This one seems much more likely to work for its intended purpose. Even if an LLM can be trained to read it, it will probably take much more processing to get the text out of a video compared to an image.