How can I totally flatten a PDF in Mac OS on the command line?
Try using pdf2ps and ps2pdf in succession. It's a little cumbersome but it helped me reduce my 30MB document down to 17MB. Not as much as I would have liked but still better. I got this idea from a friend who used this method to get rid of password permissions on a pdf.
pdf2ps your_pdf_file.pdf your_pdf_file.ps
ps2pdf your_pdf_file.ps your_pdf_file_from_ps.pdf
Hope that helps!
EDIT 2020-07-25 This used to work but doesn't anymore. Now these commands continue to preserve the PDF layers.
So this is what I've been using lately. I think functionally it does the same thing as pdf2ps file.pdf - | ps2pdf - file_flat.pdf, but it seemed to work better for me.
gs -sDEVICE=pdfwrite -dPDFSETTINGS=/default -dNOPAUSE -dQUIET -dBATCH -sOutputFile=<newfile> <oldfile>
WARNING: gs and pdf2ps|ps2pdf DO NOT Flatten PDFs!
Using gs or pdf2ps followed by ps2pdf will yield a multi-layer PDF with the content under annotations present in original form. You can verify this flaw in Preview by using Select All, then Copy, then Paste into a TextEdit window (in rich text mode). You will see the text or graphics under redaction annotations for example. This is clearly very bad if you legally need that content to be gone from the output.
A Working Solution
ImageMagick can produce a configurable quality, multi-page, single-layer flattened PDF with rasters of each page using the following command:
convert -density 150 document_original.pdf document_flat.pdf
This command rasterizes document_original.pdf, making an pixel-based image of each page, at 150 DPI, and outputs the result as document_flat.pdf.
A Note on Image Quality
Due to the rasterization, it produces a non-scalable (zoom and you'll see the text or original vector images become pixellated) PDF. It will likely have a larger filesize unless the original has very complex vector content like million-point scatter plots.
By changing density, you can trade larger file size for higher resolution output.
All text will be converted to raw pixels in each page image. Text and vector diagrams suffer the most, so experiment with the DPI until you get usable output files.