GPU Screen Shows Artifacts or Flicker to Your Eyes but Screenshots Look Clean? Screenshot, Photo & Display-Chain Diagnosis Guide

2026-11-04 | GPU Repair | 38 minutes | HK GPU Fix Editorial

GPU Screen Shows Artifacts or Flicker to Your Eyes but Screenshots Look Clean? Screenshot, Photo & Display-Chain Diagnosis Guide

Key Takeaways

When your eyes see artifacts or flicker on the screen but a screenshot looks clean, the screenshot has not contradicted your eyes: a system screenshot reads the framebuffer, which is only the first half of the display chain, and it never passes through the output encoding, the cable, or the panel, so a clean screenshot proves only that rendering reached the framebuffer without visible error, and the fault, if there is one, sits in the second half of the chain or in a timing-dependent symptom that a still screenshot cannot capture, which is why this guide treats the screenshot as the first of three independent pieces of evidence — alongside a phone photograph of the panel and the monitor's own on-screen menu — and works through them in a fixed order, from characterising the symptom, through the screenshot test, the cross-device screenshot check, the photograph comparison, the menu test, and the one-variable-at-a-time swap comparisons, so that each result narrows the suspect segment instead of inviting the mistaken conclusion that a clean screenshot means nothing is wrong.

Two Halves of One Chain

A picture on a screen is the end of a chain, and the chain divides naturally into two halves. The first half is rendering: a program draws a picture, the graphics card writes the finished frame into a memory area called the framebuffer, and that framebuffer holds the picture as the system understands it. The second half is display: the graphics card encodes the framebuffer's contents into an output signal, the signal travels along a cable, the monitor scales and processes it, the panel and its backlight turn it into light, and the eye receives that light. Every visible symptom lives somewhere along this chain, and the diagnostic task is to find which segment it lives in.

The reason a screenshot is so often misunderstood is that it samples the chain at its midpoint. A system screenshot tool reads the framebuffer. It captures the picture as it exists after rendering and before display. It never sees the output encoding, the cable, the monitor's processing, the panel, or the backlight. A screenshot therefore answers a narrow question — "is the picture correct as far as the framebuffer?" — and it answers no other question. When the screenshot is clean, the first half of the chain is exonerated, and the second half remains entirely unexamined. When the screenshot shows the same defect the eye sees, the first half is implicated, and the second half is largely exonerated, because the defect exists before the second half ever touches the picture.

This division is the whole foundation of the guide. Everything that follows — the four-result table, the photograph comparison, the monitor menu test, the swap comparisons — is a way of refining that division, of splitting the chain into smaller segments until the suspect segment is small enough to act on. Readers who internalise the division will rarely be misled by a clean screenshot again, because they will hear "screenshot is clean" not as "nothing is wrong" but as "the first half is clean; now examine the second half."

The Four Results That Decide the Direction

The screenshot test produces four possible combinations with what the eye sees, and each combination points in a different direction. The table below is the core framework of this guide, and it should be consulted before any component is swapped or any driver is reinstalled, because it determines which of those actions could even be relevant.

Screenshot resultWhat the eye seesWhere the fault sitsWhat it does not mean
Screenshot shows the defectEye sees the defectBefore or at the framebuffer: program, driver, or graphics card rendering layerIt does not implicate the panel or the cable, which the screenshot never passed through
Screenshot is cleanEye sees the defectAfter the framebuffer: output encoding, cable, monitor processing, panel, backlight, or a timing-dependent symptomIt does not mean the graphics card is fault-free, because output encoding is part of the graphics card
Screenshot is clean, phone photograph shows the defectEye sees the defectConfirmed after the framebuffer, at the panel output sideIt does not mean the viewer imagined the symptom; the photograph is independent confirmation
Screenshot shows the defect, viewed on another machine still shows itEye sees the defectConfirmed before or at the framebuffer, travelling with the framebuffer contentsIt does not mean the panel is at fault, because the defect travels with the file, not the panel

The second row is the one this guide is named for, and it is also the row most often misread. A clean screenshot with a defective-looking screen is routinely reported as "the graphics card is fine," but that reading exceeds what the evidence supports. Output encoding — the conversion of the framebuffer into a signal the monitor can accept — is performed by the graphics card, and it sits after the framebuffer. A fault in that encoding would leave the screenshot clean and the screen defective, exactly as observed. The accurate statement is narrower and more useful: the rendering half is clean; the display half, including the graphics card's own output stage, remains suspect.

The third row introduces the phone photograph as a second, independent kind of evidence. A photograph captures what the panel actually emits as light, which is the opposite end of the chain from the screenshot. When the photograph shows the defect and the screenshot does not, the two pieces of evidence agree on the location: the defect arises after the framebuffer, somewhere between encoding and light. The photograph also serves a human purpose: it converts a symptom that only one person has seen into a record that others can examine, which matters when the symptom is intermittent and may not be present when help arrives.

The fourth row closes a small but real loophole. A screenshot viewed on the same machine passes through that machine's display chain a second time when it is shown, so a defect in the viewing path could theoretically contaminate the observation. Viewing the screenshot file on a different machine or phone removes that contamination: if the defect is still visible in the file on another display, the defect is in the file itself, which means it was in the framebuffer, and the panel is exonerated. If the defect vanishes on the other machine, the viewing path on the first machine becomes a candidate, and the investigation shifts to that machine's display segment.

The Monitor Menu Test: A Third Kind of Evidence

The monitor's own on-screen menu provides a third piece of evidence, and its logic differs from both the screenshot and the photograph. The menu is drawn by the monitor itself and overlaid on top of the incoming picture. It does not pass through the graphics card, the cable, or the input processing at all. It is, in effect, a picture whose source is the monitor alone.

The test is simple to perform and decisive in its reading. Open the monitor's menu while the defect is visible, and look at the menu's own text and icons. If the menu's text is itself distorted, flickering, or discoloured — if the menu, which the monitor drew, is affected — then the panel itself is the leading suspect, because a fault in the panel corrupts everything the panel shows, including its own menu. If the menu is crisp and clean while only the incoming picture behind it shows the defect, then the panel's drawing ability is intact, and suspicion returns to the input side: the signal arriving at the monitor, the cable carrying it, or the output encoding that produced it.

Used together, the screenshot test and the menu test narrow the suspect segment from both ends. The screenshot test splits the chain at the framebuffer; the menu test splits it at the panel. A defect that appears in the screenshot but not in the menu sits in the rendering half. A defect that appears in neither the screenshot nor the menu, but appears in the incoming picture, sits between the framebuffer and the panel — the output, cable, and input-processing segment. A defect that appears in the menu itself sits at the panel. The two tests together therefore divide the chain into three segments, which is enough precision to decide the next action without guessing.

Screen Recording Is the Same Evidence as a Screenshot, Not a Second One

A common belief is that a screen recording provides independent confirmation of a screenshot: if both the screenshot and the recording look clean, surely the case is doubly proven. That belief is mistaken, because screen recording reads the same source as a screenshot. Recording software captures the framebuffer, frame after frame, and writes those frames into a video file. It never passes through the output encoding, the cable, or the panel either. A clean recording and a clean screenshot are therefore the same conclusion stated twice, not two conclusions that corroborate each other.

This matters because the false sense of double confirmation is exactly what leads readers to close the investigation prematurely. "Screenshot and recording both clean" feels like a thorough check, and it is easy to report it as though two independent methods agreed. In fact, only one method was used — framebuffer capture — applied in two forms. The independent methods are the ones that sample different points in the chain: the phone photograph, which samples the panel's light output, and the monitor menu test, which samples the panel's own drawing. A genuinely thorough check combines one framebuffer method with one panel-side method, not two framebuffer methods.

The same logic applies to streaming or sharing a screen. A shared screen viewed by someone else is still a framebuffer capture, transmitted over a network. If the remote viewer sees a clean picture, that proves the framebuffer is clean, not that the local panel is healthy — and if the remote viewer sees the defect, that proves the defect is in the rendering half, because the defect travelled with the framebuffer contents across the network. Remote viewing is a useful diagnostic tool, but its direction must be stated correctly, or it will mislead in the same way as a local screenshot.

Timing-Dependent Symptoms: Why a Still Screenshot Cannot Capture Flicker

A screenshot is a single still frame. It freezes one instant of a picture that, in normal operation, is redrawn many times per second. Any symptom that lives in time rather than in a single frame — a flicker, a brightness oscillation, a black screen that lasts a blink and then vanishes — will usually be absent from a still screenshot, not because the symptom is unreal, but because the screenshot sampled an instant when the symptom was not visible, or because the symptom is a property of the sequence of frames rather than of any one frame.

Several well-known symptom families belong to this class. Brightness flicker under variable refresh rate operation, where the panel's brightness varies as the refresh timing varies, is a sequence property: no single frame is wrong, but the succession of frames shimmers. Backlight flicker, where the panel's illumination pulses, is likewise invisible in a still framebuffer capture, because the framebuffer contains no backlight information at all. An intermittent black screen that appears for a moment and then clears will only be caught by a screenshot taken during that moment, which is a matter of luck rather than method. Refresh-related flicker, where the symptom depends on the rate at which frames are presented, is similarly a property of presentation, not of frame contents.

For these symptoms, a clean screenshot is the expected result, not a contradiction, and the diagnostic method must change. The tools for timing-dependent symptoms are comparisons over time: switching the variable refresh feature off and observing whether the flicker stops, trying a different refresh rate and observing whether the symptom changes, recording the panel with a phone in slow-motion video so that a brief event becomes visible frame by frame, and swapping the panel to see whether the symptom travels with the panel or stays with the machine. Each of these is a comparison, not a single observation, because timing symptoms can only be characterised by how they change when something else changes.

The human cost of misunderstanding this point is real. A reader who is told, or who concludes, that a clean screenshot means their flicker is imaginary, or that their eyes are at fault, has been misled by a method limit being mistaken for evidence. The correct response to a clean screenshot alongside visible flicker is not dismissal but a change of method: put the still camera down and pick up the tools that can see in time.

Remote Desktop: The Same Logic, Applied at a Distance

Remote desktop viewing follows the same framebuffer logic, and it is worth stating explicitly because it is a practical tool that many readers already have. When you view a machine's screen remotely, the picture you receive is a transmission of that machine's framebuffer. It has never passed through the local machine's panel, cable, or output encoding. It proves things about the remote machine's rendering half, and nothing about its display half.

The two directions mirror the screenshot table. If the remote view is clean while the local screen shows the defect, the remote machine's framebuffer is clean, and the defect sits in the local machine's display half — its output, cable, or panel. If the remote view shows the defect, the defect is in the remote machine's rendering half, because it travelled with the framebuffer across the network. If the remote view is clean and the local screen is also clean, nothing has been proven about the display half beyond the absence of a currently visible symptom, which is a weaker statement than it sounds for intermittent symptoms.

Remote viewing is therefore a genuinely useful first split when a second machine is available, especially for a machine whose local display is hard to inspect. But its result must be reported with its direction intact: "remote view clean" means "framebuffer clean," not "display healthy." Losing that distinction converts a good tool into a misleading one.

Two Screenshot Exceptions That Must Not Be Misdiagnosed

Two situations produce screenshot results that look like faults but are not, and both must be recognised before any other conclusion is drawn, because acting on them as faults leads to wasted effort or, in one case, to improper conduct.

The first is protected content. Streaming platforms and similar services protect certain video content so that it cannot be captured: a screenshot or recording of protected video may come out as a black picture. That black picture is the protection mechanism working as designed, not a graphics fault, and the distinguishing feature is scope: the black capture appears only for protected content, while screenshots of ordinary content — the desktop, documents, unprotected video — come out normally. The diagnostic response is to note the scope and stop there. This guide does not describe, and the reader must not seek, any method of circumventing content protection; the legitimate alternatives are to view the protected content normally on an authorised display, and to use ordinary, unprotected content for any display testing. The boundary is stated plainly because the temptation to "fix" a black screenshot of protected content leads directly toward conduct that this guide does not condone and does not assist.

The second exception is high-dynamic-range content. A screenshot of high-dynamic-range material passes through tone mapping when it is captured and stored, and the colours in the resulting file may not match the colours the eye sees on a high-dynamic-range panel. Using screenshot colours to judge a high-dynamic-range colour complaint is therefore unreliable: the capture process itself changes the colours being judged. Colour complaints confined to high-dynamic-range content belong to the high-dynamic-range mapping discussion, not to the framebuffer-versus-panel framework, and the screenshot should not be used as a colour reference for them.

The Diagnostic Sequence, in Order

As with any chain-splitting diagnosis, the order of the steps matters, because each step eliminates a segment that later steps assume is already eliminated. Working out of order — swapping a cable before taking a screenshot, for example — leaves the result ambiguous: if the symptom changes, it is unclear whether the cable, the reconnection, or a coincident change in the picture was responsible.

Step 1: Characterise the symptom before capturing anything

Record what the symptom looks like and when it appears. Is it a fixed pattern — persistent lines, blocks of wrong colour, distorted shapes that stay in place — or is it intermittent — a flicker, a flash, a brief black screen that comes and goes? Does it cover the whole desktop, or only one program? Does it appear before the operating system loads, on the startup or firmware screens, or only after the system has started? Fixed patterns travel the screenshot route; intermittent symptoms need the timing route as well, and a symptom already present before the operating system loads sits outside the screenshot method entirely, because the operating system's screenshot tool is not yet available and the driver has not yet loaded. Characterising first prevents the most common waste in this topic: applying the screenshot test to a symptom it cannot capture, and then misreading the clean result.

Step 2: The screenshot test, taken at the moment of the symptom

When the symptom is visible, capture a screenshot immediately, and view it on the same machine. If the screenshot shows the defect, the investigation goes to the rendering half: the program, the driver, and the graphics card's rendering layer, with the program comparison and the clean driver installation path as the next branches. If the screenshot is clean, do not close the investigation. A clean screenshot at this stage is a result, not a verdict: it moves suspicion to the display half and to timing-dependent causes, and the investigation continues to step 3.

The timing of the capture matters for intermittent symptoms. A screenshot taken after a flicker has passed captures a normal frame, and its clean result says nothing about the flicker. For intermittent symptoms, the screenshot test should be attempted more than once, at different moments, and its result should be reported together with whether the symptom was actually visible at the moment of capture. A screenshot taken while the symptom was not visible is not a test of the symptom at all.

Step 3: Verify the screenshot on another machine

Transfer the screenshot file to another machine or a phone and view it there. If the file looks clean on the second display, the framebuffer is confirmed clean, and the possibility that the first machine's viewing path contaminated the observation is excluded. If the file shows the defect on the second display, the defect is confirmed to travel with the framebuffer contents, and the panel is exonerated. This step is quick, it costs nothing, and it closes a loophole that otherwise leaves a shadow over every later conclusion. It should not be skipped on the grounds that the first viewing "looked fine," because the first viewing passed through the very display chain under suspicion.

Step 4: The phone photograph comparison

Photograph the panel with a phone while the symptom is visible, so that the photograph and the screenshot depict the same moment as closely as possible. Then compare the two. A photograph showing the defect alongside a clean screenshot is strong evidence that the defect arises after the framebuffer, at the panel output side. A photograph showing no defect, where the eye sees one, suggests either that the symptom is too brief for the photograph to catch — in which case the timing route applies — or that the symptom's visibility depends on viewing conditions that the photograph does not reproduce. Either way, the comparison is informative, and the photograph becomes a durable record that can be shown to someone else, which matters for intermittent symptoms that may not recur on demand.

A note on photograph technique: the photograph should include enough of the screen to show the symptom in context, and it should be taken straight on where possible, because angled photographs introduce reflections and perspective distortion that can mimic or mask a defect. If the symptom is a colour problem, be aware that a phone camera's own colour processing will shift colours somewhat; the photograph is best used to confirm the presence and shape of a defect, not as a precise colour reference.

Step 5: The monitor menu test

Open the monitor's own menu while the symptom is visible and examine the menu's text and icons. If the menu itself is distorted, flickering, or discoloured, the panel is the leading suspect. If the menu is crisp while the incoming picture behind it shows the defect, suspicion returns to the input side: output encoding, cable, or input processing. Record the menu result separately from the screenshot and photograph results, because the three together form the evidence set, and a later reader should be able to see which of the three pointed where. A diagnosis that reports only "screenshot clean" is incomplete; a diagnosis that reports all three is actionable.

Step 6: Swap comparisons, one variable at a time

With the evidence set in hand, perform swap comparisons, changing one thing at a time. Swap the panel: if the symptom follows the panel to another machine, the panel is implicated; if the symptom stays with the original machine on a different panel, the machine's output layer is implicated. Swap the cable, and then the connector, each as a separate trial, observing after each whether the symptom changed. Changing the panel, cable, and connector together and observing an improvement proves nothing about which of them mattered, and it leaves the reader unable to reproduce the result or to know which component to replace. The discipline of one variable at a time is the difference between a diagnosis and a coincidence.

Step 7: The timing route, for flicker and brief symptoms

If the symptom is flicker, brightness oscillation, or a brief black screen, run the timing comparisons alongside, or instead of, the still-image steps. Switch the variable refresh feature off and observe whether the flicker stops. Try a different refresh rate and observe whether the symptom changes character. Record the panel with a phone in slow-motion video, which stretches a brief event into visible frames. Swap the panel as in step 6, because a timing symptom that follows the panel implicates the panel, while one that stays with the machine implicates the machine's output timing. A clean screenshot is the expected result throughout this route, and it must not be reported as contradicting the symptom. The timing route's results are changes — "flicker stopped when this feature was switched off" — not single observations, and they should be recorded as changes.

Step 8: The scope split

Finally, split by scope. If only one program shows the symptom, the program layer is the suspect: its settings, its shaders, or its files, and the comparison to make is with other programs on the same machine. If the whole desktop shows it, the system, driver, or hardware layer is the suspect. If the symptom is already present on the startup or firmware screens, before the operating system loads, the driver is exonerated — it has not loaded yet — and the hardware layer is the suspect. Scope is a powerful splitter because it is cheap to test and because it immediately halves the search space, but it must be tested deliberately: "only one program" must mean that other programs were actually tried and found clean, not that only one program was ever opened.

What Each Combination of Evidence Means

The evidence set — screenshot, cross-device screenshot, photograph, and menu test — combines into a small number of meaningful patterns. The table below summarises them, so that a reader who has gathered the evidence can read off the conclusion without re-deriving the chain logic each time.

ScreenshotPhotographMonitor menuMost likely segmentNext action
Shows defectShows defectCleanRendering half, before the framebufferProgram comparison, driver path
Shows defectShows defectShows defect tooRendering half confirmed; panel also suspect only if menu distortion is independentRendering half first; panel separately
CleanShows defectCleanAfter the framebuffer: output, cable, input processingSwap comparisons, one variable at a time
CleanShows defectShows defect tooPanel itselfPanel comparison and panel-focused examination
CleanCannot capture itCleanTiming-dependent symptomTiming route: feature off, rate change, slow-motion recording
Clean for protected content only, black captureNot applicableCleanContent protection working as designedNo fault; use ordinary content for display testing
Colours differ from eye, high-dynamic-range onlyColours differCleanTone mapping in capture, not a panel faultHigh-dynamic-range mapping discussion

Two cautions apply to this table. First, the second row — defect in both screenshot and menu — is ambiguous and should not be over-read: a rendering defect shown behind a menu can make the menu look affected when it is merely overlaid on a defective picture, so the menu must be examined on its own, for example against a clean background area, before concluding that the panel is also at fault. Second, the table's "most likely segment" is a direction for the next test, not a final verdict; each row's next action is itself a test, and the investigation continues until a swap comparison or a scope split confirms the segment.

Boundaries: When the Problem Is Not This Guide's Problem

Several lookalike problems share surface features with this guide's topic and must be separated early. If the symptom appears only at high resolution or high refresh settings and vanishes at lower settings, the bandwidth, load, and signal-integrity discussion applies, and this guide's screenshot framework serves only as a one-line split. If the startup or firmware screens already show artifacts, the driver has not loaded, the screenshot method cannot be applied at that stage, and the hardware-focused path applies. If the screen is divided by a single straight vertical boundary, with one side normal and the other defective, a dedicated combination of screenshot, menu, and torch tests applies, and this guide's general framework should not be copied over it. If a screenshot is black only for protected streaming content, the protection mechanism applies, as described above. If screenshot colours differ from the eye's colours only under high-dynamic-range content, the high-dynamic-range mapping discussion applies. If the eye sees trailing or ghost images that a screenshot cannot capture, the panel response discussion applies, because trailing is a property of how the panel transitions between frames, which a still framebuffer capture cannot show. Naming the boundary early prevents the framework from being stretched onto problems it was not designed to solve, where its conclusions would be wrong even if its steps were followed faithfully.

Common Misconceptions, Stated Plainly

"A clean screenshot means the graphics card is fine." This is wrong. A clean screenshot proves the rendering half, up to the framebuffer, is clean. The graphics card's output encoding sits after the framebuffer and remains unexamined by the screenshot. The accurate statement is that rendering is clean, not that the card is fault-free.

"A screenshot showing artifacts means the monitor is broken." This is wrong in the opposite direction. A screenshot never passes through the monitor, so a defect visible in a screenshot cannot have been caused by the monitor. A screenshot defect exonerates the panel and implicates the rendering half.

"Screenshot and recording both clean means double confirmation." This is wrong, for the reason given above: both read the framebuffer, so they are one kind of evidence in two forms. Genuine confirmation requires evidence from a different point in the chain — a photograph of the panel's light, or the monitor's own menu.

"A photograph showing artifacts means the graphics card is broken." This is wrong. A photograph captures the panel's output, and a defect visible in a photograph could sit in the panel, the cable, the input processing, or the output encoding. The photograph proves the defect is real and panel-side in the broad sense; it does not single out the graphics card.

"A screenshot cannot capture flicker, so there is no problem." This is wrong, and it is the most harmful of the misconceptions, because it dismisses a real symptom on the basis of a method limit. A still screenshot cannot capture timing-dependent symptoms by design. The correct response is to change method, not to dismiss the symptom.

"Remote viewing clean means the local monitor is healthy." This is wrong. A remote view never passes through the local monitor, so it proves nothing about it. It proves the remote machine's framebuffer is clean, which is a different statement entirely.

Recording the Evidence Set

Whatever the outcome, record the evidence set before closing the case, in a form that another person can follow. Note the symptom's character — fixed or intermittent, whole-desktop or single-program, present before the operating system loads or only after — and whether it was visible at the moment of each capture. Note the screenshot result, the cross-device screenshot result, the photograph result, and the menu result separately, with the menu's own text condition described on its own, not merely as "menu fine." Note each swap comparison as a change — what was swapped, what stayed the same, and whether the symptom followed the swapped component or stayed with the machine. Note each timing comparison as a change as well — which feature was switched off, which rate was tried, and whether the symptom changed. A record in this form turns a one-off observation into a repeatable diagnosis, and it is what allows help to start from the comparisons already made rather than from the beginning.

When to Seek Help

If the evidence set points to the rendering half and the program comparison and driver path have been worked through without resolution, or if it points to the display half and the swap comparisons have not isolated a component, or if the symptom is intermittent and the timing route has not characterised it, hands-on examination may be needed. HK GPU Fix is located at Room 1209, Dali Plaza, Shatin. The shop can be reached on 2186 8010, and on WhatsApp at 6558 6806. Published diagnostic fees on the website run from HK$400 to HK$2,800 across three tiers; any individual assessment depends on what the examination finds, and no figure beyond that published range is quoted here. Bringing the evidence set — the screenshot file, the photograph, and the menu, screenshot, and swap results recorded separately — will make that examination faster, because it starts from evidence already gathered rather than from a bare report that the screen "looks wrong."

Frequently Asked Questions

My eyes see artifacts but my screenshot is clean. Is my monitor broken?

Not necessarily, and the screenshot alone cannot answer that. A clean screenshot proves only that the picture was correct as far as the framebuffer. The fault could sit in the output encoding, the cable, the monitor's processing, the panel, or the backlight, or it could be a timing-dependent symptom that a still screenshot cannot capture. The next steps are the photograph comparison and the monitor menu test, which sample the panel side of the chain, followed by swap comparisons one variable at a time.

If my screenshot shows artifacts, does that rule out the monitor?

Yes, in the important sense: a screenshot never passes through the monitor, so a defect visible in a screenshot must have existed before the display half. The monitor is exonerated as the origin of that defect, and the investigation goes to the rendering half — the program, the driver, and the graphics card's rendering layer. The monitor could still have an unrelated problem, but it cannot be the cause of a defect that is already present in the framebuffer.

Why do both my screenshot and my screen recording look clean?

Because both read the same source: the framebuffer. A screen recording is a sequence of framebuffer captures, not a capture of the panel's light. A clean recording and a clean screenshot are the same conclusion in two forms, not two independent confirmations. For independent confirmation, use a method that samples a different point in the chain: a phone photograph of the panel, or the monitor's own menu.

What does it mean if my phone photograph shows the defect but my screenshot does not?

It confirms that the defect is real and that it arises after the framebuffer, on the panel output side. The photograph captures the panel's emitted light, while the screenshot captures the framebuffer, so their disagreement locates the defect between those two points: output encoding, cable, monitor processing, panel, or backlight. The photograph also turns a symptom only you have seen into a record others can examine, which is valuable for intermittent symptoms.

How does the monitor menu test work?

The monitor's menu is drawn by the monitor itself and overlaid on the incoming picture, without passing through the graphics card or the cable. If the menu's own text and icons are distorted or flickering, the panel itself is the leading suspect, because a panel fault corrupts everything it shows, including its own menu. If the menu is crisp while only the incoming picture is defective, the panel's drawing ability is intact, and suspicion returns to the input side: output encoding, cable, or input processing.

My screen flickers but every screenshot looks normal. Am I imagining it?

No. A still screenshot cannot capture timing-dependent symptoms, because it freezes a single instant of a picture that is normally redrawn continuously. Brightness flicker under variable refresh operation, backlight flicker, and brief black screens are all sequence properties that a still framebuffer capture will usually miss. The correct response is to change method: switch the variable refresh feature off and observe, try a different refresh rate, record the panel in slow-motion video, and swap the panel — and to record each result as a change.

If someone viewing my screen remotely sees a clean picture, does that mean my monitor is fine?

No. A remote view is a transmission of your machine's framebuffer, and it never passes through your local monitor. A clean remote view proves your framebuffer is clean, not that your local display is healthy. Conversely, if the remote viewer sees the defect, the defect is in the rendering half, because it travelled with the framebuffer across the network. Remote viewing is a useful split, but its direction must be reported correctly.

Why is my screenshot black, but only for streaming video?

That is most likely content protection working as designed: certain protected video cannot be captured, and a screenshot or recording of it comes out black. The distinguishing feature is scope — the black capture appears only for protected content, while screenshots of ordinary content come out normally. It is not a graphics fault, and this guide does not describe any method of circumventing content protection. Use ordinary, unprotected content for display testing, and view protected content normally on an authorised display.

Why do screenshot colours differ from what I see on screen for high-dynamic-range content?

Because a screenshot of high-dynamic-range material passes through tone mapping when captured and stored, which changes its colours relative to what the eye sees on a high-dynamic-range panel. The screenshot is therefore not a reliable colour reference for high-dynamic-range complaints. Colour problems confined to high-dynamic-range content belong to the high-dynamic-range mapping discussion, not to the framebuffer-versus-panel framework.

Should I reinstall my driver if my screenshot is clean?

Not as a first step. A clean screenshot points away from the rendering half, where the driver lives, and toward the display half and timing-dependent causes. Reinstalling a driver to chase a clean-screenshot symptom changes a component that the evidence has already largely exonerated, while leaving the suspect segment untouched. The driver path belongs to the branch where the screenshot shows the defect, or where the scope split points to the system layer — not to the clean-screenshot branch.

What if the symptom is already visible on the startup screens, before the operating system loads?

Then the driver is exonerated, because it has not loaded yet, and the screenshot method cannot be applied at that stage, because the operating system's screenshot tool is not available. The hardware-focused path applies: the symptom exists independently of the operating system and its driver, so the investigation goes to the hardware layer. Reporting this scope fact early saves a great deal of wasted driver work.

What should I record before asking for help?

Record the symptom's character — fixed or intermittent, whole-desktop or single-program, present before the operating system loads or only after — and whether it was visible at each capture. Record the screenshot result, the cross-device screenshot result, the photograph result, and the menu result separately. Record each swap as a change, stating what was swapped and whether the symptom followed the component or stayed with the machine. Record each timing comparison as a change as well. That evidence set lets help start from comparisons already made, rather than from the beginning.

Walking One Frame Through the Chain

It helps to follow a single frame from its creation to the eye, and to note at each stage what a screenshot, a photograph, and the menu would each show if that stage were the faulty one. A program draws a scene and the graphics card writes the finished frame into the framebuffer. At this point, a screenshot would capture exactly what the program drew, including any error the program or the driver introduced. If the program drew a corrupted shape, or the driver misrendered a surface, the screenshot shows it, and the chain downstream is irrelevant to that defect, because the defect is already baked into the frame.

Next, the graphics card's output stage encodes the framebuffer into a signal. Encoding involves timing, colour format conversion, and protocol handling, and an error here produces a signal that differs from the framebuffer. A screenshot taken at this point would still be clean, because the screenshot reads the framebuffer before encoding. Only a device downstream of the encoder — the monitor, or a capture device placed after the output — could see the error. This is the stage that the phrase "clean screenshot, faulty screen" most often conceals, and it is why the graphics card cannot be exonerated by a screenshot: part of the card sits on each side of the screenshot's sampling point.

The signal then travels along the cable. A cable fault typically announces itself in the time domain — dropouts, brief black screens, sparkling points that come and go — rather than as a stable, clean geometric defect, because a cable either delivers its signal or it does not, and partial delivery shows up as instability. A stable, unchanging artifact that looks identical in every frame is therefore a poor fit for a cable fault, while an intermittent symptom that changes with cable movement or connector reseating is a good fit. That distinction, between stable and unstable symptoms, is a useful secondary clue, though it never replaces the screenshot and photograph tests.

At the monitor, the incoming signal is processed — scaled to the panel's native grid, colour-adjusted, and timed to the panel's refresh — and then the panel and backlight convert it into light. A fault in the monitor's processing can distort an otherwise clean signal, and a fault in the panel can distort everything, including the monitor's own menu. The photograph captures this final stage as light, which is why it is the natural counterpart to the screenshot: the screenshot samples the chain's midpoint as data, and the photograph samples its endpoint as light. Between them, they bracket the whole second half.

Finally, the eye receives the light. Human vision itself introduces one more variable that no capture method shares: sensitivity to flicker varies between people and with viewing conditions, and a flicker that one person notices readily may be less obvious to another, or less obvious in a photograph. That variability is not a reason to doubt a reported flicker, but it is a reason to prefer the objective comparisons — feature off versus on, rate changed versus unchanged, slow-motion recording — over arguments about who can see what. The chain ends at the eye, but diagnosis should end at the comparisons.

When the Photograph Disagrees With the Eye

Occasionally the photograph and the eye disagree, and the disagreement itself needs interpretation rather than being treated as a contradiction. The most common case is a symptom too brief for the photograph: a flash lasting a fraction of a second may be gone before the phone's shutter opens, or may fall between the phone's own frames. In that case the photograph's clean result does not refute the eye; it only shows that the photograph, like the screenshot, sampled the wrong instant. Slow-motion video, which records at a higher frame rate than ordinary photography, is the appropriate next tool, because it increases the chance of catching a brief event and stretches it into visible frames.

A second case is a symptom that depends on viewing angle or ambient light. Some panel defects are most visible head-on and fade at an angle, or most visible in a dark room and washed out in bright light. A photograph taken at an angle or in different lighting may therefore miss a defect that the eye sees clearly in its usual conditions. The remedy is to photograph under the same conditions in which the symptom is seen — same seating position, same room lighting — rather than to conclude from a mismatched photograph that the symptom is unreal.

A third case is colour. A phone camera applies its own colour processing, white balance, and exposure decisions, so the colours in a photograph will differ somewhat from the colours the eye perceives, and from the colours another phone would record. A photograph is therefore reliable for the presence, shape, and location of a defect, and less reliable for its exact colour. A colour complaint should be described in words alongside the photograph — which areas shift, in which direction, and under which content — rather than being left to the photograph alone to convey.

In all three cases, the principle is the same: a capture method that disagrees with the eye should prompt a question about the method's limits, not an immediate dismissal of the eye. Methods have sampling instants, viewing angles, and processing of their own. Diagnosis advances by matching the method to the symptom, not by declaring one witness wrong.

Separating Panel Faults From Input Faults in More Detail

The monitor menu test draws the line between panel faults and input faults, but the line deserves a closer look, because the two sides call for different next actions and different expectations. A panel fault corrupts the panel's own output: its menu, its test patterns if it offers them, and every input shown on it, all pass through the same faulty panel. An input fault corrupts only the arriving signal: the menu, drawn locally, stays clean, and a different input on the same panel may look fine. That asymmetry is what makes the menu test decisive, and it is worth testing deliberately rather than glancing at the menu in passing.

To test it deliberately, open the menu while the symptom is at its worst, and examine the menu's text closely, including small text and fine icons, because a subtle panel fault may distort fine detail before it distorts large shapes. Move the menu, if the monitor allows it, to different parts of the screen, and see whether the distortion follows the menu or stays with a screen region — a fault that follows the menu points to the menu's rendering, while a fault fixed to a screen region points to that region of the panel. Switch to a different input, if one is available, and see whether the symptom appears there too: a symptom present on every input points to the panel, while a symptom confined to one input points to that input's signal path.

These refinements do not replace the swap comparisons, but they sharpen them. A reader who arrives at the swap stage already knowing whether the menu is affected, whether the symptom follows the menu, and whether it appears on other inputs, can interpret the swap result immediately: a symptom that followed the panel in the swap, and that affected the menu, is a panel conclusion with two independent lines of evidence behind it, not one. That redundancy is the point of the whole evidence-set approach. No single test in this guide is meant to carry the diagnosis alone; each test exists to corroborate or to correct the others, and a conclusion supported by two tests pointing the same way is worth far more than a conclusion shouted by one.

Related articles

Related services

If you need Display output repair, see these service pages: