PDF Lab / Original specimens

OBSERVED · REPRODUCIBLE · LIMITED

Why a phone photo can turn sideways in a PDF

We imported an original chart with EXIF orientations 1–8 through the actual workbench. Orientation 6 changes both rotation and pixel dimensions.

Start with the problem

Symptom: A photo is upright in a viewer but appears sideways or swaps width and height in the PDF.
Main cause: JPEG pixel order and EXIF display orientation are separate. Different handling of the orientation tag can change how the same file appears.
What to do now: Check the top edge and subject direction immediately after import, then compare the reviewed and downloaded PDF. If it is wrong, save the source photo upright and select it again. For an image inserted into an existing PDF, you can also adjust Rotation in the editor. The app cannot infer an incorrect tag’s intent or straighten a tilted scene. Orientation-1 JPEG location metadata may need separate removal.

Use the same TOP strip and corner A as landmarks

These PDFs use the same test chart with orientation tags 1 and 6. They are two tag-controlled outputs, not a faulty photo followed by its correction.

Tag 1 output · TOP along the top. A is at the upper left and B at the upper right. The inserted chart measures 60×40 mm. — **Tag 1 output · TOP along the top**
A is at the upper left and B at the upper right. The inserted chart measures 60×40 mm.

Tag 6 output · TOP along the right. A moves to the upper right. The tag’s 90-degree clockwise rotation changes placement to 40×60 mm. — **Tag 6 output · TOP along the right**
A moves to the upper right. The tag’s 90-degree clockwise rotation changes placement to 40×60 mm.

Actual PDF crops: same region, 216 dpi. Select an image to open the full PDF.

What changed

The TOP strip and corner A move with the tag, and the stored image dimensions swap. These views use the same PDF region and scale; neither image is stretched to match the other.

Why it matters

A photo viewer that applies EXIF may rotate the original JPEG before you compare it. Use the subject and fixed landmarks, rather than thumbnail shape, to judge the PDF’s intended orientation.

What to do

Follow TOP and A through the source JPEG, normalized PNG and saved PDF. For a real photo, use the subject’s top edge. Save the source upright or adjust Rotation if the result differs from your intent.

Compare full pages and open the original PDFs

Measured observations

Item	Input / reference	Observed output
EXIF 1	1200 × 800	Pixel mean error 0.000
EXIF 2	1200 × 800	Pixel mean error 0.000
EXIF 3	1200 × 800	Pixel mean error 0.000
EXIF 4	1200 × 800	Pixel mean error 0.000
EXIF 5	800 × 1200	Pixel mean error 0.000
EXIF 6	800 × 1200	Pixel mean error 0.000
EXIF 7	800 × 1200	Pixel mean error 0.000
EXIF 8	800 × 1200	Pixel mean error 0.000

These observations apply to these specimens, not every PDF or reader.

When it happens

JPEG pixel order and EXIF display orientation are separate. Readers can disagree if they handle the tag differently. This controlled experiment changes tags on a 1200×800 original chart; it is not a test of particular phone cameras.

What we did

Using the Chrome file picker, we imported eight JPEGs and saved projects and checked PDFs. Orientation 1 retains JPEG; the others become upright PNGs after browser decoding. Independent Pillow EXIF transforms are compared with stored pixels. We also inspect embedded image sizes and retained text.

Compare orientation 6

Locate TOP and corners A–D in the source. A photo viewer may already rotate it automatically.
Open the normalized PNG and check that its pixels, rather than an orientation tag, are portrait-oriented.
Distinguish the small added chart from the original landscape chart in the PDF.
For a photo record, also align the subject and viewpoint of each before-and-after pair.

Exceptions and privacy

The app cannot infer intent from an incorrect tag or straighten a tilted scene. Orientation-1 JPEGs retain their source compression stream and may retain EXIF location metadata. Remove sensitive metadata before import. Physical phone capture and OS clipboard workflows remain separately unverified.

Files to inspect yourself

Independent observations and method (JSON) · Authorship and file hashes

Crop coordinates, scale, source hashes and fresh extraction · Rechecked: 2026-10-24 · pdftoppm version 26.07.0 · pypdf 6.10.0 · pdfplumber 0.11.9. Original PDFs were not modified.

Environment, date and evidence boundaries

Checked: 2026-10-23 · Windows / Chrome 155 · pdf-lib 1.17.1 · PDF.js 6.4.299 · fontkit 1.1.1 · pypdf 6.10.0 · pdftoppm version 26.07.0. PDFs and artwork are original; NanumGothic uses OFL-1.1. Tables and files record direct observations; explanations interpret those structures. Physical OS keyboard input, phones and every PDF reader are outside this test.

Official reference explanations (separate from measurements): Pillow: EXIF transpose reference

Reproduction code: scripts/lab-inputs.py, lab-outputs.mjs, lab-derived.py, tests/readiness-browser.mjs and tests/lab-pdf.py.

Keep this guide handy

Keep the guide address or share it with someone who needs these instructions. It contains no personal PDF or editable project.

Public guide address