A-CAM Lens Flicker Digital 2.39:1

Add a light to see how this lens flares

Drag a light · Shift-click to move the selected one there · A add · Del remove · H handles

T2.8 24.00 FPS 180.0° EI 800 ±0.0 EV No filter 4300K

Flare Bench

Physically based lens flares from the lens, the camera and the lights you tag.

Sequence

Shot

Flare amount

Flare lights 0 / 4

Tag only the lights that should flare: the sun, a practical, a stage spot. Fill and cheat lights stay out.

Show the light sourcesOff: the flare without the lamp, flame or sun itself

Haze glow

Shot

Camera motion

Before / afterDrag the line in the frame

Camera

Focal length

T-stopT2.8

Focus∞

Focus on the lightKeeps the selected light sharp; move Focus to rack off it

Exposure±0.0

Focal length sets the view, the sun's size and depth of field; the ghosts come from the lens design.

Look

Ray density

Elements

Arc size1.00×

Lens X-ray

The lens in cross-section with live rays from the selected light: the image rays in its colour, and in amber the ghost you picked (or the strongest one) bouncing between its two surfaces.

Lens tricks

What cinematographers do to a lens to make it flare more. Each is simulated from the physics and works with any lens.

Coating

Diffraction grating filter

Filter

Grooves

Rotation

Softness

Strength

Free-lensing

Lens tilt

Direction

Lens & camera details

Iris

Body

Frame rate

Shutter angle

Mains power

Front filter

Matte box flagsCut light from sources outside the frame

Front glassClean

Smears

Light leaks

Light handlesPress H to toggle

Share a look

Saves every setting, your lights and a snapshot of the frame to one file. Send it to someone to recreate exactly what you see, or load it later to pick up where you left off.

How it works

Flare lights: Only the lights an artist tags flare: the sun, a headlight, a stage spot. Each flares from its real position, size, colour, brightness and beam direction, and has its own flare gain, so a light can flare more or less without relighting the shot. Lights just past the frame edge still flare, pouring glare and barrel reflections in from that side, as on set. Add a light with + Add light under the frame (or press A), or from the Flare lights group at the top of the panel; drag its handle to place it, or Shift-click the frame to move the selected light there (a plain click never moves a light); remove it with its × or Delete. A shot takes up to four. S solos a light's flare and M mutes it; the light itself stays.
Flare model: Ray-traced follows thousands of rays per ghost through each lens's surfaces, glass, anti-reflection coatings and iris, across the spectrum (Hullin et al., SIGGRAPH 2011), plus diffraction for the starburst and a scatter model for glare. Hybrid traces the ghosts and streak and hand-shapes the rest. Artistic is fully hand-shaped and fastest. The lens designs are real optical types matched to each lens family, not the makers' private prescriptions.
Choosing a lens: The lenses are grouped by look, each tuned against real frames shot on the lenses named under it. Clean and large-format modern glass barely flares. Warm character lenses glow. Vintage glass throws bold coloured rings. Classic anamorphics give the thin blue streak and tall oval ghosts; modern anamorphics a paler streak and sideways pills. Cinema zooms sit close to modern primes. It opens on Large-format modern, the ARRI Signature Prime class of glass Roger Deakins shot 1917 on: clean, with almost no ghosts, which suits the restrained flares of modern features. The flare amount opens low for the same reason.
Light sources: The list is kept to the lights that turn up in an animated feature and really flare: the sun, headlights, streetlights, stage spotlights, bare bulbs and flashlights, plus the three classic ways a crew aims a light into the lens from just off frame. Each is calibrated by how far above the exposure it really sits and by its real spectrum: the sun or a headlight aimed at the lens is many stops over and flares hard, and sodium streetlights barely streak through blue-coated anamorphic glass. Large soft sources, fire, neon and screens are left out on purpose: they glow rather than flare, which a glow or convolution in comp handles better.
Ghosts and their cut edges: Ghosts are images of the lens opening, lined up through the frame centre opposite the light. Near the frame edge the glass rims clip them into crescents and cat's eyes, which change as the camera moves; stopping down shrinks the clipping.
Flare lights on set: Cinematographers often make the flare on purpose with a light just out of shot aimed at the lens: a xenon torch panned on and off the lens (the Star Trek method), a small spot trimmed in from the frame edge, a mirror bouncing the sun. The three off-frame lights in the picker land where a crew would put them, just past the frame edge, so the light itself stays out of shot while its flare pours in.
Starburst, streak, dirt: The starburst is diffraction off the iris blades. The anamorphic streak comes from reflections inside the cylindrical front group. Dust on the front glass shows as soft iris-shaped blobs; wipe marks and fingerprints throw streaky fans. The glass starts clean, the way a camera crew keeps it, and Lens & camera details can dirty it.
Lens tricks: Cinematographers who want more flare change the lens rather than paint it on, and Lens tricks simulates each one from the physics, on any lens. Coating swaps the lens's anti-reflection coatings for a single MgF₂ layer (as on 1950s–70s glass, and the custom single-coated sets some films order) or none at all (the stripped "uncoated" sets): every ghost gets much brighter, since it is two reflections and each one grows from about 0.3% to 1.3% or 5%, and single-coated ghosts turn the magenta and amber of a quarter-wave layer. A modern lens with 15 or more elements stripped bare washes out completely, which is why the real uncoated sets are older, simpler designs. A diffraction grating filter splits everything bright into spectra at exactly the angles its groove spacing sets: rows of rainbow (Lines, Cross) or fans of rainbow rays (Star 6 and 8). Free-lensing tilts the lens off the mount, so the flare lines up through a point off the frame centre and light leaks in through the gap.
Big soft rainbow flares: The huge rainbow arcs and washes of spectrum in many films come from things in front of the lens rather than the lens itself: gratings (simulated under Lens tricks), prisms, crystals and textured glass, or a flat filter whose reflection off the front element makes a giant ring. Rainbow arcs in the mixer is the art-directed version of those huge soft rings, for when a look calls for one, with its own Arc size.
Flicker: Sodium streetlights and bulbs on mains power pulse at twice the mains frequency, and LED headlights at their PWM rate; off-speed frame rates and shutters make them beat, and the rolling shutter turns that into bands.
Using it in comp: The flare is meant to go over a render, where the light itself already exists. Turn off Sources under the frame to see the flare alone: the ghosts, star, streak and the glow the lens adds around the light, without the sun's disc or the bulb. Handles hides the light markers for a clean view, and Present (P) hides the panel.
Layers: The menu under the frame (or keys 1–5) shows the layers the Nuke node would output, each over black: Beauty, Flare only, Ghosts, Star + streak and Glare (veil, dirt, barrel light and bloom). In the mixer, S solos an element and M mutes it, to find out what is making a particular shape.
Comparing looks: A and B under the frame hold two sets of look settings (lens, T-stop, coatings, mixer, tricks) over the same shot and lights. The slot that is lit follows your changes; click the other one, or press X, to switch. Ctrl/⌘+Z undoes the last 50 changes (Shift to redo), and double-clicking a slider puts it back to its default. Each panel section folds away with a click on its title; the flare amount stays pinned at the top.
Fixing one ghost: Physically driven by default, with a surgical override: Pick a ghost in the Look section, then click a ghost in the frame. The tool traces every ghost path to find the one under the cursor, names the two surfaces it bounces between and how much each reflects, and gives that path its own gain (down to off), tint and nudge, plus a solo to see it alone. It works on the traced ghosts (Ray-traced and Hybrid) and applies to that path's ghost from every light. Lens X-ray draws the lens in cross-section with live rays from the selected light: the image rays, and the picked (or strongest) ghost bouncing between its two surfaces, clipped where the real rims and iris clip it.

Seeing it in the shot

The tool builds a rough model of the shot from the render's own utility pass, in a second or two, and the flare lights live inside it.

Pick the shot: Choose a Sequence and Shot at the top of the panel and the tool finds that shot's utility pass on its own. In production that list would come from the pipeline; until it is connected, three demo shots stand in: a golden-hour meadow shooting into the sun, a concert stage with spotlights on the truss, and a night street. Each demo's utility pass is made the way a renderer makes one, by casting the shot camera's rays into a scene, and the tool reads it exactly as it would a real one. (For testing, utility EXRs can also be dropped on the frame.)
A rough model from the utility pass: The pass's world position, normals and depth, plus surface colour and emission when it has them, turn every pixel into a point of the scene as the shot camera saw it, so the model has the shot's exact framing and perspective. The shot's tagged lights come in with it.
The flare lights live in it: Drag a light onto the scene and it takes the distance of the surface under it (a streetlamp head, a car's headlight), so its flare, its size and its depth of field are right for where it really is. A light behind something the camera can see, such as a post, a tree or a building edge, is hidden, partly hidden lights dim, and the flare follows. The lights also light the model, with their real falloff, beam and colour and screen-space shadows, so a streetlight pools on the road and a headlight washes the street ahead of the car. View shows the surface colour, normals or depth to check the pass was read right.
The camera comes with it: The shot camera is read from the EXR metadata where the renderer wrote it: transform, field of view, focal length, f-stop and focus distance, checked against the pixels before it is trusted. Without it, the camera is solved from the position and depth passes themselves. Scene units (metres or centimetres) are guessed from how far away things are, and can be set by hand.
What it reads: Scanline EXRs, single- or multi-part, with ZIP, ZIPS, RLE or no compression, the lossless kinds used for data passes. Layers are found by their usual names: P, pos or position; N or normal; depth, Z or distance (an unfiltered depth is preferred when there is one); surfaceColor, albedo or diffuse colour; emission. Anything wider than 2048 pixels is read at half size or less.

From prototype to a Nuke tool

Flare Bench is the look and the behaviour. In production the same model would run as a Nuke node fed by the shot itself, so nobody places a flare by hand. The goal is one artist step: tag the light.

1. Tag the light in Houdini: Lights get a Flare switch (a parameter on the studio's light asset) and an optional flare gain in stops. The sun is tagged by default; fill and cheat lights never are. That is the only thing a lighter has to do.
2. The render carries everything: At render time the pipeline writes the shot camera (focal length, filmback, f-stop, focus distance, lens package; many renderers already put these in the EXR header) and, for every tagged light on every frame, its position in frame, distance, colour, intensity, size and how visible it is (a few rays toward each light, so the sun going behind a tree dims its flare) into the EXR's metadata. There are no locators to export or keep in sync: the data travels with the frames, so a re-render or a camera change updates the flare automatically. Big shaped lights can also get their own light-group pass, which the node reads as the light's real shape at whatever angle it is seen.
3. One node in comp: A single Flare node goes on the beauty. It reads the camera and the tagged lights from the metadata, sets the lens from the show's camera package, holds the tagged lights out of the screen flares so nothing flares twice, and outputs the flare as its own layers. The compositor only touches taste: the overall amount, a light's gain, the elements mix, and, when a director notes one ring on a face, that one ghost's gain, tint or position.
Why this way: It needs no new habits. Lighters already tag light groups and comp already reads EXR metadata, so the flare follows the real camera and lights with no hand-matching, and it stays right through every revision. A starting-point Houdini exporter is in the project's GitHub repository. It reads a Solaris stage (tags as flare:enable, flare:type, flare:gain attributes on any USD light; distances from the stage's own units) or old-style /obj lights, and writes each frame's camera and tagged lights into the rendered EXR's header as one flare/shot attribute. Drop such EXRs on the frame and the tool reads the lights straight from the header.

Under the hood

None of the ghosts, rings or streaks in the traced modes are drawn shapes. They are what comes out the back of a simulated lens when light is sent through it. This is how.

A lens, as an optical prescription: Each lens family is a real optical design written the way a lens designer writes one: a table of surfaces, each with its radius of curvature, the spacing to the next, the glass's refractive index and Abbe number (its dispersion), its clear diameter, and where the iris sits. The spherical primes are built around a double-Gauss core. The anamorphics add a cylindrical front group, solved numerically to be afocal in one axis so it squeezes only horizontally. Digital bodies add the IR-cut filter, cover glass and sensor behind the lens, because they reflect light too. The designs represent each lens type; they are not the makers' private prescriptions.
Paraxial optics: From that table the tool works out the focal length, back focus and entrance pupil: where the iris appears from the front, and how big. Anamorphics get this separately for each axis. Rays are aimed through that pupil, so stopping down really does shrink the bundle of light that gets in.
Where ghosts come from: A ghost is light that reflects twice inside the lens before it reaches the sensor: backwards off one surface, then forwards again off an earlier one. A lens with N surfaces has N(N−1)/2 such paths, so a 20-surface lens has 190. Each path is test-traced at several positions across the frame and ranked by how concentrated its light is, and the strongest 48 to 128 (set by Ray density) are kept. The ranking is done wide open and again three stops down, so a path that only shows once the iris closes (and stops clipping it) is not dropped. Paths with four or more reflections are dropped: each extra pair of reflections makes the light another 10,000 times or more dimmer. Lens designers also check where every ghost focuses and reshape the glass until none lands sharp on the sensor, where it would show as a crisp copy of the light mirrored through the centre. The lens models here are representative rather than optimised that way, so a path that would focus is given a designed lens's minimum blur, with its light kept, and an anamorphic line path that would draw a second streak across the frame is left out.
Real ray tracing, on the graphics card: For every ghost path, a grid of up to 128×128 rays is spread evenly across the pupil (using an equal-area square-to-disc mapping) and traced through every surface. Each surface gets an exact ray–sphere or ray–cylinder intersection and Snell's law refraction, with mirror reflection at the path's two ghost surfaces. Rays that miss a surface, strike outside an element's rim, or totally internally reflect are lost, as they would be in real glass. Each ray is one pixel of a shader, so at the higher settings millions of rays are traced for every frame.
Coatings and Fresnel reflection: At every surface, the fraction reflected is computed for each wavelength and angle. Bare glass and single- and two-layer coatings use the thin-film characteristic-matrix method, averaged over both polarisations. Modern broadband multicoats follow curves shaped like published reflectance data, including the faint residual colour (green, amber, violet) that tints each maker's ghosts. A ghost's brightness and colour are the product of its two reflecting surfaces and the transmission through every other surface on its path.
Spectral, not RGB: Six wavelengths from 420 to 660 nm are traced separately, with each glass's index changing with wavelength, so ghosts fringe and rainbow the way real dispersion makes them. Every source has an emission spectrum: a blackbody for the sun and tungsten, line spectra for sodium and discharge lamps, narrow bands for LEDs. Each traced wavelength is turned into colour with the CIE 1931 colour-matching functions, and sub-bands interpolated between neighbouring wavelengths keep rainbow edges continuous.
Glass dispersion: How much a glass's index changes across the spectrum comes from real catalogue glasses. Each glass in a lens design takes the Sellmeier dispersion curve (the standard fit to measured indices) of the nearest Schott glass by index and Abbe number, scaled to the design's own values. That keeps the real shape of the curve: crowns barely change across the spectrum, while flints and dense flints rise steeply in the blue, which is what pushes violet fringes outward on ghost rims. The curve is fitted as a polynomial for the graphics card.
Coatings you can change: The Coating choice re-coats every glass surface of the lens before tracing: a quarter-wave MgF₂ layer tuned to 550 nm, or bare glass. The thin-film calculation then does the rest. A quarter-wave layer reflects least in the green and most at the ends of the spectrum, so single-coated ghosts come out magenta and amber, and bare glass reflects about 4–6% evenly, so uncoated ghosts are bright and neutral. The camera's IR filter, cover glass and sensor keep their own coatings. In the Artistic model the same choice scales the ghosts and haze by the equivalent amount.
Diffraction grating: A grating filter is a plate ruled with fine grooves. Light of wavelength λ leaves in order m at the angle where sin θ = mλ/d, with d the groove spacing, so each order is a full spectrum, violet nearest the source and red furthest out, at a distance set by the lens's focal length; finer gratings throw it further. Orders 1 to 3 are drawn with typical efficiencies (about 10%, 2.5% and 0.8% of the light). Straight grooves throw the spectra along one axis; cross and star filters have two, three or four sets of grooves at angles, so every light grows four, six or eight rainbow rays. The filter sits in front of the lens and acts on everything coming in, not just the tagged lights: the lamp, its glow, lit haze, windows, highlights. Because the lens's glare and the filter are both convolutions, each diffracted copy carries the light's own glow, which is what makes the rays broad and soft on an overexposed source. The incoming image is gathered at reduced resolution over a field two and a half frame heights either side, so lights outside the frame throw their rays in too. Softness spreads each copy further, as a frosted, smeared or dirty filter and haze do.
Light grazing the barrel: A light at the frame edge or just outside it also reaches parts of the lens that aren't meant to be seen: the blackened baffles inside the barrel and the ground bevel round the front element. That path is non-sequential, so it is simulated separately. For each part of the ring the light hits, its reflection off the baffle's cone is computed with a Fresnel reflectance that rises sharply at grazing angles (why even matte black glints there), and its refraction in through the bevel and out the back of the front element uses that element's own glass dispersion. Each new direction is imaged where the lens puts it, so the light lands as a soft crescent on the side of the frame opposite the light, fringed with spectrum where it went through glass. A matte box cuts it.
Free-lensing: Holding a lens slightly off the mount tilts its optical axis, and every ghost lines up along the axis rather than the frame centre, so the whole flare swings off-centre. The tool moves the axis by the tilt angle through the focal length, traces the ghosts about the new axis, and lets light leak through the open gap between lens and body.
From rays to an image: Each ghost's traced grid is drawn as a mesh of triangles. Brightness at any point is the light carried by the pupil cells around a ray divided by the area they land on. Where the mesh is squeezed, the ghost gets brighter, so the bright caustic rims, line ghosts and pinched shapes come straight out of the optics, as in Hullin et al. (SIGGRAPH 2011). The iris is applied per pixel from each ray's distance to the blade polygon, so every ghost carries the real iris shape, blade count and rotation.
Clipped ghosts and off-frame light: Rays that pass outside an element's clear diameter are stopped. This is mechanical vignetting: near the frame edge it cuts ghosts into crescents and cat's eyes, and it lets a light outside the frame still throw ghosts into it. Both happen with real glass.
Sources with real size and distance: Lights are extended sources, not points. The sun covers its true half a degree, so the focal length sets exactly how big it is in frame. Every ghost is that source blurred by the source's size times the ghost's own magnification: how far the ghost moves when the light moves, measured for each ghost path from the lens itself. A ghost that nearly focuses barely moves and stays crisp; a big ghost moves more than the light and goes soft, which is why sun ghosts look the way they do. The magnification travels with each ghost's light, so the blur is set pixel by pixel. A light at a finite distance sends diverging rays rather than parallel ones, which shifts and refocuses its ghosts. Brightness is set in stops above exposure from real-world luminances: the sun is about 11 stops over a daylight exposure, a headlight aimed at the lens about 10, a bare bulb about 7.
Focus and depth of field: Focusing moves the lens away from the sensor as a real focus mechanism does, so ghosts drift and change size as you rack focus ("flare breathing"). Out-of-focus sources get the thin-lens blur circle, (f²/N)·|1/D − 1/F|/2, squeezed horizontally on anamorphics.
Manufacturing imperfections: Every element is given a small decentre, tilt and polishing error, so its ghosts never line up perfectly through the centre of frame.
Starburst from diffraction: The star is Fraunhofer diffraction from the lens opening as each light actually sees it. Far from a bright light, the diffraction of any opening is set by its outline: each straight edge throws a spike along its normal, with energy per unit length (edge length / opening area)·λN/(2π²x²), and each curved stretch lights the directions its normals face with a halo falling as 1/x³, in proportion to its radius of curvature. So an even number of blades gives that many spikes and an odd number twice as many, stopping down grows the star, and red spikes are slightly wider. For each light the iris polygon (or its rounded version) is cut by the lens rims into the cat's eye it really sees off axis: blade edges that get shortened throw weaker spikes, ones cut away throw none, and the rim arcs add a halo across the cat's eye. The star keeps the same orientation wherever the light is, because the iris doesn't turn. A full numerical FFT of the opening was considered: at true scale the star's fine structure (fringes along each spike, a few micrometres apart) is far below one pixel of a cinema frame, and the visible far field is exactly what the outline gives.
Glare and veiling haze: Micro-roughness on the glass and haze inside the lens scatter a small fraction of the light into a glare spread function, normalised so no energy is created. It is strong right beside the source and falls off slowly, about as 1/r^2.6, so a source as bright as the sun lifts the blacks and blooms a wide soft mass around itself, as real sun shots do. Each ghost also gets a soft halo scattered by its own coatings.
Independent of screen size: Every size is worked out in fractions of the frame from the optics (the sun's angle, a ghost's magnification, the spread of the glare), not in screen pixels, so a small viewport shows the same flare as a 4K frame, with less detail. The only pixel limits are where nothing can be smaller than a pixel: star spikes are at least one pixel wide (with their light kept the same, so they get dimmer, not brighter) and the softening of a ghost is at least a few pixels.
Aiming rays where they count: For a light far off axis much of the lens opening is cut off by the rims and the iris, and the opening itself shifts ("pupil walking"), so rays spread over the whole on-axis pupil would mostly be wasted exactly where ghosts are clipped into crescents. Every ghost path first runs forward through the front of the lens to its first reflecting surface, so for each light the tool traces that shared stretch, finds the part of the opening that survives it, and spreads the whole ray grid over just that window (each ray then carries less of the opening's area). Crescent edges come out cleaner at the same ray count.
Foreground in front of the lens: Rays from different points of the lens opening travel toward the light a few millimetres apart. Something close to the camera (leaves in the foreground, a character passing the lens) blocks some of them and not others, so its silhouette shows inside the ghosts. In a shot, each traced ray is marched back toward the light through the shot's depth, and dropped if it passes behind a surface within 5 m of the camera; anything farther away dims the whole flare through the usual visibility.
Motion blur: When the camera moves during the shutter, everything on the light (its star, streak and glow) moves with it, but a ghost sitting at k times the light's position along the line through the lens axis moves k times as far, and the other way for ghosts across the centre. The flare is blurred along that line accordingly, separately from the scene.
Converging like a path tracer: While the shot holds still, each new frame traces a fresh, rotated set of rays, source samples and wavelengths and averages it in. This is Monte Carlo accumulation, so a locked-off frame keeps refining toward a smooth result.
Linear light throughout: Everything is computed in scene-referred linear light, in 16-bit floating-point buffers. The flare is added to the image the way a lens adds it, before the camera's response curve. Flare gain scales what the lens does with a light, not the light itself, as in Animal Logic's flare pass.
What is tuned rather than derived: Some parts are modelled by hand: the Artistic flare model, the extra rainbow rings, the rainbow arcs, dust and smears on the front glass, light leaks, and the relative strength of glare and streak for each lens family. Those were matched by eye against real footage shot on the lenses each family is named after. In the Ray-traced model, the ghosts themselves come only from the optics.
Further reading: Hullin, Eisemann, Seidel and Lee, Physically-Based Real-Time Lens Flare Rendering, SIGGRAPH 2011. Ritschel, Ihrke, Frisvad, Coppens, Myszkowski and Seidel, Temporal Glare, Eurographics 2009. Animal Logic, Physical Based Lens Flare Rendering in The LEGO Movie 2, DigiPro 2019. Shirley and Chiu, A Low Distortion Map Between Disk and Square, 1997. Wyman, Sloan and Shirley, Simple Analytic Approximations to the CIE XYZ Color Matching Functions, 2013.