Methods and systems for compressing images include generating custom quantization tables for quantizing frequency information associated with an image. Specifically, one or more embodiments determine acceptable error percentages during compression of a digital image based on content of the digital image. For example, the acceptable error percentages are defined by compression error thresholds that limit how much error a quantizer in a quantization table can introduce during compression of the digital image. One or more embodiments generate the custom quantization table by determining quantizer values that produce compression errors that meet the compression error thresholds. One or more embodiments compress the digital image using the custom quantization table.