Applications
One measure, many pipelines.
φ is a general structural prior. The same measure powers super-resolution, enhancement, astrophoto gradient removal and video today — and reaches into new domains next.
Live today
Super-Resolution ×2
● LiveUpscale images and video, structure-first, no invented detail.
φ Enhancement
● LiveContrast and sharpness placed exactly where φ sees structure.
Denoising
● LiveRemoves sensor noise and grain while φ preserves real detail — a small model that keeps structure sharp.
Gradient removal
● LiveAstrophotography: φ lifts the light-pollution and vignetting gradient while preserving stars and nebulosity.
Video ×2
● LiveFrame-by-frame upscaling, encoded to MP4 in your browser.
Measured, on every kind of image
Each pair is a real run of the exact model you can try in the studio — 100% zoom, bicubic left, phiXplorer right, PSNR measured against the original. Pick your field.
Nature & wildlife+1.5 dBUpscale ×2
Nature & wildlife+2.1 dBUpscale ×2
Nature & wildlife+2.3 dBUpscale ×2
Architecture & detail+1.6 dBUpscale ×2
Architecture & detail+0.9 dBUpscale ×2
Nature & wildlife+0.4 dBUpscale ×2
Astrophotography+0.3 dBUpscale ×2
Astrophotography+0.5 dBUpscale ×2
Documents & text+0.9 dBUpscale ×2
Old photossharpness ×2.9φ Enhance
Nature & wildlifesharpness ×2.6φ Enhance
Astrophotographysharpness ×3.4φ Enhance
Architecture & detail+10.7 dBDenoise
Nature & wildlife+8.8 dBDenoise
Nature & wildlife+7.7 dBDenoise×8 pairs: a real low-resolution crop enlarged ×8 — no downscale, no reference, nothing hallucinated. ×2 pairs: original → ÷2 reduction (area) → ×2 upscale, PSNR measured against the original. Enhance pairs: same resolution, sharpness = Laplacian-variance ratio. Denoise pairs: Gaussian noise (σ≈22) added to a clean original, left = noisy, right = model output, PSNR measured against the clean original. Real-noise pairs: two frames of the same scene from the same camera — one at base ISO (the reference), one at ISO 6400 (left). No noise was added: it comes from the sensor. The frames are aligned by integer pixel shift and exposure-matched per channel (no resampling, so the noise is untouched); PSNR is measured on the exact crop shown, against the base-ISO frame. Crops are picked for legibility and structure, never for the best score. Source: NIND (Wikimedia Commons, CC BY 4.0), scenes held out of training.
Moon: NASA/LRO · Nebula: NASA/ESA/CSA (Webb) · The Tetons (1942): Ansel Adams, US National Archives — public domain. Denoise samples (CC0, Wikimedia Commons): Blue Mountain Village — Transportfan70 · Autumn forest path — Vovogov90 · Mara Park woodland — Extemporalist.
Research & roadmap
Super-Resolution ×4 & ×8
● On requestHigher zoom factors, run offline on your own hardware — 4.66 MB model, φ inside. See the ×8 previews in the gallery.
Denoising — real sensor noise
● On requestThe in-browser model learns synthetic noise. This one is trained on real sensor noise — signal-dependent, spatially correlated, chroma-dominant. Same 0.71 MB model, φ inside: on real photographs it gains +3.4 dB over the in-browser one, up to +7.7 dB on the noisiest shots. See the pairs in the gallery. Runs offline on your own hardware.
φ Detection — RGB & IR
● On requestStructure detection validated in our publications, across visible and infrared. Not yet in-browser — available on request.
ROI extraction
● On requestSemantic-free regions of interest, driven by φ alone — published and functional. Available on request.
Compression
Researchφ-guided bit allocation — spend bits where structure lives.
Scientific Imaging
RoadmapStructure-aware processing for microscopy, astronomy, remote sensing.
Quality Assessment
RoadmapA structural score to rank and audit image pipelines.
Some capabilities — φ detection (RGB & IR) and ROI extraction — are already validated in our publications but not yet packaged for the browser. They are available on request.