I am currently experimenting with after effects and premiere to see what sort of strange things they might be capable of and was hoping someone had some insight on an Idea I had.

In this situation I have some text, such as some subtitles or even a time code, that is burned into the footage. I do not have any of the project files for this footage, the only thing I have to work with is the final video.

Is there any way to have either Adobe Premiere or After Effects recognize what text is on screen automatically? Bonus points if it can then take this information and feed it into a new title element.

P.S. This is all assuming the source video's text is of decent quality (i.e. the text is static, decently distinguished from the video behind it, and not too tiny.)

  • You could export the frames as a frame sequence and then use OCR software on the frames. This would be a job for shell scripting or a programming language rather than AE or PP though.
    – stib
    Commented Feb 11, 2016 at 3:10
  • @stib if you add this as an answer and provide some links to suggested software I will accept it.
    – Malco
    Commented Feb 11, 2016 at 14:12

1 Answer 1


I do not know of any existing solutions for this in Premiere or AE.

However, if you know a little about programming, you might be able to write a script using OpenCV that runs Tesseract OCR on video frames, and comes out with the text you want. Once you have the text and time codes, you definitely could write a script to build the titles at the appropriate time codes inside of AE or Premiere.

(I became aware of these open-source tools since I am currently building a mobile app that runs OCR on documents.)

Check out the Scene Text Recognition docs for OpenCV, if you want to try that route.

  • So for this method you would need to output a series of still frames correct? Or does tesseract accept Video? If it doesn't accept video how would you easily tie the frame number to a timecode? Just count the frames?
    – Malco
    Commented Jul 15, 2016 at 13:17
  • OpenCV can handle video, and the Scene Text Recognition docs I pointed to show how to call Tesseract on a frame through OpenCV. They have some C++ sample code there that might get you started, but OpenCV is also useable from Python, if that's more comfortable. Commented Jul 16, 2016 at 17:53
  • As for timecode, OpenCV works in integer frame count, but you could use that directly when scripting in AE (and I assume Premiere), assuming the frame rate of the sequence matches. Commented Jul 16, 2016 at 17:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.