Video Editing Timeline Preview Green Screen, Artifacts or Glitches While the Export Is Fine? Preview Rendering, Decode & Effects Diagnosis Guide

2026-11-16 | GPU Repair | 28 minutes | HK GPU Fix Editorial Team

Video Editing Timeline Preview Green Screen, Artifacts or Glitches While the Export Is Fine? Preview Rendering, Decode & Effects Diagnosis Guide

Key Summary

A timeline preview that turns green or broken while the export plays perfectly has not necessarily exposed a faulty card or a damaged clip. Preview and export are two different jobs, and a clean export proves the decode, render, and encode work already succeeds. This guide shows which preview layer is at fault, and how to prove it.

The Shape of the Problem: Preview Broken, Export Perfect

This guide covers one specific shape, and the shape itself is the whole diagnosis. The same timeline, in the same editing software: the program monitor shows a green screen, green blocks, a scrambled picture, or frames that break apart — while the exported file, played in an ordinary player, is completely normal. The preview is broken and the export is fine, on the same clips, with the same effects, on the same machine.

That combination is what makes this symptom different from its neighbours. As #474 explains, a preview that stutters while the export plays smoothly is a real-time problem: the preview cannot keep up with the clock, but the picture itself is correct. Here the picture itself is wrong — green, blocky, or broken — which is a different fault in the same preview path. The mirror image belongs to #476: there the preview plays smoothly and the export fails, which proves the fault sits on the export path instead. And a green screen in a browser or a media player belongs to #173, which covers the playback decode path rather than the editing timeline. None of those is this symptom. Here the preview picture is corrupt and the export is clean, on the same timeline.

The first conclusion most editors reach is that the graphics card is failing, and the first remedy most reach for is a new card. Both of those first reactions overlook the single result that settles the question: the exported file came out clean. A clean export proves that decoding works, rendering works, and encoding works. The fault can only sit in the part of the chain that the export does not share with the preview — the real-time preview path itself. What remains to be found out is not whether the hardware is capable of the job at all. The question is which part of the preview path produces a corrupt picture, and the answer is found by testing the layers one at a time.

Preview and Export Are Two Different Jobs

As #474 explains, preview and export are two different jobs with two different time limits. Timeline preview has to decode each frame, apply every effect on that frame, render it through the graphics card, and show it on the monitor, all inside the time allowed by the sequence frame rate — at a typical sequence frame rate, each frame gets only a fraction of a second, and there is no second chance. Export works to no such clock. Each frame can be given whatever time it needs to pass through the same decoding, effects, and rendering, without hurrying.

That difference is exactly why a preview can be corrupt while an export is clean. The preview path runs the same work under a deadline, through a live renderer, with effects applied in sequence and frames handed between the processor and the graphics card in real time. The export path runs the same work without the deadline, and often through a different render configuration. A fault that only appears under the deadline, or only in the live renderer, will show in the preview and never in the export. The export is therefore not just a consolation — it is the decisive test. When the same stretch of timeline is corrupt in preview but clean in the export, the fault is locked to the preview path, and every step that follows works inside that boundary.

There is a second consequence worth stating plainly: a clean export does not mean nothing is wrong. It means the export path is sound. The preview path can still be faulty, and the two paths have to be judged separately. An editor who concludes from a clean export that the machine is fine will keep fighting a corrupt preview indefinitely, because the evidence that would point at the preview path has been read as evidence that nothing is wrong.

What a Green Screen Actually Is

A green screen is alarming, but alarming does not mean a hardware fault. As #173 explains, a green screen means the decoded picture has gone wrong at the colour handover step, so the whole frame leans green or shows green blocks. The picture has been decoded, but somewhere in the handover between decoding and display — the colour handling, the tone mapping, or the renderer setting — the colour information has been misread, and green is what comes out the other side.

In the editing preview path, the same mechanism can appear at the renderer step rather than the player step. The usual suspects are a mismatch between the hardware decoding path and the driver or software colour handling, and incorrect tone mapping when high dynamic range content is converted for a standard dynamic range display. Both are configuration and path problems, not signs that the card itself has failed. That distinction matters because it changes the remedy completely: a colour handover fault is separated by changing the path — the renderer, the decoding mode, the effect order — and watching whether the green follows the setting. A fault that follows a setting predictably is a path problem, not a dying card.

The same logic applies to the other preview corruptions in this guide. Blocky colour patches, a scrambled picture, and frames that break apart are all pictures that have gone wrong somewhere between decoding and display. What they share is that the source material is intact — the export proves that — and the corruption is introduced on the way to the monitor. The layers below are the places on that road where it can happen.

What the Old Community Reports Do and Do Not Prove

Two old Adobe Community threads report exactly this shape, and they need to be handled honestly, because they are old and because they disagree with this guide on one point.

The first thread, from around 2018, reports that every imported clip looked broken in both preview windows, while the finished render played normally — and that switching the renderer from Mercury Playback Engine GPU Acceleration to Software Only made the preview normal at once. The second thread, also from around 2018, reports video turning green and purple in the source and program monitors when GPU acceleration was switched on, and a reply pointing at a driver update. Both threads are around eight years old, and the driver versions and software versions they name are long out of date — no version number from either thread should be treated as current guidance. What still applies is the mechanism: a renderer switch changing the preview picture proves the fault sits in the preview render path, not in the clip and not in the card as a whole.

The point of disagreement is what the switch means. The old thread presents switching to Software Only as the repair. This guide treats it as a diagnostic test only, and that is a deliberate choice, following the way newer guides describe the same switch. Switching to Software Only removes GPU acceleration from the preview entirely, so a preview that turns normal proves the fault was in the GPU preview render path — but running on Software Only permanently gives up GPU acceleration, and the underlying fault, whether in the driver, the effect order, or the codec, is still there, only bypassed. A test that proves where a fault sits is not the same thing as a repair, and this guide keeps the two apart throughout.

The Five Suspicion Layers

A corrupt preview with a clean export can fail at five layers, and they must be kept separate. Changing everything at once and finding the preview normal proves nothing about which change did the work, and the problem will come back the next time only one of those conditions is present. The layers below run from the cheapest to check to the hardest to fix, and each one has its own deciding test.

LayerWhat happensDeciding test
1. RendererMercury Playback Engine set to GPU acceleration or Software Only; setting reset by a software updateRenderer switch comparison: corrupt on GPU acceleration, normal on Software Only means this layer
2. Effect orderA GPU-accelerated effect runs after a processor effect, and the handover between the two goes wrongSwitch effects off one at a time: whichever one restores the picture is the trigger
3. DecodeInter-frame codec load, hardware decoding support, or HDR to SDR tone mappingClip swap comparison: simple intra-frame clip normal, original clip corrupt means this layer
4. Memory and loadHigh-resolution timeline, many effect layers, and video memory pressurePreview resolution isolation: dropping from Full to 1/2 to 1/4 restores the picture means this layer
5. Driver and media cacheDriver version that does not sit well with the editing software, or a damaged media cacheTimeline comparison, with media cache clearing only where damage is suspected

Layer One: The Renderer

As #407 explains, GPU acceleration in Premiere Pro is controlled by the Mercury Playback Engine: under File, then Project Settings, then General, then Video Rendering and Playback, the Renderer has to be set to Mercury Playback Engine GPU Acceleration for the graphics card to be in use. If only Mercury Playback Engine Software Only is available, Premiere is not recognising a usable card — but that is a different symptom from this guide, where the card is recognised, the renderer is set to GPU acceleration, the preview runs, and only the picture is corrupt.

The deciding test is the renderer switch: change the Renderer from GPU Acceleration to Software Only and play the same stretch of timeline again. If the preview turns normal at once, the fault sits in the GPU preview render path. If the preview stays corrupt on Software Only as well, the renderer is not the layer, and the investigation moves on. Two cautions apply. First, the Renderer setting is stored per project, so one project can use GPU acceleration while another uses Software Only — the fact that another project previews normally proves nothing about this project's setting, and each project has to be checked on its own. Second, as #407 explains, settings can be reset to their defaults when the software is updated, so if the preview turned corrupt immediately after an update, check the Renderer setting first before reinstalling anything. The switch is a test, not a repair: after the test, record the result, and do not leave the project on Software Only as if the matter were settled.

Premiere can also show you what it has detected for itself. As #407 explains, Help, then System Compatibility Report, lists the graphics card and driver that Premiere has detected, along with any warnings. That report shows what Premiere itself sees, rather than what the editor assumes it sees, and it is the starting point for any investigation at this layer.

Layer Two: Effect Order

Editing software applies effects in sequence, and that sequence can be the fault. A GPU-accelerated effect running after a processor effect can force the software to move that stretch of frames from the processor back to the graphics card, and that handover can go wrong — producing a green screen, colour blocks, or a broken picture exactly where the two effects meet. The deciding test is to switch the effects on the timeline off one at a time, playing the same stretch after each change. Whichever effect, switched off, restores the picture is the trigger condition. If every effect is switched off and the preview is still corrupt, the fault is not in the effect layer.

The discipline here is the same as everywhere else in this guide: one change at a time. Switching all effects off at once and finding the picture normal proves only that something in the group was responsible. Switching them back on one at a time is what identifies which one, and the order matters — an effect that is harmless on its own can be the trigger when it runs after a different effect, so the test has to reproduce the original order rather than testing each effect in isolation.

Layer Three: Decode

As #474 explains, modern camera formats such as H.264 and H.265 use inter-frame compression: each frame is stored with reference to the frames around it, so playing one frame means decoding several frames at once. That arrangement saves disk space but costs a lot of effort at playback time. With a high-resolution clip in one of these formats, the decoder has a much heavier job on every frame than it does with a clip whose frames are each stored on their own. ProRes and DNxHR take the other approach: bigger files, yet every frame stands alone for decoding, so the work per frame drops sharply.

The deciding test is a clip swap: on the same timeline, replace the clip with a simple one — low resolution, intra-frame codec — and play it again. If the simple clip previews normally and the original clip is corrupt, the fault sits in the decode layer. If the simple clip is corrupt too, the fault is not in the decode layer. Two further checks belong here. First, whether hardware decoding is switched on for the codec in question, and whether the hardware supports that codec at all — a codec the hardware cannot decode falls back to the processor, and that fallback can produce the corruption. Second, whether the clip is high dynamic range content being converted for a standard dynamic range display, since incorrect tone mapping at that conversion is one of the known routes to a green screen, as #173 explains.

Layer Four: Memory and Load

A high-resolution timeline with many effect layers can press on video memory, and that pressure can corrupt the preview picture. As #381 explains, monitoring numbers mislead in both directions here: an engine can pre-allocate a large pool of video memory as a cache, so a high reading can mean caching rather than actual use, and a comfortable-looking reading can still hide a failed allocation, because the number reports total free space while the allocation needed one continuous block. A monitoring number is a reference, not a verdict, and it must be read together with the comparison tests rather than trusted on its own.

The deciding test is preview resolution isolation: in the program monitor, drop the playback resolution from Full to 1/2, then to 1/4, playing the same stretch after each change. If the picture turns normal as the resolution drops, the fault sits in the load layer — the preview path could not carry the full-resolution load in real time. Proxy media is the same test in a different form: cutting with a low-resolution, easy-to-play proxy linked to the original by timecode, and exporting from the original, leaves the finished file unaffected while removing the decode load from the preview. If proxies or a lower preview resolution restore the picture, the load layer is confirmed; if the preview stays corrupt at 1/4 with proxies on, the fault is not load.

Layer Five: Driver and Media Cache

This layer is suspected last, and only on evidence. A driver version that does not sit well with the editing software version can corrupt the preview render path, and the timeline is the first clue: if the corrupt preview began immediately after a driver update or a software update, that timing points here. The comparison is against the current official release guidance for the driver version to use — never against a version number from an old community thread.

Media cache belongs here too, and only where damage is suspected. As #474 explains, a damaged or unfinished media cache slows playback, and clearing the cache is a test in this layer — with a warning that belongs to it: after the cache is cleared, the first playback has to rebuild the rendered frames, so it will stutter for a while, and that stutter is normal. That stutter on its own does not show the fault has survived. Clear the cache only through the software's own media cache controls, never by deleting cache folders by hand, since deleting the wrong folder can damage a project.

The Seven Steps

Work through these steps in order, changing one thing at a time and writing down the result after each step. A test that changes two things at once proves nothing about which change mattered, and the fault will return the next time only one of those conditions is present.

Step 1: Characterise the Symptom

Before changing anything, write down the facts. How is the preview corrupt — the whole picture green, local green blocks, a scrambled picture, broken frames, or flicker? Which clip, which position on the timeline, and which effects are switched on when it happens? Is the corruption fixed to one position every time, or does it wander to a different place on each play? And is the exported file genuinely normal — if the timeline has never been exported, it cannot be called normal yet. Note the editing software and its version, the renderer setting, and whether a driver or software update happened recently. That note is the baseline every later step is compared against.

The fixed-versus-wandering distinction deserves attention at this step, because it separates two layers before any setting is changed. Corruption that appears at the same picture position every time points at the clip or the effect at that position. Corruption that appears at a different place on each play points at memory or driver load, which has no fixed address. This fixed-versus-wandering logic follows the same idea as #476, which uses whether an export failure always stops at the same timecode to separate a clip fault from a system fault.

Step 2: Confirm the Fork With an Export

Export the same stretch of timeline and play the finished file in an ordinary player. If the export is clean, the fault is locked to the preview path, and the investigation continues in this guide. If the export shows the same corruption, this is not the right guide: the fault sits earlier in the chain, and the export-failure guide #476 is the right path. This step must be done rather than assumed, because the entire diagnosis rests on the export being genuinely normal — a preview fault diagnosed on an untested export is a guess, not a finding.

Step 3: Swap in a Simple Clip

On the same timeline, replace the clip with a simple one: low resolution, intra-frame codec such as ProRes or DNxHR. If the simple clip previews normally and the original clip is corrupt, the fault sits in the decode or effect layer. If the simple clip is corrupt too, the fault is not in the decode layer, and the investigation moves to the renderer. This comparison follows #474, which uses the same swap to separate decode load from the rest of the preview chain.

Step 4: Switch the Renderer

Change the Renderer from GPU Acceleration to Software Only and play the same stretch again. If the preview turns normal at once, the fault sits in the GPU preview render path. If the preview stays corrupt on Software Only as well, the renderer is not the layer. Remember that this is a comparison test, not a repair: switching to Software Only gives up GPU acceleration, so record the result and do not leave the project on Software Only as if the matter were settled. Check the System Compatibility Report under Help as well, since it shows what the software itself has detected. And check the setting per project, because one project can use GPU acceleration while another uses Software Only.

Step 5: Switch Effects Off One at a Time

Switch the effects on the timeline off one at a time, playing the same stretch after each change. Whichever effect, switched off, restores the picture is the trigger condition. If every effect is switched off and the preview is still corrupt, the fault is not in the effect layer. Keep the original effect order when switching them back on, because an effect that is harmless on its own can be the trigger when it runs after a different effect.

Step 6: Drop Preview Resolution and Try Proxies

In the program monitor, drop the playback resolution from Full to 1/2, then to 1/4, playing the same stretch after each change. Then switch on proxy media and play it again. If the picture turns normal at a lower resolution or with proxies on, the fault sits in the load layer. If the preview stays corrupt at 1/4 with proxies on, the fault is not load. Lowering preview resolution changes only what the monitor shows while editing: the source files and the export setup are left untouched, and you can raise the setting back whenever you want.

Step 7: Compare Drivers and Media Cache, Then Consider Inspection

Confirm the driver version against the current official release guidance, and clear the media cache through the software's own controls only where cache damage is suspected. After clearing, the first playback has to rebuild the rendered frames and will stutter for a while — that is normal, not evidence that the fault remains. If all seven steps have been worked through and the preview is still corrupt, bring the machine in with the complete record: which clip and codec, which effects, the preview resolution, the renderer setting, the fixed or wandering position of the corruption, and the result of every step. That record is what makes an in-shop inspection efficient, because it shows exactly which layers were tested and which evidence each one produced.

Six Common Misconceptions

Misconception 1: "A green preview means the graphics card is broken"

Wrong. A clean export already proves that decoding, rendering, and encoding work. A corrupt preview means only that the real-time preview path has a fault — the renderer, the effect order, the decode load, the memory load, or the driver. A card fault would not spare the export so neatly.

Misconception 2: "Switching to Software Only fixed it"

Wrong. Switching to Software Only is a diagnostic test, not a repair. It proves the fault sits in the GPU preview render path, but it gives up GPU acceleration, and the underlying fault is still there, only bypassed. The repair has to follow the test result — driver, effect order, codec, or memory — rather than stopping at the switch. The old community thread that presents the switch as a repair is describing a test result, not a fix.

Misconception 3: "A clean export means nothing is wrong"

Wrong. A clean export proves only that the export path is sound. Preview and export are two different jobs: preview has to deliver every frame inside the sequence frame rate, while export can take its own time with each frame. As #476 explains in the mirror direction, a smooth preview does not guarantee a working export — and the reverse holds too. The two paths have to be judged separately.

Misconception 4: "Clearing the media cache will always fix it"

Wrong. Clearing the cache only helps where the cache is damaged. After clearing, the first playback has to rebuild the rendered frames and will stutter for a while, which is normal and must not be mistaken for the fault remaining. Clearing as routine maintenance destroys a working cache and creates the very rebuilding work it was meant to remove.

Misconception 5: "A stronger graphics card will fix a green preview"

Wrong. A green preview can be a renderer setting, an effect order, a codec, or a driver version — none of which is a question of how much computing power the card has. Buying a stronger card before running the seven steps is guessing at the most expensive possible cause for a fault that may sit in none of those places.

Misconception 6: "A green screen means the clip is damaged"

Wrong. The same clip playing normally in an ordinary player, and an export of the same timeline playing normally, proves the source material is intact. A corrupt preview points at the preview render path, not at the clip. Until the export comparison and a player comparison have been done, the clip must not be called damaged.

Boundaries: Problems That Look Similar but Are Not This One

Several problems look like a corrupt editing preview but belong to other guides, and naming them here saves the reader from following the wrong diagnosis.

A preview that stutters — playing slowly, dropping frames, moving in bursts — while the export plays smoothly belongs to #474, which is a real-time performance problem with a correct picture, not a corrupt picture. A preview that plays smoothly while the export fails, stalls, or crashes belongs to #476, the precise mirror image of this guide. A green screen or stutter in a browser or a media player belongs to #173, which covers the playback decode path rather than the editing timeline. Editing software that cannot find the graphics card at all — where the renderer offers only Software Only — belongs to #407, which starts from a card that is not recognised, while this guide starts from a card that is recognised and working. An encoder that overloads during a live stream or a recording, with dropped frames in the recording itself, is an encoding problem rather than a preview problem. And a timeline that is corrupt in the export as well as the preview is not this guide at all: the fault sits earlier in the chain than either the preview path or the export path, and the renderer and decoding layers have to be settled first.

When to Stop and Bring the Machine In

Stop when all seven steps have been worked through, one variable at a time, and the layers still cannot be separated. Stop sooner if the corruption appears in the export as well, if the machine shuts itself down, or if any step involves opening the machine or changing firmware — those belong to professional inspection.

Bring the machine in with the notes from the seven steps: which clip and codec, which effects were on, the preview resolution, the renderer setting, whether the corruption stays fixed at one position or wanders, and the result of every comparison. That record is what makes an in-shop inspection efficient, because it shows exactly which layers were tested and which evidence each one produced.

In-store inspection takes place at Room 1209, Dali Plaza, Shek Mun. You can call 2186 8010, or reach the shop on WhatsApp at 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800 (Tier 1 HK$400–780, Tier 2 HK$480–1,380, Tier 3 HK$680–2,800), with the actual price subject to a post-inspection quote. There is no home-visit service and no Mong Kok shop; inspection is carried out in the shop.

A diagnosis that ends at the stop line is not a failed diagnosis. A note that says "Same timeline: preview green at the same position every time, export clean, simple ProRes clip previews normally, preview turns normal on Software Only, System Compatibility Report shows no warnings" is a precise description of where the fault lives, and it is exactly what an in-shop inspection needs to start from the right place instead of from the beginning.

FAQ

Q1: My timeline preview is green but the export is fine — is my graphics card broken?

Not necessarily. A clean export proves that decoding, rendering, and encoding all work, so the fault sits only in the real-time preview path — the renderer, the effect order, the decode load, the memory load, or the driver. A card fault would not spare the export so neatly. Run the renderer switch test first: if the preview turns normal on Software Only, the fault is in the GPU preview render path.

Q2: Switching to Software Only made my preview normal — is that the fix?

No. Switching to Software Only is a diagnostic test, not a repair. It proves the fault sits in the GPU preview render path, but it gives up GPU acceleration, and the underlying fault is still there, only bypassed. The repair has to follow the test result — driver, effect order, codec, or memory. An old community thread from around 2018 presents the switch as a repair; this guide treats it as a test, following the way newer guides describe the same switch.

Q3: Another clip previews normally — does that rule out the renderer?

No. The Renderer setting is stored per project, so one project can use GPU acceleration while another uses Software Only, and each project has to be checked on its own. The clips can also differ: H.264 and H.265 use inter-frame compression, so playing one frame means decoding several frames at once, while ProRes and DNxHR store each frame independently and are far lighter to decode. Settings can also be reset to their defaults when the software is updated, so check the Renderer setting first if the corruption began after an update. Another clip previewing normally rules out neither the renderer nor the decode layer — check the setting per project and run the clip swap comparison.

Q4: What does a green screen actually mean?

As #173 explains, a green screen means the decoded picture has gone wrong at the colour handover step, so the whole frame leans green or shows green blocks. The picture has been decoded, but the colour information has been misread somewhere between decoding and display — in the colour handling, the tone mapping, or the renderer setting. It is a path problem, not a sign that the card itself has failed.

Q5: How do I tell whether the fault is an effect or the decode layer?

Run two comparisons. Switch the effects off one at a time, playing the same stretch after each change — whichever effect, switched off, restores the picture is the trigger condition. Then swap in a simple clip, low resolution and intra-frame codec such as ProRes or DNxHR: if the simple clip previews normally and the original is corrupt, the fault sits in the decode or effect layer; if the simple clip is corrupt too, it is not the decode layer. Keep the original effect order when switching effects back on, because an effect that is harmless on its own can be the trigger when it runs after a different effect.

Q6: The preview turns normal when I drop the playback resolution from Full to 1/4 — what does that mean?

It points at the memory and load layer. Lowering the playback resolution changes only what the monitor shows while editing — the original clips and the export settings stay exactly as they are, and the resolution can be raised again whenever you like. If the picture turns normal at a lower resolution, the preview path could not carry the full-resolution load in real time.

Q7: After clearing the media cache, the first playback stutters for a while — does that mean it is not fixed?

No. Clearing the cache is a test in the driver and cache layer, and only helps where the cache is damaged. After clearing, the first playback has to rebuild the rendered frames and will stutter for a while — that is normal, not evidence that the fault remains. Clear the cache only through the software's own media cache controls, never by deleting cache folders by hand, since deleting the wrong folder can damage a project.

Q8: The corruption always appears at the same spot on the timeline — what does that mean?

It points at the clip or the effect at that position, rather than at memory or driver load. Corruption fixed to one position every time follows the material at that position; corruption that wanders to a different place on each play has no fixed address and points at memory or driver load instead. This fixed-versus-wandering logic follows the same idea as #476, which uses whether an export failure always stops at the same timecode.

Q9: The same clip plays fine in an ordinary player and the export is clean — does that mean the clip is undamaged, and can I ignore the green preview?

Together with a clean export it proves the source material is intact — a corrupt preview points at the preview render path, not at the clip. But it should not be ignored: a corrupt preview makes editing judgement unreliable, because colour, effects, and transitions all have to be judged from the preview. Work through the seven steps to separate the layers before deciding what to do next.

Q10: If all seven steps still cannot separate the layers, what should I bring for inspection?

The notes from the seven steps: which clip and codec, which effects were on, the preview resolution, the renderer setting, whether the corruption stays fixed at one position or wanders, and the result of every comparison. The shop is at Room 1209, Dali Plaza, Shek Mun, telephone 2186 8010, WhatsApp 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. There is no home-visit service and no Mong Kok shop.

Related articles

Related services

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