You can control the output through prompt engineering.

For example, try with a prompt like "Provide a 100 word summary of the content at the following url and give me the url of the image on the page: [url]". You can expand the prompt as detailed as possible to get the most accurate output. Imaging explaining things to your colleague, use the same words to explain things to the model.

To dynamically change the size, you'll have to build the prompt using code and pass it on to Bedrock.