How This Image
Was Rendered

The Gaia catalog directly provides each star’s G magnitude and BP-RP color index. Magnitude is a logarithmic scale, not linear brightness. BP-RP is Gaia’s own photometric difference between its blue and red passbands, not ordinary photographic RGB. The screen ultimately needs three linear sRGB channels, so two translations are required: magnitude into relative luminous flux, and color index into display color.

Magnitude is converted to brightness with the Pogson formula: L = 10^(-0.4 × (m - m_ref)). A difference of 5 magnitudes means a factor of 100 in brightness. The smaller the magnitude value, the brighter the star. m_ref is only a brightness anchor. It decides where the whole image falls within the screen’s dynamic range without changing the relative brightness relationship between stars.

The color chain is longer. BP-RP is first converted to effective temperature using the Pecaut & Mamajek main-sequence calibration. The Sun’s G2V type corresponds to BP-RP of about 0.82 and a temperature of 5772K. The temperature then enters a blackbody radiation spectrum, which is integrated against the CIE 1931 color-matching functions to produce linear sRGB. Finally, the Sun is used as the white point: the 5772K solar color temperature is set to neutral white, and other stars become bluer or redder relative to that white point.

The easiest mistake here is treating Gaia BP-RP as Johnson B-V. They come from different filter systems, and their values are not directly interchangeable. If BP-RP is used as B-V, the many moderately red stars in the Galactic disk are pushed toward warm yellow, then amplified by display-layer saturation, making the entire Milky Way look yellowish. With the correct calibration, the disk returns to a nearly neutral gray-white, the Galactic center keeps a slight warmth, and the differences between blue-white and orange-red stars remain visible.

The stars in the Gaia catalog are point sources, but a point source on a screen cannot be drawn as a single pixel. In a real optical system, a point light source passes through a lens, sensor, and display, and becomes a point spread function (PSF) with width. That shape is a property of the imaging system, not the star. Faint and bright stars share the same core shape; they differ only in energy.

The renderer gives all stars a shared narrow Gaussian PSF with a core width of 0.6 pixels. This value is intentionally narrow, because the core PSF only removes the hard-pixel feel from point sources. It is not responsible for making large blobs around bright stars. If the core grows beyond 1 pixel, faint-star grains and Milky Way texture get smoothed away, and the image looks as if seen through a thin haze.

Bright-star size is created by saturated spillover. The portion of linear brightness above the saturation line is first clipped, then redistributed back into the image through two wider Gaussian wings, with widths of 3 and 9 pixels. This process conserves energy: the spillover wings only spread out energy from the over-bright core and do not create extra light. The saturation line is also attached to the magnitude ladder rather than to a fixed screen brightness. That way, when higher sensitivity is simulated, the whole brightness ladder shifts together, instead of incorrectly washing the entire Milky Way into diffuse glare.

The visual result is that faint stars remain sharp pinpoints, mid-brightness stars become small disks, and bright stars acquire a disk and soft halo. The apparent size of stars changes continuously with brightness instead of being artificially split into brush categories.

The Milky Way seen by the eye is not a pattern made from a few bright stars. It is integrated light from vast numbers of faint stars. A single faint star may be unresolved, but when millions or hundreds of millions of them stack along the Galactic plane, they form a continuous milky glow. The deeper the Gaia catalog, the closer this integrated light gets to the real sky. The shallower the catalog cutoff, the more the renderer has to proxy the faint stars that were not drawn.

At the shallow-catalog stage, the renderer compensates with a luminosity function. Faint stars in the catalog with G≥11 are multiplied by a 3.8× gain to proxy still fainter stellar populations that have not yet been included one by one. This number comes from Gaia’s own magnitude distribution: changes in the total luminous flux of the last few faint magnitude bins can be extrapolated to estimate how much integrated light remains missing after the cutoff.

But gain can only proxy total light. It cannot proxy grain statistics. A small number of relatively bright stars multiplied by a large gain and a large number of fainter stars accumulated one by one can have the same total flux, but they look different. The former tends to be coarse; the latter creates a finer milky glow and more stable dark-lane boundaries. The main image therefore uses a deep-catalog cache generated from the Flatiron bulk mirror: about 616 million G<20 stars in the Guangzhou Galactic-center field of view. At this depth, the shape of the Great Rift is no longer emphasized by a screen mask or dust texture. It becomes readable because the nebulous star fields on both sides brighten while star-poor regions become relatively darker. When real stars can be used, clever approximations are unnecessary.

The same Milky Way is clear under dark skies and disappears in the city. The reason is not that the catalog changed, but that an extra layer of skyglow sits in front of the observer. City lights scatter through the atmosphere and become a background added on top of linear starlight. The brighter the background, the lower the contrast between starlight and background. Faint stars and the diffuse Milky Way are drowned out first, leaving only the brightest stars.

The Bortle scale provides the observational anchor for this background layer. Bortle 1 corresponds to top-tier dark sky, with a naked-eye limiting magnitude of about 7.8. Bortle 6, a bright suburb, is around 5.3. Bortle 9, an urban center, is around 4.0. The renderer maps each level to a sky-background surface brightness and uses NELM to decide the magnitude boundary for visible point sources.

Point-like stars and the broad Milky Way band cannot be judged by the same rule, because the human eye detects point sources and diffuse light differently. Diffuse light uses a Weber contrast threshold: when a structure’s brightness increment relative to the background falls below a few percent, it becomes hard for the eye to detect. The default threshold is 0.035, applied to the low-frequency Milky Way light without erasing high-frequency point sources. The result is that under Bortle 1 the Milky Way is nearly complete, while around Bortle 7 it disappears from naked-eye view, matching real observing experience.

The first four mechanisms all compute linear light: stellar brightness, color, PSF, saturated spillover, skyglow, and human visual thresholds. But a screen is not a linear astronomical instrument. Gaia star brightness spans more than a million-fold range, while an ordinary SDR screen can display only limited contrast. The final tone-mapping step compresses the linear physical result into the screen without breaking the relative relationships established earlier.

This is also where the separation between physical layer and display layer becomes most concrete. The renderer first estimates the sky floor in each frame, using a low percentile of sky-background brightness, and anchors it to a fixed dark gray value. It then enhances only the starlight signal above the sky floor, avoiding pulling the background itself into a gray curtain. Multiple Bortle panels share the same stretch instead of being normalized independently. Otherwise, the city panels would be artificially brightened, misleading the reader into thinking a similar star field remains visible under light pollution.

Color is released after brightness compression. The chroma parameter is 1.8, bringing back the color-temperature differences already provided by BP-RP from the compressed state. Highlights use a soft shoulder, so the Galactic center and the areas around bright stars roll smoothly toward the display limit while retaining internal structure. Finally, the image is output with gamma 2.2 for ordinary screens. The display layer acknowledges that it serves the screen, but it does not rewrite the sky already computed by the physical layer.