diff --git a/mani_skill2/utils/wrappers/record.py b/mani_skill2/utils/wrappers/record.py
index ba284acbf..e19ff181b 100644
--- a/mani_skill2/utils/wrappers/record.py
+++ b/mani_skill2/utils/wrappers/record.py
@@ -185,7 +185,9 @@ class RecordEpisode(gym.Wrapper):
         save_trajectory: whether to save trajectory
         trajectory_name: name of trajectory file (.h5). Use timestamp if not provided.
         save_video: whether to save video
-        info_on_video: whether to write data about reward and data in the info object to the video
+        info_on_video: whether to write data about reward, action, and data in the info object to the video. The first video frame is generally the result
+            of the first env.reset() (visualizing the first observation). Text is written on frames after that, showing the action taken to get to that
+            environment state and reward.
         save_on_reset: whether to save the previous trajectory (and video of it if `save_video` is True) automatically when resetting.
             Not that for environments simulated on the GPU (to leverage fast parallel rendering) you must
             set `max_steps_per_video` to a fixed number so that every `max_steps_per_video` steps a video is saved. This is
@@ -385,12 +387,14 @@ def recursive_replace(x, y):
                     )
                 if self._trajectory_buffer.fail is not None:
                     recursive_replace(self._trajectory_buffer.fail, first_step.fail)
-        if self.save_video:
-            self._render_images.append(self.capture_image())
 
         return obs, info
 
     def step(self, action):
+        if self.save_video and self._video_steps == 0:
+            # save the first frame of the video here (s_0) instead of inside reset as user
+            # may call env.reset(...) multiple times but we want to ignore empty trajectories
+            self._render_images.append(self.capture_image())
         obs, rew, terminated, truncated, info = super().step(action)
 
         if self.save_trajectory:
diff --git a/manualtest/record_test.py b/manualtest/record_test.py
index d5561bb9a..fc7d8c28b 100644
--- a/manualtest/record_test.py
+++ b/manualtest/record_test.py
@@ -10,7 +10,7 @@
 if __name__ == "__main__":
     # sapien.set_log_level("info")
     # , "StackCube-v1", "PickCube-v1", "PushCube-v1", "PickSingleYCB-v1", "OpenCabinet-v1"
-    num_envs = 1
+    num_envs = 2
     for env_id in ["PickCube-v1"]:
         env = gym.make(
             env_id,
@@ -39,7 +39,11 @@
         )
         # env = ManiSkillVectorEnv(env)
 
-        env.reset(seed=52, options=dict(reconfigure=True))
+        # env.reset(seed=52, options=dict(reconfigure=True))
+        env.reset()
+        env.reset()
+        env.reset()
+        env.reset()
         # for i in range(180):
         #     env.step(env.action_space.sample())
         # env.step(env.action_space.sample())