diff --git a/models/Wav2Lip/vocodes_process_inference.py b/models/Wav2Lip/vocodes_process_inference.py
index 40f6da7d86..5f604c0995 100644
--- a/models/Wav2Lip/vocodes_process_inference.py
+++ b/models/Wav2Lip/vocodes_process_inference.py
@@ -47,11 +47,14 @@ parser.add_argument('--cached_faces_filename', type=str,
 parser.add_argument('--audio_filename', type=str,
                     help='Filepath of video/audio file to use as raw audio source', required=True)
 
+parser.add_argument('--output_video_filename', type=str,
+                    help='Output filename for the final video', required=True)
+
 parser.add_argument('--output_metadata_filename', type=str,
                     help='Output filename for the JSON containing width, height, etc.', required=True)
 
-parser.add_argument('--outfile', type=str, help='Video path to save result. See default for an e.g.',
-                                default='results/result_voice.mp4')
+#parser.add_argument('--outfile', type=str, help='Video path to save result. See default for an e.g.',
+#                                default='results/result_voice.mp4')
 
 parser.add_argument('--static', type=bool,
                     help='If True, then use only first video frame for inference', default=False)
@@ -309,7 +312,7 @@ def maybe_concatenate_end_bump(tempdir, args, frame_w, frame_h):
     # More info on filters: https://stackoverflow.com/a/22958746
     command = ' '.join([
         'ffmpeg',
-        '-i {}'.format(args.outfile),
+        '-i {}'.format(args.output_video_filename),
         '-i {}'.format(bumpfile),
         '-filter_complex "concat=n=2:v=1:a=1"',
         '{}'.format(padded_wav_filename)
@@ -321,7 +324,7 @@ def maybe_concatenate_end_bump(tempdir, args, frame_w, frame_h):
         print('Failed to concatenate end bump')
     else:
         print('Done concatenating end bump; renaming file...', flush=True)
-        os.rename(padded_wav_filename, args.outfile)
+        os.rename(padded_wav_filename, args.output_video_filename)
 
 
 def main(tempdir):
@@ -476,7 +479,7 @@ def main(tempdir):
 
     out.release()
 
-    command = 'ffmpeg -y -i {} -i {} -strict -2 -q:v 1 {}'.format(args.audio_filename, output_video_filename, args.outfile)
+    command = 'ffmpeg -y -i {} -i {} -strict -2 -q:v 1 {}'.format(args.audio_filename, output_video_filename, args.output_video_filename)
     print('command:', command, flush=True)
     subprocess.call(command, shell=True)
 
diff --git a/src/script_execution/wav2lip_inference_command.rs b/src/script_execution/wav2lip_inference_command.rs
index 5826f2ac87..b33f7e0497 100644
--- a/src/script_execution/wav2lip_inference_command.rs
+++ b/src/script_execution/wav2lip_inference_command.rs
@@ -32,7 +32,7 @@ impl Wav2LipInferenceCommand {
                  media_template_filename: P,
                  cached_faces_filename: P,
                  output_metadata_filename: P,
-                 output_media_filename: P,
+                 output_video_filename: P,
                  is_image: bool,
                  spawn_process: bool) -> AnyhowResult<()>
   {
@@ -54,6 +54,8 @@ impl Wav2LipInferenceCommand {
     command.push_str(&audio_filename.as_ref().display().to_string());
     command.push_str(" --cached_faces_filename ");
     command.push_str(&cached_faces_filename.as_ref().display().to_string());
+    command.push_str(" --output_video_filename ");
+    command.push_str(&output_video_filename.as_ref().display().to_string());
     command.push_str(" --output_metadata_filename ");
     command.push_str(&output_metadata_filename.as_ref().display().to_string());
 
diff --git a/src/util/semi_persistent_cache_dir.rs b/src/util/semi_persistent_cache_dir.rs
index fd03496b03..1aa0f8d51f 100644
--- a/src/util/semi_persistent_cache_dir.rs
+++ b/src/util/semi_persistent_cache_dir.rs
@@ -1,6 +1,7 @@
 use std::path::{PathBuf, Path};
 use std::fs;
 use crate::util::anyhow_result::AnyhowResult;
+use tempdir::TempDir;
 
 /// These are for files on the worker filesystems
 pub struct SemiPersistentCacheDir {
@@ -103,4 +104,35 @@ impl SemiPersistentCacheDir {
     let _ = fs::create_dir_all(self.video_asset_path())?;
     Ok(())
   }
+
+  // ==================== W2L OUTPUT RESULTS ====================
+
+  /// We cache W2L faces here.
+  /// We'll likely need to LRU cache them.
+  pub fn w2l_output_results_path(&self, temp_dir: &TempDir, inference_job_token: &str) -> PathBuf {
+    // NB: We don't want colons from the token in the filename.
+    let filename = inference_job_token.replace(":", "");
+    let filename = format!("{}_result.mp4", filename);
+
+    temp_dir.path().join(&filename)
+  }
+}
+
+#[cfg(test)]
+mod tests {
+  use crate::util::semi_persistent_cache_dir::SemiPersistentCacheDir;
+  use tempdir::TempDir;
+
+  #[test]
+  fn test_w2l_output_results_path() {
+    let dirs = SemiPersistentCacheDir::default_paths();
+    let temp_dir = TempDir::new("test").unwrap();
+
+    let filename = dirs.w2l_output_results_path(&temp_dir, "TYPE:TOKEN")
+      .to_str()
+      .map(|s| s.to_string())
+      .unwrap();
+
+    assert!(&filename.ends_with("TYPETOKEN_result.mp4"));
+  }
 }
diff --git a/src/w2l_inference_job.rs b/src/w2l_inference_job.rs
index 2f4a7a8758..6f5bd64aa4 100644
--- a/src/w2l_inference_job.rs
+++ b/src/w2l_inference_job.rs
@@ -441,6 +441,8 @@ async fn process_job(inferencer: &Inferencer, job: &W2lInferenceJobRecord) -> An
     false
   )?;
 
+  let output_video_fs_path = temp_dir.path().join(&audio_bucket_hash);
+
   if true {
     info!("FAKE DONE");
     thread::sleep(Duration::from_millis(5000));