after too much testing - including writing custom html parsers and checking cpython generated bytecode - this is the fastest general-purpose way to upgrade an html doc's images into amp i could come up with. if you expect a lot of images in a document, it would be faster to write a second function which ensure unique `imgs` and eliminates duplic…

amp_img_upgrade.py

import re
RE_img = re.compile("""<img[^>]+>""", re.I)

def upgrade(
    html,
    img_layout=None,          # STRING
    img_fallback=None,        # STRING
    img_noscript=None,        # BOOL
    img_noloading=None,       # BOOL
    img_default_height=None,  # must be a STRING, never an INT
    img_default_width=None,   # must be a STRING, never an INT
):
    imgs = RE_img.findall(html)
    tag_end = ' noloading >' if img_noloading else '>'
    img_fallback = "<div fallback>%s</div>" % img_fallback if img_fallback else ''
    for _img in imgs:
        _r_idx = -1  # ">" default
        if _img[-3:] == " />":
            _r_idx = -3
        elif _img[-2:] == "/>":
            _r_idx = -2
        elif _img[-2:] == " >":
            _r_idx = -2
        _tag_new = "<amp-" + _img[1:_r_idx]
        if img_layout and (' layout=' not in _tag_new):
            _tag_new += ' layout="' + img_layout + '"'
        if img_default_width and (' width=' not in _tag_new):
            _tag_new += ' width="' + img_default_width + '"'
        if img_default_height and (' height=' not in _tag_new):
            _tag_new += ' height="' + img_default_height + '"'
        _tag_new += tag_end
        _tag_new += img_fallback
        if img_noscript:
            _tag_new += "<noscript>%s</noscript></amp-img>" % _img
        else:
            _tag_new += '</amp-img>'
        html = html.replace(_img, _tag_new)
    return html