_feat, device)
-    batch["character_position"] = torch.tensor(character_position, device=device)
-
+    batch["character_position"] = torch.tensor(
+        [float(coord) for coord in character_position.strip("[]").split(",")],
+        device=device,
+    )
+    
     with torch.no_grad():
         out = diffuser.predict_step(batch, 0)