Anaglyph 3D: Red and Cyan Glasses Effect
Each pixel is displaced by an amount that depends on how far away you tell the tool it is, which is what produces depth rather than a flat double image. Includes the Dubois matrices, derived to keep usable colour instead of throwing channels away.
Drop images here
or click to browse · or paste with Ctrl+V
JPG · PNG · WebP · GIF · BMP · AVIF
Depth model
Anaglyph type
Dubois uses matrices derived to minimise colour error. Greyscale has the least ghosting and full colour the most.
How far the two eye views are pushed apart. More is not better; past about 20 px most people lose the fusion and just see doubles.
Which depth appears to sit exactly at the screen. Everything nearer comes forward and everything further recedes.
A single photograph contains no depth information, so this cannot be real stereo. The depth model is a guess you choose, and it works when the guess matches your scene.
Presets
Your photos are never uploaded. The parallax and the export both run inside this browser tab.
People also use
Everything this tool does
Parallax that varies with depth
Each pixel is displaced according to the depth model you choose, so different parts of the picture separate by different amounts. A single uniform shift produces a double image rather than depth.
The real Dubois matrices
A pure red is 255, 0, 0 under a naive split, so the cyan eye sees nothing and the object vanishes for that eye. Under Dubois it comes out 109, 80, 0, and each matrix row sums to one so neutrals pass through with 0 levels of deviation.
Convergence control
Sets which depth appears to sit exactly at the screen, with everything nearer coming forward and everything further receding. Putting it on your subject is usually the most comfortable choice.
Four depth models
Ground plane, centre subject, brightness as a proxy, and a flat uniform shift. Each is a different assumption about how your scene is arranged.
Honest about the limits
A single photograph carries no depth information at all. This page says so rather than implying the tool is recovering something real.
Four anaglyph types
Greyscale is the most comfortable to view, full colour takes channels whole and loses strongly coloured objects from one eye, half colour sits between, and Dubois is the best all-round default.
Bilinear sampling
Parallax puts sample positions between pixels almost everywhere, so smooth sampling is what keeps edges clean rather than stepped.
Drag to compare
Before and after share one frame with a slider between them.
Batch with one ZIP
Apply the same depth model and separation across a set and take everything as a single archive.
Nothing is uploaded
No server, no queue, nothing to delete afterwards. Close the tab and every trace of the image goes with it.
Depth models and anaglyph types
Two separate choices. The depth model decides which parts of your photo come forward, and the anaglyph type decides how the two eye views are packed into one image.
| Setting | What it assumes | Try | Good for |
|---|---|---|---|
| Ground plane | the bottom is nearest | separation 10-16 | Landscapes, streets, rooms |
| Centre subject | the middle is nearest | separation 8-14 | Portraits, single objects |
| Bright is near | brighter means closer | separation 6-12 | Backlit and studio subjects |
| Flat | everything at one depth | separation 6-12 | The classic fake 3D look |
| Dubois | minimised colour error | any depth | Best all-round quality |
| Half colour | grey left eye, colour right | any | Less ghosting than full colour |
| Greyscale | no colour at all | any | Least ghosting, most comfortable |
4
depth models
4
anaglyph types
0
files uploaded
Dubois
matrices included
ZIP
batch download
Free
no account needed
FlipMyFormat vs other anaglyph tools
| Capability | FlipMyFormat | Typical free tool |
|---|---|---|
| Files stay in your browser | ||
| Depth-dependent parallax, not one flat shift | ||
| Dubois matrices for minimal colour error | ||
| Convergence control for the screen plane | ||
| Honest that a single photo has no real depth | ||
| Batch ZIP download without paying | ||
| No watermark on the output |
How to make an anaglyph that people can actually fuse
An anaglyph works by showing each eye a slightly different view of the same scene, filtered so that the red lens sees one and the cyan lens sees the other. Your visual system compares the two and reads the horizontal difference between them as depth. That last part is the whole game and it is why the most common mistake is pushing the separation up: a bigger difference is not more depth, it is a harder puzzle, and past roughly twenty pixels most people stop fusing the two images and simply see a doubled picture with a headache attached. Start low, around ten to fourteen, and only go further if the fusion still holds. The second decision is convergence, which is subtler and more useful than it sounds. Whatever depth you set as the convergence point will appear to sit exactly at the surface of the screen, with everything nearer coming out toward you and everything further receding behind it. Putting it in the middle is safe; putting it on your subject makes the subject sit at the screen and the background fall away, which is usually the more comfortable arrangement. The third choice is the anaglyph type, and it is a trade between colour and comfort. Full colour keeps channels whole, which sounds ideal and is not: a strongly red object ends up visible to one eye only, so it fights the fusion rather than helping it. Greyscale ghosts least and is the easiest to look at for any length of time. Dubois sits between them using matrices derived specifically to minimise the perceived error, and it is the right default for almost everything.
- 1
Add a photo
Drag it in, click to browse, or paste with Ctrl+V. Scenes with a clear near and far, such as a street or a path, work far better than flat subjects.
- 2
Choose a depth model that matches the scene
Ground plane for anything shot looking along the ground, centre subject for a portrait or single object.
- 3
Set separation and convergence, then export
Keep separation modest and put convergence on whatever should sit at the screen. Export PNG so the colour separation survives.
Where the depth comes from, and where it does not
A photograph has no depth in it
Real stereo photography uses two cameras a few centimetres apart, and the difference between their images is genuine measured parallax. A single photograph has none of that: the information about how far away things were was lost the moment the scene was projected onto a flat sensor. So this tool cannot recover depth, and any tool claiming to from one ordinary photo without a trained model is describing something impossible. What it does instead is let you choose a depth model, which is an assumption about how the scene is arranged, and then displace each pixel accordingly. A ground plane model assumes the bottom of the frame is nearest, which is true of most photographs taken standing up and looking outward. A radial model assumes the centre is nearest, which is true of most portraits. When the assumption matches your scene the result reads convincingly, and when it does not you get depth in the wrong places, which is worth knowing before you blame the tool.
Why the Dubois matrices exist
The naive way to build an anaglyph is to take the red channel from one eye and the green and blue from the other. It works, and it destroys colour: anything strongly red or strongly cyan in the scene ends up visible to only one eye, which both breaks the fusion and shifts the colour badly. Eric Dubois derived a pair of three by three matrices that mix all three channels of both views, chosen to minimise the perceived colour difference between the anaglyph and the original when viewed through real red and cyan filters. Those exact matrices are implemented here, and what they buy is easy to demonstrate. Take a pure red object. Under a naive channel split the output is 255, 0, 0, so the cyan eye receives nothing at all and that object simply vanishes for one eye, which is what breaks the fusion. Under Dubois the same red comes out as 109, 80, 0, so the cyan eye still receives a signal. The matrices are also constructed so that each row sums to one, which means neutral tones pass straight through: measured across a grey ramp from black to white, the worst deviation was 0 levels.
What people use it for
Retro 3D posters
Red and cyan anaglyphs are instantly recognisable as 1950s cinema, which is a strong design association.
Party and event graphics
Hand out cheap red cyan glasses and an anaglyph print becomes an activity rather than a picture.
Album and zine artwork
The colour fringing reads as deliberate and lo-fi even to people without glasses.
Teaching about stereo vision
A working anaglyph is the simplest way to demonstrate how two slightly different views become depth.
Museum and exhibition panels
Anaglyphs are cheap to print and need only paper glasses, unlike any other 3D display.
Social media posts
The red and cyan look is distinctive at thumbnail size, glasses or not.
Book and comic covers
Strong period association with pulp science fiction and horror.
Product and packaging design
Used as a graphic device rather than a viewing method, the fringing signals retro immediately.
Six tips worth knowing
Keep the separation modest. Past about twenty pixels most people stop fusing the images and just see doubles.
Put the convergence on your subject so it sits at the screen and the background falls away behind it. That is the most comfortable arrangement.
Match the depth model to the scene. Ground plane for anything shot looking outward, centre subject for portraits.
Dubois is the right default. Use greyscale when the image will be looked at for a long time, since it ghosts least.
Scenes with a clear near and far work best. A flat wall or a close-up has nothing for the depth model to separate.
Export PNG. Anaglyphs depend on precise colour separation and JPEG stores colour at half resolution.
Troubleshooting
I just see a double image
The separation is too high to fuse. Bring it down to around ten and build up slowly.
The depth is in the wrong places
The depth model does not match your scene. A ground plane model on a photograph shot straight at a wall will put depth where there is none.
The colours look wrong even without glasses
That is inherent to anaglyphs, since each eye's view occupies different channels. Dubois minimises it; greyscale avoids the issue by having no colour at all.
There is heavy ghosting
Strong reds or cyans in the scene leak into the wrong eye. Switch to half colour or greyscale, which is exactly the trade those modes exist to make.
Frequently asked questions
How does an anaglyph work?
Each eye is shown a slightly different view of the scene, filtered so the red lens sees one and the cyan lens the other. Your visual system reads the horizontal difference between them as depth.
Is this real 3D?
No, and it cannot be. Real stereo needs two cameras a few centimetres apart. A single photograph lost all depth information when the scene was projected onto a flat sensor, so this displaces pixels according to a depth model you choose.
What is a depth model?
An assumption about how your scene is arranged. Ground plane assumes the bottom of the frame is nearest, which is true of most photos taken standing up. Centre subject assumes the middle is nearest, which is true of most portraits.
Which depth model should I choose?
Whichever matches your photograph. Ground plane for streets, landscapes and rooms; centre subject for portraits and single objects; brightness for backlit or studio subjects where lighter genuinely does mean closer.
How much separation should I use?
Around ten to fourteen pixels. More is not more depth, it is a harder puzzle, and past roughly twenty most people stop fusing the two images and see doubles instead.
What does convergence do?
It sets which depth appears to sit exactly at the screen surface. Everything nearer comes toward you and everything further recedes. Putting it on your subject is usually the most comfortable arrangement.
What are the Dubois matrices?
A pair of three by three matrices derived by Eric Dubois to minimise the perceived colour error of an anaglyph viewed through real red and cyan filters. They mix all three channels of both views rather than taking channels whole.
How much better is Dubois than a channel split?
The clearest case is a strongly coloured object. A pure red comes out as 255, 0, 0 under a naive split, so the cyan eye receives nothing and that object disappears for one eye. Under Dubois the same red is 109, 80, 0, so both eyes still get a signal, which is what keeps the fusion working.
Which anaglyph type is most comfortable?
Greyscale, because it has no colour to leak into the wrong eye and therefore ghosts least. It is the right choice for anything that will be looked at for a while.
Why is there ghosting?
Strong reds and cyans in the scene are visible to only one eye through the filters, so the other eye sees a gap. Half colour and greyscale reduce it, which is the trade those modes exist to make.
What glasses do I need?
Standard red and cyan anaglyph glasses, red over the left eye. They cost very little and are the reason anaglyphs remain popular despite better 3D methods existing.
Which photos work best?
Scenes with a clear sense of near and far: a path receding, a street, a room with a foreground object. A flat wall or a tight close-up gives the depth model nothing to separate.
Why does my image look wrong without glasses?
That is normal and unavoidable. The two eye views occupy different colour channels, so the unaided image always shows coloured fringing. Many people use that look as a graphic device on its own.
Which format should I export?
PNG. Anaglyphs depend on precise separation between the red and cyan channels, and JPEG stores colour information at half resolution.
Are my photos uploaded to a server?
No. The parallax and the export both run in your browser using the canvas API. Nothing is sent anywhere.
Can I process a batch?
Yes. Add as many photos as you like and the same depth model, separation and convergence are applied to each.
Can I use a real stereo pair instead?
Not with this tool, which takes one image. A genuine stereo pair would give real parallax and a much better result, but it needs two photographs taken from slightly different positions.
Does it add a watermark or need an account?
No to both. No sign-up, no email wall, no daily limit and no watermark on the output.