The three algorithms in Compress PDF do different work. Choosing one is a decision about what your output must retain, not just how small you want it to be. This guide is based on CommandPDF’s processor and interface. It contains no measured compression benchmark and makes no comparative speed or size claim.
Standard: structure first, optional image processing second
Standard sends the input to the compression worker to rewrite its structure. With Optimize embedded images enabled, the resulting bytes then go through a separate image-optimization pass using the selected quality preset.
That second pass can fail independently—for example, if the processing engine cannot load. The processor catches that failure and returns the worker’s structure-compressed result. This keeps the operation usable, but it means the displayed file size is the evidence of what you obtained; selecting image optimization is not proof that it completed.
Low, Medium, High and Maximum describe output quality for the image pass. A higher quality preset is not “more compression.” The amount saved depends on the input, so two files with the same page count can produce very different results.
Condense: optimize images and clean the structure
Condense calls the same image-and-structure optimization routine used by Standard’s optional second pass, without first running the separate compression worker. It can re-encode sufficiently large embedded images as JPEG, replace them when the routine finds a size saving, and clean and recompress the document’s structure. The quality preset applies here too.
It works with existing PDF objects rather than taking a screenshot of every page. That makes it different from Photon, but it does not make it lossless. Compare photographs, fine graphics and small text in scanned images, as well as the document’s interactive features.
An intention to retain interactivity is not a guarantee for every PDF. A document can contain unusual annotations, embedded files, form scripts or signatures. Open the output and exercise the features your recipient needs. Keep the source so you can return to it if the rewritten copy behaves differently.
Photon: rebuild pages from rendered images
Photon renders each page at an output DPI and creates a new PDF from those images. This makes its tradeoffs easier to understand:
- Text becomes pixels rather than the original selectable characters.
- Form fields and other interactive objects should not be expected to survive as editable controls.
- Fine lines and small text depend on the selected pixel resolution.
- An already compact, text-based PDF may become larger when pages are turned into images.
For an A4-sized page, a 150-DPI render is roughly 1,240 by 1,754 pixels. Doubling the DPI approximately quadruples the pixel count, because both width and height double. This is geometry, not a benchmark of the final file size: image content and encoding also affect the bytes.
Photon has a dedicated DPI control. The general quality selector is not passed to Photon as the same Low-to-Maximum preset, so use the algorithm’s own controls and inspect its output.
A repeatable comparison on your own document
Use a non-confidential copy that represents the documents you actually work with. Always start each run from that same original; compressing the previous output changes the input and makes the comparison misleading.
Record the original byte size and the features that must survive. For each run, note:
- The algorithm and its settings, including whether image optimization or metadata removal was selected.
- The downloaded size in bytes.
- Whether a known phrase remains searchable and copyable.
- Whether required links, bookmarks and fields still function.
- Whether the smallest text and fine graphics are acceptable at the intended viewing or print size.
Calculate the size reduction as (original bytes - output bytes) / original bytes × 100. A negative result means the file grew. Report the actual byte sizes with the percentage; a large percentage on a small file may be unimportant for your delivery limit.
Do not call an output “lossless” solely because it looks similar at normal zoom. Visual appearance, extractable text, interactive fields and metadata are different properties. A file can preserve one and lose another.
What to do with an unexpected result
The file grew: keep the original, try Standard with image optimization disabled, or consider whether you really need an image-only copy. Rasterizing clean text is often the wrong tradeoff.
Savings were small: the document may already store its content efficiently, or Standard’s optional image pass may not have completed. Do not infer an error from a small percentage alone.
The document looks fine but search fails: compare it with the original. Photon replaces the original text layer with pixels; scanned inputs may already be image-only.
The size target can only be met by making text unreadable: split the document if the recipient permits multiple files, remove unneeded pages, or use another accepted delivery method. A smaller file that cannot be used has not solved the task.
For an attachment-specific checklist, follow compressing for a file-size limit. To practice checking page order and output structure without private documents, use the sample merge and split packet.