Create the RBAC permission module that maps roles to permissions and
provides a requirePermission middleware to replace requireAdmin. Update
the test server to use requirePermission for the admin health check.
Merge CPU, CUDA, and lite Docker images into a single unified image.
One tag (latest) works on all platforms: amd64 (NVIDIA CUDA) and arm64 (CPU).
GPU auto-detected at runtime. All ML models and packages baked in.
Key changes:
- Platform-conditional Dockerfile (nvidia/cuda on amd64, node on arm64)
- tini as PID 1 for proper signal handling
- Fix FILES_STORAGE_PATH data loss bug
- Fix RealESRGAN upscaler (was broken, always fell back to Lanczos)
- Fix PaddleOCR language codes and stdout corruption
- Simplified CI/CD (single build, single tag)
- Expanded model pre-download with verification
- Shutdown timeout, improved health endpoint
- Remove unused lama-cleaner
Users running lite mode had no way to tell why AI tools were greyed out.
Now the public health endpoint reports the variant, and a visible banner
appears in the tool panel when running in lite mode.
- Replace OpenCV Haar Cascades with MediaPipe for face detection, using
short-range model first with full-range fallback for better accuracy
- Add auto-orient to remove-background route for EXIF-rotated photos
- Change default background removal model from u2net to birefnet-general-lite
- Fix flaky test by setting SQLite busy_timeout before journal_mode pragma
Co-authored-by: Siddharth Kumar Sah <siddharth123sk@gmail.com>
Batch progress was broken because JobProgress events lacked a `type`
field. The frontend checks `data.type === "batch"` to distinguish batch
from single-file SSE events, so batch progress was silently discarded
and multi-file processing appeared stuck at 15%.
Also improves the processing UX for non-AI (Sharp-based) tools: the
progress bar now pulses during the server processing phase and shows
a "This may take a moment" hint after 10 seconds.
Co-authored-by: Siddharth Kumar Sah <siddharth123sk@gmail.com>
AI routes (remove-background, erase-object, ocr, blur-faces, upscale)
were silently swallowing errors - failures returned HTTP 422 to the
client but never appeared in server logs. This made it impossible for
self-hosters to diagnose issues like 504 timeouts from reverse proxies.
Adds request.log.info() at processing start (tool name, image size, key
settings) and request.log.error() in catch blocks, matching the existing
tool-factory pattern.
Co-authored-by: Siddharth Kumar Sah <siddharth123sk@gmail.com>
* feat: add resolveOutputFormat utility for input format preservation
* fix: preserve file order in batch processing with X-File-Results header
Collect all results before streaming the ZIP to guarantee upload order.
Replace X-File-Order with index-based X-File-Results header that maps
each upload index to its processed filename, handling failures and
duplicate filenames correctly.
Closes#13
* fix: use X-File-Results for index-based batch file matching
The frontend now matches processed files to entries by upload index
instead of fragile name/position matching.
* feat: preserve input format in smart-crop with quality control
Smart crop now outputs in the same format as the input (JPG in, JPG out)
instead of always converting to PNG. Adds an optional quality setting
(default 95) for lossy formats.
Closes#14
* feat: add output quality slider to smart crop settings UI
* feat: preserve input format in crop tool
* feat: preserve input format in color adjustment tools
Applies to brightness-contrast, saturation, color-channels, and
color-effects tool routes.
* refactor: avoid double encode in smart-crop content mode
For the simple trim path (no pad-to-square), chain .toFormat() on the
trim pipeline directly instead of creating a second Sharp instance.
This eliminates a redundant intermediate encode that degraded quality
for lossy formats. Also use trimmed.info dimensions instead of a
separate metadata() call for the pad-to-square path.
---------
Co-authored-by: Siddharth Kumar Sah <siddharth123sk@gmail.com>
Add a :cuda Docker image tag that auto-detects NVIDIA GPU at runtime
and falls back gracefully to CPU. Same pattern as Immich.
- New gpu.py shared utility for cached CUDA detection
- Background removal (rembg): pass CUDAExecutionProvider to ONNX Runtime
- Upscaling (Real-ESRGAN): use CUDA device + FP16 when GPU available
- OCR (PaddleOCR): enable use_gpu when CUDA detected
- Dispatcher reports GPU status at startup via readiness signal
- Admin health endpoint exposes GPU availability
- Dockerfile uses ARG GPU=false with conditional NVIDIA CUDA base image
- docker-compose.gpu.yml override for GPU users
- CI/CD workflows build and publish :cuda tag (amd64 only)
Three tags: :latest (CPU), :lite (no AI), :cuda (GPU with CPU fallback)
Linux libheif packages provide heif-convert instead of heif-dec (which
is macOS-only). The decoder now tries heif-convert first, then falls
back to heif-dec. Both accept the same argument syntax.
Adds a new mode that trims uniform-color borders around the subject,
like GIMP's "Crop to Content." Includes configurable tolerance threshold
and optional pad-to-square with target size for e-commerce workflows.
The original attention-based crop is preserved as "Focus Crop" mode.
Closes#7
- Health endpoint returns "healthy" instead of "ok" for consistency
- MAX_USERS now configurable via env var (default 5)
- People API returns team names instead of UUIDs in register/list
- PUT user update accepts team names (name-first lookup, fallback to ID)
- Login rate limit follows global rate limit when RATE_LIMIT_PER_MIN > 1000
- Strip-metadata preserves original format encoding instead of always PNG
- Fix e2e tests: rotate/crop/border button selectors match actual UI
- Fix e2e tests: create Engineering/Design teams in people test setup
- Fix e2e tests: people UI uses select for team field, not text input
- Update visual regression baseline for tablet home page
Worker thread initialization imports the tool registry which reads SQLite.
Under Docker volume filesystems, this can deadlock silently on SQLITE_BUSY,
causing APPLY to spin at 0% forever. Sharp operations complete in milliseconds
and don't need worker offloading. Added 30s AbortSignal timeout as defense
in depth for future re-enablement.
Siddharth Kumar Sah