Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

add frame resizing

EnzoFanAccounte629eec48094
diff --git a/browser_use/browser/video_recorder.py b/browser_use/browser/video_recorder.py
index 5d260fe66..2445cfde0 100644
--- a/browser_use/browser/video_recorder.py
+++ b/browser_use/browser/video_recorder.py
@@ -2,13 +2,18 @@
import base64
import logging
+import math
+import subprocess
from pathlib import Path
from typing import Optional
+import numpy as np
+
from browser_use.browser.profile import ViewportSize
try:
import imageio.v2 as iio
+ import imageio_ffmpeg
from imageio.core.format import Format
IMAGEIO_AVAILABLE = True
@@ -18,6 +23,13 @@ except ImportError:
logger = logging.getLogger(__name__)
+def _get_padded_size(size: ViewportSize, macro_block_size: int = 16) -> ViewportSize:
+ """Calculates the dimensions padded to the nearest multiple of macro_block_size."""
+ width = int(math.ceil(size['width'] / macro_block_size)) * macro_block_size
+ height = int(math.ceil(size['height'] / macro_block_size)) * macro_block_size
+ return ViewportSize(width=width, height=height)
+
+
class VideoRecorderService:
"""
Handles the video encoding process for a browser session using imageio.
@@ -41,6 +53,7 @@ class VideoRecorderService:
self.framerate = framerate
self._writer: Optional['Format.Writer'] = None
self._is_active = False
+ self.padded_size = _get_padded_size(self.size)
def start(self) -> None:
"""
@@ -57,13 +70,14 @@ class VideoRecorderService:
try:
self.output_path.parent.mkdir(parents=True, exist_ok=True)
+ # The macro_block_size is set to None because we handle padding ourselves
self._writer = iio.get_writer(
str(self.output_path),
fps=self.framerate,
codec='libx264',
quality=8, # A good balance of quality and file size (1-10 scale)
pixelformat='yuv420p', # Ensures compatibility with most players
- macro_block_size=16, # Recommended for h264
+ macro_block_size=None,
)
self._is_active = True
logger.debug(f'Video recorder started. Output will be saved to {self.output_path}')
@@ -73,10 +87,8 @@ class VideoRecorderService:
def add_frame(self, frame_data_b64: str) -> None:
"""
- Decodes a base64-encoded PNG frame and appends it to the video.
-
- This method is designed to be fast and non-blocking. It will
- gracefully handle corrupted frames.
+ Decodes a base64-encoded PNG frame, resizes it, pads it to be codec-compatible,
+ and appends it to the video.
Args:
frame_data_b64: A base64-encoded string of the PNG frame data.
@@ -86,21 +98,47 @@ class VideoRecorderService:
try:
frame_bytes = base64.b64decode(frame_data_b64)
- # imageio reads bytes directly and converts to a numpy array
- # The format is auto-detected from the bytes.
- img_array = iio.imread(frame_bytes)
-
- # Ensure frame dimensions match video dimensions
- h, w, _ = img_array.shape
- if w != self.size['width'] or h != self.size['height']:
- # This can happen if the viewport changes mid-recording.
- # A more robust solution could involve resizing, but that is non-trivial.
- # For now, the video size must be the same as the viewport
- logger.warning(
- f'Frame size ({w}x{h}) does not match video size '
- f'({self.size["width"]}x{self.size["height"]}). Skipping frame.'
- )
- return
+
+ # Build a filter chain for ffmpeg:
+ # 1. scale: Resizes the frame to the user-specified dimensions.
+ # 2. pad: Adds black bars to meet codec's macro-block requirements,
+ # centering the original content.
+ vf_chain = (
+ f'scale={self.size["width"]}:{self.size["height"]},'
+ f'pad={self.padded_size["width"]}:{self.padded_size["height"]}:(ow-iw)/2:(oh-ih)/2:color=black'
+ )
+
+ output_pix_fmt = 'rgb24'
+ command = [
+ imageio_ffmpeg.get_ffmpeg_exe(),
+ '-f',
+ 'image2pipe', # Input format from a pipe
+ '-c:v',
+ 'png', # Specify input codec is PNG
+ '-i',
+ '-', # Input from stdin
+ '-vf',
+ vf_chain, # Video filter for resizing and padding
+ '-f',
+ 'rawvideo', # Output format is raw video
+ '-pix_fmt',
+ output_pix_fmt, # Output pixel format
+ '-', # Output to stdout
+ ]
+
+ # Execute ffmpeg as a subprocess
+ proc = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+ out, err = proc.communicate(input=frame_bytes)
+
+ if proc.returncode != 0:
+ err_msg = err.decode(errors='ignore').strip()
+ if 'deprecated pixel format used' not in err_msg.lower():
+ raise IOError(f'ffmpeg error during resizing/padding: {err_msg}')
+ else:
+ logger.debug(f'ffmpeg warning during resizing/padding: {err_msg}')
+
+ # Convert the raw output bytes to a numpy array with the padded dimensions
+ img_array = np.frombuffer(out, dtype=np.uint8).reshape((self.padded_size['height'], self.padded_size['width'], 3))
self._writer.append_data(img_array)
except Exception as e:
diff --git a/docs/customize/browser-settings.mdx b/docs/customize/browser-settings.mdx
index dbec43905..defae8b76 100644
--- a/docs/customize/browser-settings.mdx
+++ b/docs/customize/browser-settings.mdx
@@ -354,13 +354,31 @@ window_position: dict | None = {"width": 0, "height": 0}
Window position from top-left corner.
-#### `save_recording_path`
+#### `record_video_dir`
+
+```python
+record_video_dir: str | Path | None = None
+```
+
+(alias: `save_recording_path`)
+
+Directory to save video recordings. If a path is provided, a video of the session will be recorded to an MP4 file in this directory.
+
+#### `record_video_size`
+
+```python
+record_video_size: dict | None = None
+```
+
+The frame size (width, height) of the video recording. If not set, it will default to the browser's viewport size. Example: `{"width": 1280, "height": 720}`
+
+#### `record_video_framerate`
```python
-save_recording_path: str | None = None
+record_video_framerate: int = 30
```
-Directory path for saving video recordings.
+The framerate for the video recording. Defaults to 30 frames per second.
#### `trace_path`
@@ -779,32 +797,6 @@ Client certificates to be used with requests.
Note: Browser Use also provides some of our own recording-related options not listed below (see above).
-#### `record_video_dir`
-
-<a name="record-video-dir"></a>
-<a name="save-recording-path"></a>
-
-```python
-record_video_dir: str | Path | None = None
-```
-
-Directory to save `.webm` video recordings. [Playwright Docs: `record_video_dir`](https://playwright.dev/python/docs/api/class-browsertype#browser-type-launch-persistent-context-option-record-video-dir)
-
-<Note>
- This parameter also has an alias `save_recording_path` for backwards
- compatibility with past versions, but we recommend using the standard
- Playwright name `record_video_dir` going forward.
-</Note>
-
-#### `record_video_size`
-
-```python
-record_video_size: dict | None = None. [Playwright Docs: `record_video_size`](https://playwright.dev/python/docs/api/class-browsertype#browser-type-launch-persistent-context-option-record-video-size)
-
-```
-
-Video size. Example: `{"width": 1280, "height": 720}`
-
#### `record_har_path`
<a name="record-har-path"></a>
diff --git a/examples/features/video_recording.py b/examples/features/video_recording.py
index 0a3f90783..87c7e1e10 100644
--- a/examples/features/video_recording.py
+++ b/examples/features/video_recording.py
@@ -11,7 +11,7 @@ async def main():
browser_session = BrowserSession(browser_profile=video_profile)
agent = Agent(
- task='Go to github.com/trending then navigate to the first trending repository.',
+ task='Go to github.com/trending then navigate to the first trending repository and report how many commits it has.',
llm=ChatOpenAI(model='gpt-4.1-mini'),
browser_session=browser_session,
)