Djeex
974690132e
New layout:
src/script/app/ Python package (was app/)
src/assets/img/ SVGs (was static/*.svg)
src/assets/css/ stylesheet (was static/style.css)
src/templates/ HTML templates (was templates/)
illustration/ README screenshot (was screenshots/), kept at
the repo root since it's a doc asset, not
something the app serves
main.py, jobs/, VERSION, docker/ unchanged locations
main.py stays at the repo root as the entry point (matches how
Docker and local runs already invoke it) and prepends src/script to
sys.path before importing app, rather than moving into src/script/
itself and needing every entry point updated to match.
config.py's path constants are recomputed for app/config.py's new
depth (src/script/app/), with SRC_DIR and REPO_ROOT resolved
explicitly rather than a single ambiguous BASE_DIR, since "templates
and assets live under src/" and "jobs and VERSION live at the repo
root" are no longer the same directory. Flask's static_folder now
points at src/assets, which serves img/ and css/ under it at
whatever URL prefix Flask derives from the folder name (/assets, not
/static); templates already reference assets via the "static"
endpoint name through url_for, so this needed the url_for() calls'
filename argument updated with the img/ or css/ prefix, not a
hardcoded URL change.
Docker copies src/ into the image preserving this same internal
structure, so the path arithmetic in config.py resolves identically
in both a local checkout and the container, no --chdir or other
container-specific path handling needed.
Verified end to end after the move: local Flask dev server, a full
Docker rebuild, and a real HTTP conversion request through the
running container (TIFF HDR pipeline, numpy/tifffile imports
included) all succeed.
Note
⚠️ This is an heavily vibe-coded proof of concept — do not expose it to the internet — use it at your own risk. Github repo is a mirror of https://git.djeex.fr/Djeex/insta-hdr-converter. You'll find full package, history and release note there.
Instagram HDR Assembler
A small web front end for assembling Instagram-compatible HDR JPEGs (gain map) from an SDR export and an HDR export out of Lightroom or Camera Raw.
The gain-map assembly logic (src/script/app/assembler.py) is adapted from kostis-kounadis/instagram-hdr-assembler, itself based on the reverse-engineering work of karachungen/instagram-hdr-converter. Original MIT license kept in LICENSE.
How it works
- Upload an SDR file (
.jpg/.jpeg) and an HDR file: either a gain-map JPEG (.jpg/.jpeg) or Lightroom's 32-bit float linear HDR TIFF export (.tif/.tiff). - Both files must have identical dimensions and match one of Instagram's
feed resolutions exactly:
1080x1080(1:1),1080x1350(4:5), or1080x566(1.91:1). - The real HDR pixel data is obtained from the HDR file: decoded back
out of a gain-map JPEG (
ultrahdr_app, undoing its own embedded base image and gain map), or read directly from a linear TIFF and given HLG's system-gamma OOTF so it's on the same display-referred scale a gain-map JPEG's HDR intent would be. A gain map is only correct against the exact base it was computed from, and that's rarely the SDR file you upload, so reusing an existing one as-is would quietly reconstruct the wrong HDR look. Your SDR file is cleaned up and re-encoded at 4:2:0 chroma subsampling (Pillow), then a fresh gain map is computed against it and packaged into the final UltraHDR JPEG (ultrahdr_app). The output's visible image is exactly your SDR file; only the gain map is recomputed. The TIFF route skips the gain-map JPEG's 8-bit-in-8-bit precision ceiling, giving a more accurate result. - The validation report is shown on the result page, including the "GainMapMin is negative" warning when present, with a download link for the finished JPEG.
Uploads and results don't stick around: SDR/HDR originals are deleted
right after conversion, the result a few seconds after you leave the
result page (heartbeat + background reaper), and jobs/ is wiped on every
app startup.
Running locally, without Docker
pip install -r requirements.txt
# also install exiftool and compile libultrahdr, see docker/Dockerfile
# for the exact packages and commands (validated on Alpine)
PORT=5050 python3 main.py # PORT is optional, defaults to 5000
Running with Docker
docker compose -f docker/compose.yaml up -d --build
Opens on http://<host>:5050 (port 5000 collides with AirPlay Receiver on
macOS, hence 5050 by default; change it in docker/compose.yaml if
needed).
Running this on the open internet
Don't do it.