Current example evaluation
57-frame source video
This local version evaluates the new calibration order on every frame. The first calibration removes stationary horizontal, vertical and non-separable sensor structure. The second stage applies an adaptive per-frame cloud pseudoflat controlled by the unrest index.
Cloud correction can normalise broad background and transparency variations, but it cannot reconstruct stars or comet signal physically blocked by cloud.
Uncalibrated sequence
Geometric alignment without sensor or cloud correction.
Star reference — expanded
Object reference — expanded
Star reference — common
Object reference — common
Mean products
Stationary fixed-pattern calibration
Persistent bad pixels and additive sensor bands are removed before a broad multiplicative flat is applied.
Original
Star reference — expanded
Object reference — expanded
Star reference — common
Object reference — common
Mean products
Pseudoflat plus unrest-index cloud correction
A sigma-24 dynamic pseudoflat is applied per frame. Measured on the common star field, the median unrest index falls from 6.042 to 0.713 (88.2% lower).
Original
Star reference — expanded
Object reference — expanded
Star reference — common
Object reference — common
Mean products
Gaia DR3 field census
Stellar population
The diagrams use the complete Gaia DR3 field sample returned down to G = 13.5. Click either graphic to open it at its native resolution.
Download Gaia field photometry CSV Download HR-diagram source CSV
Development transparency
Video Align was developed by Jost Jahn with assistance from OpenAI Codex. AI assisted with drafting, reviewing and testing code and documentation. Scientific objectives, processing decisions, evaluation and release approval remain under human responsibility.
No AI model is embedded in Video Align, and no AI service receives the user’s videos. Star registration, calibration, object tracking and aggregation use deterministic algorithms running locally.
The application uses Python, NumPy, OpenCV, Astropy, FFmpeg and PyInstaller. AI-assisted code can contain errors; scientific results should be independently reviewed and compared with measured calibration frames whenever possible.
Contrast and tail dynamics
Normal and inverted 3×3 contrast matrices
All nine variants use every frame and preserve the original object-centred pixel scale. Rows change the black-point percentile; columns change the asinh strength. The recommended variant is black p1, asinh 4 with a diffuse-structure score of 0.969.
Normal contrast matrix
Inverted contrast matrix
Recommended rendering
Recommended · score 0.969
Normal
Black point p1 (59 DN), white point p99.9 (207 DN), asinh strength 4.
Recommended · inverted
Inverted
The same pixels and parameters, with luminance inverted for dark-detail inspection.
Exploratory physical estimate
Candidate moving knot in the tail
A temporal star model was subtracted in the star-aligned coordinates, fixed stellar cores were masked, and the residual was centred on the comet. A weak coherent ridge then moves outward by 0.3125 px/source frame. It is above one robust sigma in only 36.8% of frames, so this is a low-confidence candidate rather than an unambiguous knot identification.
Sky plane
15.4 km/s
Projection angle
32.1°
Deprojection
×1.88
Model estimate
29.0 km/s
Interpretation limit: The ridge is weak and supported above one robust sigma in only part of the sequence. Residual stellar structure and changing tail morphology can imitate a moving knot. The 29.0 km/s value is conditional on the object being C/2025 K1 (ATLAS), on the inferred 9.13-minute source cadence, and on motion parallel to the anti-solar tail. It is not a spectroscopically measured three-dimensional velocity.
Candidate overlay: contrast and star-masked residual
