Nights of astrophotography in. Verified, plate-solved masters out. In about a minute, not an evening.
简体中文 · Why · Screenshots · Download · Quick start · Getting started · Docs · Contributing
Blink: every Light of a channel, aligned and stretched alike, with its measurements beside it. Light and dark follow the system. (Browser demo with labelled interface values; the native app shows your real frames and masters.)
- An evening of WBPP becomes about a minute. 61 Lights (26 MP, four nights, L R G B) took PixInsight WBPP 3.0.1 24 min 11 s on an M3 Pro. Ultra-Fast WBPP turned the same Lights into solved, verified masters on the same Mac in 76 s (features §1).
- Import everything, press Start. Every Light is measured: sky brightness, extinction, star counts, native-resolution PSF, trailing, clouds, occlusion and field agreement. Automatic screening, on by default, leaves out the frames the pipeline cannot use and lists each one with its reason and a preview; several panels of one target become a mosaic in the same run (automatic screening, mosaics).
- Or decide every frame yourself. Uncheck automatic screening and Blink shows the same numbers beside each frame and its reference; the run then applies exactly your decisions (Blink screening).
- The sky coordinates are verified, not assumed. Every master is plate-solved and checked against catalog stars before it is published. With the Gaia tiles installed, the WCS is refitted on catalog stars to about 0.15″ on real data (astrometry).
- Free and open. MIT licensed, runs on macOS Apple Silicon and Windows x64, needs no PixInsight. Your Lights are never written to, results go into a new folder, and nothing is uploaded anywhere.
| Ultra-Fast WBPP | PixInsight WBPP | |
|---|---|---|
| 61 × 26 MP Lights, four nights, L R G B, M3 Pro | 76 s | 24 min 11 s (WBPP 3.0.1, its default route with LocalNormalization) |
| Frame selection | measured evidence per Light; automatic screening by default, with cross-night baselines, suspect nights measured inside the integration and region weight maps for partial cloud, or your own decisions in Blink | whole-frame weights and threshold rejection |
| Calibration pairing | WBPP's rules, night by night (NIGHT_…, DATE_… folders); a folder prepared for WBPP imports as it is |
WBPP's rules |
| Plate solving | every master solved and verified on catalog stars; refitted to TAN-SIP up to order 3 on Gaia stars when the tiles are installed | astrometric solution in the default route |
| Mosaics | several panels placed by one astrometric fit of all panels on one shared canvas, matched by overlap stars; a real six-panel M31 (713 Lights) publishes | not part of WBPP |
| Trust | read-only sources, create-only results, receipts with hashes and algorithm ids, verified WCS before "done", bit-identical reruns of mono masters | console log |
| Master quality | historical G₈ 1.01–1.03 against the WBPP masters of the same data; full equivalence not certified | the reference |
| Drizzle | native 1×–4× from exactly the integration's inputs, Bayer drizzle | DrizzleIntegration |
| Platforms, license | macOS 15+ Apple Silicon, Windows x64 · MIT, no PixInsight needed | macOS, Windows, Linux · commercial |
| After preprocessing | linear masters, SuperL.fits, previews, receipts; no post-processing |
a complete platform |
The WBPP time is one measurement of its default route on identical inputs; the two tools' stage lists are not identical, and the reference Lights are no longer available for a re-measurement. Later builds measured the same project at 64–67 s, and at 79 s with the Gaia catalog refit, one run each (CHANGELOG). The complete comparison with the evidence behind every row, and the list of what this project does not do, is docs/features.md.
- You shoot mono or one-shot-colour deep sky with N.I.N.A. or a similar sequencer, over several nights, and want calibrated, stacked, plate-solved linear masters without babysitting a script for an evening.
- You own PixInsight and want to keep it for post-processing: the masters open there as ordinary linear FITS with a verified WCS.
- You do not own PixInsight and want a free preprocessing path whose results are measured against WBPP's rather than described.
- You want to see why a frame was kept or dropped, and to overrule the machine when you disagree.
It is preprocessing only: no stretching, gradient removal, deconvolution or colour calibration.
| 1 · Import | 2 · Blink |
|---|---|
![]() |
![]() |
| 3 · Processing | 4 · Result |
![]() |
![]() |
- Import. Drop all your nights at once: Lights, raw Flat/Dark/Bias or existing masters, N.I.N.A. folder layouts, or a folder you prepared for PixInsight WBPP. Types, targets and filters come from the headers; calibration is paired by WBPP's rules, night by night (calibration).
- Screening, automatic or in Blink. With Automatic screening checked (the default) the run leaves out the frames it cannot use, and the result lists each one with its reason and a preview. Unchecked, Blink gives each channel a reference frame and one shared stretch: step or play through the Lights, compare with the reference, keep or drop a frame or a whole night, undo. The flags are measured hints; the decisions are yours, and each channel is confirmed explicitly.
- Processing. Start processing becomes available once an output folder is chosen and the solver is ready, and, with automatic screening unchecked, once every Light has been reviewed and every channel confirmed. Calibration, registration, normalization, integration, optional drizzle and the per-channel plate solve run as one job with live progress.
- Result. One linear FITS per filter,
RGB.fitsorLRGB.fitswhen R, G and B exist,SuperL.fits(a super luminance of the broadband channels, on by default),previews/,receipt.jsonanddetails/. The app shows done only after the receipt, the product hashes and the final WCS have been verified.
Screenshots are rendered from the app's labelled browser demo by apps/desktop/scripts/screenshots.py; the numbers in them are interface placeholders, not measurements. The step-by-step guide with an FAQ is docs/getting-started.md.
Releases will appear on the Releases page with checksums and bundle attestations. Today the way in is building from source (guide): one command chain on a Mac,
make bootstrap BOOTSTRAP_PYTHON=python3.12
make desktop-dev # runs the app from the clone
# or, for a local app bundle:
CARGO_PROFILE_RELEASE_STRIP=none make desktop-build-macos-prereleaseWindows builds are described in docs/windows.md.
- macOS. The app needs macOS 15 or later. It is ad-hoc signed and not notarized. On macOS 15 and later the right-click → Open bypass no longer works: open System Settings → Privacy & Security → Open Anyway, or run
xattr -dr com.apple.quarantine "Ultra-Fast WBPP.app". - Windows. The installer is unsigned; SmartScreen shows "Windows protected your PC" → More info → Run anyway.
Nothing is bundled, for licensing reasons (licensing), so each computer needs a solver:
- Install the solver. macOS:
brew install astrometry-net(givessolve-field). Windows: ASTAP (astap_cli.exe) with a star database such as D20 or D50 unpacked into the ASTAP folder (Windows). - Download the index set in the app. Under Advanced options and solver setup, accept the provider terms and download the managed Astrometry.net index set (about 350 MB), then press Recheck setup.
- Optional, for the catalog refit. The distortion-corrected WCS (about 0.15″ on real data) needs the Gaia DR2 LITE tiles (587 MiB). Install them in the same section (Gaia DR2 tiles (optional, recommended for mosaics)), or from a terminal:
ultra-fast-wbpp catalog install astrometry-net-5205-lite --accept-provider-terms <id from catalog list> --progress-json(offline solver catalogs). In a built macOS app the engine isUltra-Fast WBPP.app/Contents/Resources/resources/ufwbpp-engine/ufwbpp-engine-aarch64-apple-darwin/ufwbpp-engine-aarch64-apple-darwin.
Moving to another computer: copy the catalog folder (<data root>/catalogs/astrometry-net) to the same data root there and open the app; the engine re-points its config and receipts to the new place, while the solver program itself must be installed on each computer (moving the catalog).
Import your folders (all panels of a mosaic at once), choose an output folder (remembered next time) and press Start processing; automatic screening picks the Lights. To decide every frame yourself, uncheck Automatic screening and review every channel in Blink first. To try the interface without data, make demo opens the labelled browser demo. The language defaults to English; the selector is in the toolbar and the choice is remembered. The first-run guide: docs/getting-started.md.
The same engine runs from the command line:
.venv/bin/ultra-fast-wbpp run /data/lights /data/calibration --recipe docs/recipes/mono-standard.json --output /data/new-result --progress-jsonultra-fast-wbpp doctor --json reports the hardware, the native kernels and the solver readiness; run-project is the multi-target, multi-filter route the desktop uses. The headless default selection is legacy-gate; { "selection": { "policy": "unattended-v2" } } in the recipe turns on the automatic screening the desktop uses (automatic screening).
- Alpha, under active development. Real-data runs are validated on one M3 Pro (macOS) and one Ryzen 7 5800H laptop (Windows 11); other machines run generic, unmeasured profiles. The register of what is and is not validated is docs/validation-matrix.md.
- Mono is validated on real data. One-shot colour (Bayer) Lights are calibrated as mosaics, debayered into R/G/B channel masters and combined into RGB (recipe), but only on synthetic data so far. Mosaics are validated on real M31 mosaics (2×2, five and six panels) (recipe); drizzle onto the mosaic canvas has synthetic pipeline tests only.
- Bit-identical reruns hold for mono masters. RGB pixels are identical between reruns while the file's checksum cards differ.
- Automatic screening is the desktop default (
unattended-v2). Unchecked, the Blink review is required and your decisions control admission. Its thresholds and the Blink flags come from one site and camera and cannot guarantee detection of every defect. The command line's default remainslegacy-gate. - Plate solving needs a solver on each computer (above). Nothing is uploaded anywhere.
- Disk space for full-resolution intermediate frames: about 12 bytes per pixel per Light on the output volume, which may be an external exFAT disk.
- Inputs: FITS and XISF Lights, raw calibration frames or masters (including PixInsight WBPP's XISF masters). Not supported: DSLR RAW files, and Lights that WBPP already calibrated or registered (they are refused; import the raw Lights).
Each item links to its CHANGELOG section.
- Automatic screening by default, and mosaics in one click: a project runs from import to its masters, or to the mosaic of several panels, without a review. On RCS2 J0327-1326 (350 Lights) it kept every Light of a hand selection and 29 more, with binned noise equal or lower (G₈ 1.005–1.017).
- Mosaics on one shared canvas, placed by one fit of all panels, and drizzled onto that canvas. A real six-panel M31 mosaic of 713 Lights publishes (validation matrix).
- A catalog-refitted WCS: TAN-SIP up to order 3 on Gaia stars, about 0.15″ on held-out stars on real masters, with the Gaia DR2 tiles and a verification robust to old Tycho-2 positions.
- A filter whose own solve misses the gate takes its grid's solve, verified by the refit; a run whose solves all miss only the RMS gate still reaches the refit.
- Projects with narrowband or partial filter sets publish.
- Screening keeps uniform thin cloud and a night whose optics elongate every star, and fails a Light taken off the field.
- Importing folders arranged for PixInsight WBPP and calibration paired by WBPP's rules.
SuperL.fits: the broadband channels combined into one deeper luminance, each weighted by its noise; a channel with clearly worse stars is left out. The masters are unchanged.
| What | Result | Where |
|---|---|---|
| 61 Lights · 4 nights · L R G B · M3 Pro | 76 s end to end (170 → 97 → 82 → 76 s from three changes within days) | features §1, CHANGELOG |
| PixInsight WBPP 3.0.1, same Lights, same Mac | 24 min 11 s | features §1 |
| Same project on a Ryzen 7 5800H laptop, Windows 11 | 267–289 s | Windows |
| Two runs of one project on one machine | mono masters bit-identical; RGB pixels identical, checksum cards differ | validation matrix |
| Masters vs PixInsight WBPP (same frames) | historical G₈ 1.01–1.03; full equivalence not certified | features §5, evaluation standard |
| 2× drizzle vs the Lanczos-3 master | half-light radius 2.5–4.4 % smaller, effective noise 6–7 % lower, flux ratio 0.993–0.994 | drizzle recipe |
| Catalog refit of a real master | 0.157″ on held-out Gaia stars (NGC 7331), against the solver's 2.58″ on the same stars | astrometry, CHANGELOG |
This independent implementation does not claim algorithmic or pixel equivalence to PixInsight/WBPP; the evaluation standard is how it is compared.
flowchart LR
A["Import<br/>N.I.N.A. or WBPP folders · FITS / XISF<br/>raw Flat · Dark · Bias · masters"] --> B["Screening, automatic or Blink<br/>sky · extinction · PSF · trails<br/>clouds · occlusion · field"]
B --> C["Calibration<br/>paired by WBPP's rules<br/>masters built or reused"]
C --> D["Registration<br/>Lanczos-3 · rotations · meridian flips"]
D --> E["Normalization<br/>stellar scale · sky response"]
E --> F["Integration<br/>rejection · trail corridors<br/>weights · optional drizzle"]
F --> G["Plate solve<br/>Astrometry.net / ASTAP<br/>catalog verification · Gaia refit"]
G --> H["Publish<br/>masters · SuperL · previews · receipts"]
The hot loops are multithreaded native kernels that reproduce the NumPy reference value for value; a Light is decoded once, calibrated in memory and warped straight into the stack. The execution path, boundaries and science are in docs/architecture.md; the user guides are the recipes.
| Directory | Responsibility |
|---|---|
apps/desktop |
React interface and Tauri desktop bridge; it runs the engine's command line (guide) |
packages/engine |
The ultra-fast-wbpp engine (Python package ufwbpp): workflows, stacking, selection, solvers, products, CLI |
packages/light-frame-qc, packages/registration |
Light measurements, native PSF, quality gate, Blink flags; star detection, transform estimation, warping |
native |
C++ kernels (warp, rejection, reduction, drizzle, debayer, Lanczos table) and Metal |
tools/validation, benchmarks |
Master evaluation and tolerance gate; run tracer, kernel benchmarks (dated evidence in docs/evidence) |
scripts, packaging |
Native build chain, engine sidecar packaging, bundle attestation, public-tree and link checks |
docs |
Index · getting started · features · architecture · recipes · validation |
Contributing: CONTRIBUTING.md for setup and rules, AGENTS.md for the full working guide. make test runs the Python, Rust, frontend and native tests; make check adds formatting, lint and the public-tree checks.
Original code is MIT licensed; keep the copyright and license notice when redistributing, including commercially (NOTICE suggests an attribution). Third-party components keep their own licenses (licensing, third-party notices); the Python runtime imports no GPL-licensed library.
Not affiliated with PixInsight, Pleiades Astrophoto, N.I.N.A., Astrometry.net or ASTAP; contains no PixInsight/PCL source or binaries.


