Gigapixel GmbH is a specialized image portal for ultra-high-resolution photos from 100 megapixels for large-format applications – without AI upscaling. This page explains why the difference between genuine gigapixel photography and computationally upscaled material is decisive.
What is AI Upscaling?
AI upscaling uses neural networks (e.g., Topaz Gigapixel AI, Adobe Super Resolution) to computationally enlarge a small source image. The algorithm was trained on thousands of image pairs and can statistically interpolate missing pixels – it invents details based on patterns, but does not measure them.
Typical architecture: Convolutional Neural Networks (CNNs) operate on tiles of 256×512 pixels. For large outputs, the entire image is split into such tiles, processed, and reassembled. Visible artifacts appear at tile boundaries.
4×
AI collapses from this upscaling factor
(Kim et al., 2024)
256 px
AI training tile size – artifacts on large surfaces
(Topaz Community, 2020)
94 ppd
Human eye resolution capacity
(Ashraf et al., 2024)
r = 0.091
Aesthetic judgment correlation high/low res
(Mullin et al., 2017)
What is Genuine Gigapixel Photography?
Gigapixel photography is created through physical capture: a camera takes hundreds to thousands of individual images with a long focal length, which are assembled into a complete image using stitching software. Every pixel is based on an actual light measurement by the camera sensor – no information is invented.
Technique: Typically 70–400 mm focal length, 30–40% overlap, software such as PTGui or Autopano, output in PSB format (up to 300,000 × 300,000 pixels). Processing time: hours to days (Kopf et al., 2007).
Direct Comparison: Gigapixel vs. AI Upscaling
| Criterion | Genuine Gigapixel Photography | AI Upscaling |
|---|---|---|
| Information source | Camera sensor – physically measured | Statistically interpolated – invented |
| Quality when zooming | Always new genuine details visible | Blur and artifacts from 4× |
| Forensically valid | Yes – document authentic | No – generated details |
| Visual acuity requirement | Meets 94 ppd of human eye | Difference visible at less than 50 cm |
| Large-format print | Up to 200+ m2 without quality loss | Artifacts visible when approaching |
| Healing Environment | Full restorative effect (Salingaros 2015) | Low-quality simulation sabotages goals |
| Cost efficiency | One license – unlimited reuse | Software costs + quality risk |
Conclusion: AI upscaling can visually enlarge a small image – but it cannot create image information that was not there. Genuine gigapixel photography delivers physically measured original data that is irreplaceable for large-format printing, healing environments, forensic documentation, and interactive viewers. Gigapixel GmbH exclusively sells original images from 100 megapixels – without interpolation, without AI upscaling.
Frequently Asked Questions
Can I use AI-upscaled images for large-format printing?
For prints up to approximately 1 m2 from 2 meters viewing distance, AI upscaling may be sufficient. For areas over 5 m2, viewing distances under 1 meter, or forensic requirements, original gigapixel images are mandatory. The human eye detects AI artifacts at 50 cm below 350 ppi (Ashraf et al., 2024).
Is Topaz Gigapixel AI the same as genuine gigapixel photography?
No. Topaz Gigapixel AI computationally enlarges an existing image. Genuine gigapixel photography is created through physical camera captures with stitching software. AI creates no new image information – it interpolates statistically. A 1-gigapixel AI-upscaled image contains no more true image information than the 10-megapixel source.
How can I recognize the difference between gigapixel and AI upscaling?
The most reliable test is zooming in: genuine gigapixel images always reveal new physically correct details at deep zoom. AI-upscaled images show a muddy effect or artificial patterns. In the Gigapixel Portal every image can be inspected at full original resolution before purchase – without registration.
Why is AI upscaling unsuitable for Healing Environments?
Healing environments require the full fractal complexity of natural images with D = 1.3–1.5 (Taylor, 2021). AI algorithms cannot reconstruct this subtle complexity. Low-quality simulations actively sabotage restorative goals (Ashraf et al., 2024). Medical: AI can invent structures that do not exist and cause misdiagnoses.
From which upscaling factor do AI artifacts become visible?
From the 4x magnification factor, current algorithms collapse measurably (Kim et al., 2024). In MOS tests, AI models scored significantly worse at high zoom levels. The human eye detects the difference below 94 ppd in the fovea (Ashraf et al., 2024).
Where is AI upscaling scientifically classified as dangerous?
In medical pathology, AI can invent structures leading to misdiagnoses. In forensics, interpolation corrupts evidentiary foundations. In art restoration, AI details could overwrite historical information (Operation Night Watch, Rijksmuseum, 717 gigapixels). In industry, defects can be masked by smoothing algorithms.
Scientific Sources
- Ashraf, M. et al. (2024): Resolution Limit of the Eye. University of Cambridge / Meta. 94 ppd average, 120 ppd individual.
- Kim, B.S. et al. (2024): Chain-of-Zoom: Extreme Super-Resolution. KAIST AI. AI collapses from 4x.
- Kopf, J. et al. (2007): Capturing and Viewing Gigapixel Images. Microsoft Research.
- Mullin, C. et al. (2017): The gist of beauty. r = 0.091 correlation high/low resolution.
- Taylor, R.P. (2021): Stress reduction via fractal art. D = 1.3–1.5.
- Salingaros, N.A. (2015): Biophilia and Healing Environments.
- Stories in the Rock (2013): 4 minutes average viewing time gigapixel viewer.