How to write a multipage TIFF with different ImageFileDirectories for each frame?

[anntzer]

What did you do?

I would like to write a multipage tiff file with different ImageFileDirectories for each frame. I started with the example given at #3636 (comment) (#3636 being essentially the same issue).

What did you expect to happen?

The example code should print 1 2 3 4.

What actually happened?

The code prints 1 2 3 4, then crashes with

Exception ignored in: <PIL.TiffImagePlugin.AppendingTiffWriter object at 0x101bb2930>
Traceback (most recent call last):
  File "/path/to/lib/python3.13/site-packages/PIL/TiffImagePlugin.py", line 2197, in close
  File "/path/to/lib/python3.13/site-packages/PIL/TiffImagePlugin.py", line 2068, in finalize
ValueError: seek of closed file

What are your OS, Python and Pillow versions?

--------------------------------------------------------------------
Pillow 11.1.0
Python 3.13.0 | packaged by conda-forge | (main, Oct 17 2024, 12:32:35) [Clang 17.0.6 ]
--------------------------------------------------------------------
Python executable is /path/to/miniforge3/envs/default/bin/python
System Python files loaded from /path/to/miniforge3/envs/default
--------------------------------------------------------------------
Python Pillow modules loaded from /path/to/.local/lib/python3.13/site-packages/PIL
Binary Pillow modules loaded from /path/to/.local/lib/python3.13/site-packages/PIL
--------------------------------------------------------------------
--- PIL CORE support ok, compiled for 11.1.0
--- TKINTER support ok, loaded 8.6
--- FREETYPE2 support ok, loaded 2.13.2
--- LITTLECMS2 support ok, loaded 2.16
--- WEBP support ok, loaded 1.5.0
--- JPEG support ok, compiled for libjpeg-turbo 3.1.0
--- OPENJPEG (JPEG2000) support ok, loaded 2.5.3
--- ZLIB (PNG/ZIP) support ok, loaded 1.3.1.zlib-ng, compiled for zlib-ng 2.2.2
--- LIBTIFF support ok, loaded 4.6.0
*** RAQM (Bidirectional Text) support not installed
*** LIBIMAGEQUANT (Quantization method) support not installed
--- XCB (X protocol) support ok
--------------------------------------------------------------------

A few notes:

• The crash can be worked around by making the file-handle closure a no-op (fp.close = lambda: None immediately after opening the file), but that's a not-so-nice hack.
• The example code given in the linked issue relies on the private TiffImagePlugin._save method; it would be nice if a (perhaps simpler) public API was available.
• The example code sets a different encoderinfo on each frame, and as such actually runs afoul of the (limited...) documentation regarding encoderinfo at https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html#reading-multi-frame-tiff-images: "for correct results, all the appended images should have the same encoderinfo and encoderconfig properties."

[radarhere]

The code from #3636 (comment) is

from PIL import Image, ImageSequence, TiffImagePlugin

# Preparation
import numpy as np
images = [np.asarray(im) for im in [
	Image.new("RGB", (100, 100), "#f00"),
	Image.new("RGB", (100, 100), "#0f0")
]]
listOfTuples = [
	(55, [1, 3]),
	(56, [2, 4])
]

# Script
image_list = []
for image in images:
	image_list.append(Image.fromarray(image))
frames = []
for i, frame in enumerate(image_list):
	info = TiffImagePlugin.ImageFileDirectory()
	
	for tupleIdTag, tupleTags in listOfTuples:
		info[tupleIdTag] = tupleTags[i]
		info.tagtype[tupleIdTag] = 3
	
	frame.encoderinfo = {'tiffinfo': info}
	frames.append(frame)
with open("out.tiff", "w+b") as fp:
	with TiffImagePlugin.AppendingTiffWriter(fp) as tf:
		for frame in frames:
			frame.encoderconfig = ()
			TiffImagePlugin._save(frame, tf, "out.tiff")
			tf.newFrame()

reloaded = Image.open("out.tiff")
for tupleIdTag, _ in listOfTuples:
	print(reloaded.tag_v2.get(tupleIdTag))
reloaded.seek(1)
for tupleIdTag, _ in listOfTuples:
	print(reloaded.tag_v2.get(tupleIdTag))

If I run that with Pillow 11.1.0 in Python 3.11, I get

1
2
3
4

So I'm unable to replicate. Is there something specific that you've changed that triggers the crash?

[anntzer]

I confirm that the error does occur with this code, but looking at it again it happens when the script exits (e.g. if the script is in a test.py file and you run python test.py), not if you run it e.g. from a notebook.

[radarhere]

Running python3 test.py, I was unable to replicate it in GitHub Actions - see radarhere@5df9eb4 and https://github.com/radarhere/Pillow/actions/runs/13538366310

The example code given in the linked issue relies on the private TiffImagePlugin._save method; it would be nice if a (perhaps simpler) public API was available.

Thanks to #8483, you can now replace

frame.encoderconfig = ()
TiffImagePlugin._save(frame, tf, "out.tiff")

with

frame.save(tf, "TIFF")

[radarhere]

The example code sets a different encoderinfo on each frame, and as such actually runs afoul of the (limited...) documentation regarding encoderinfo at https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html#reading-multi-frame-tiff-images: "for correct results, all the appended images should have the same encoderinfo and encoderconfig properties."

https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html#reading-multi-frame-tiff-images lists that under append_images

append_images
A list of images to append as additional frames. Each of the images in the list can be single or multiframe images. Note however, that for correct results, all the appended images should have the same encoderinfo and encoderconfig properties.

This is because _save_all overwrites encoderinfo and encoderconfig on each subsequent image with the data from the first image.

Pillow/src/PIL/TiffImagePlugin.py

Lines 2297 to 2309 in d7d48df

	def _save_all(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
	encoderinfo = im.encoderinfo.copy()
	encoderconfig = im.encoderconfig
	append_images = list(encoderinfo.get("append_images", []))
	if not hasattr(im, "n_frames") and not append_images:
	return _save(im, fp, filename)
	cur_idx = im.tell()
	try:
	with AppendingTiffWriter(fp) as tf:
	for ims in [im] + append_images:
	ims.encoderinfo = encoderinfo
	ims.encoderconfig = encoderconfig

The documentation isn't incorrect - the documentation is talking about append_images, which the script that we've been talking about doesn't use.

[anntzer]

re: your CI test: actually CI did print the error (even if the run didn't fail), see https://github.com/radarhere/Pillow/actions/runs/13538366310/job/37834022564#step:11:28

re: encoderinfo: Sure, the referenced docs only mention the use of encoderinfo for append_images, but this paragraph is actually the only mention of encoderinfo in the entire docs (wrt. TIFF) so in the absence of any mention of the contrary it seems not unreasonable to take that as a general statement.

[radarhere]

Ok, I've created #8780 to resolve this. It improves the documentation, and also allows

    with TiffImagePlugin.AppendingTiffWriter(fp) as tf:
        for frame in frames:
            frame.encoderconfig = ()
            TiffImagePlugin._save(frame, tf, "out.tiff")
            tf.newFrame()

to be replaced with

    frames[0].save("out.tiff", save_all=True, append_images=frames[1:])

[anntzer]

The patch works great, thanks for the quick fix.
As a minor usability improvement, may I also suggest that (as a general rule, not restricted to TIFF) setting append_images to a non-empty list also implicitly requests that they be indeed saved, regardless of whether save_all had been set? After all, it seems a bit strange to pass append_images if one didn't actually want to save them...

[radarhere]

Ok, I've created #8781