GTA 6 thumbnails, redo
The redo of your GTA thumbnails. This time each video starts with the research: what the winning GTA thumbnails for that video actually do, counted, with the examples. Every thumbnail names the pattern it copies and the question it puts in the viewer's head. The first set is still at v1.
What changed from the first set
- Intrigue: every thumbnail copies a proven GTA winner pattern and names the question it puts in your head (shown under each one). A tease the video answers is allowed now; the first set only allowed literal scenes.
- Text: it follows what the GTA winners do (counted per video below), not your main-channel no-text rule. Drawn in post, never by the AI.
- Research: each video starts with a fresh pull of the winning thumbnails for that video and topic, counted on one sheet (text, circles and arrows, faces, logos, the intrigue device), shown first in every tab.
- One check, then shipped: one independent QA, including a cold read at phone size (what question does it raise?), one fix pass, done.
- Checked against the scripts: 12 of the hooks (SMELLS?, CAN THEY MARRY?, GTA 6 JUDGES YOU, NO ONLINE?, TOO OLD?, $100?, CUT?, 11X BIGGER? and more) checked line by line in the video scripts; all true. The YouTube quota ran out again on day two (another session shares the key), so view counts come from the October niche data.
Your calls
- Should you be in GTA thumbnails at all? In all 4 videos the research found almost no winner with a creator's face (0 to 1 per video). You are in 3 to 4 per video only so you can judge, and those are mostly the weaker ones.
- Text on the page now follows the GTA winners: short verdicts, questions and numbers, never a plain label.
GTA 6 in 20 Minutes: Everything Confirmed



15 Things GTA 5 Lets You Do That GTA 6 Won't



How We Imagined GTA 6 vs What We Got



GTA 6: Everything Rockstar Just Revealed



Research (26 winners vs 15 normals): text is on almost all of both; winners use fewer words (median 3 vs 5) and circle ONE thing in the game world. No winner shows a real creator, so you are in 4 here only so you can judge. The YouTube quota ran out again mid-pull, so view counts are from the 10-25 niche data.
What wins for this video (research)
# Wins for "GTA 6 in 20 Minutes: Everything Confirmed" (v2, 2026-11-05)
Sample: 26 winners (same-video GTA 6 uploads, last 18 months, 100K+ views, 3.1x to 227x) and 15 normals (the same
channels' same-topic uploads below their channel median). Ratio = lower of views/subs and views/channel median. Stats:
niche corpus 10-25/26 (API refresh: 403 quota, api_log.txt).
Winners vs normals
- Text: 24 of 26 vs 13 of 15. Text is the baseline, not the edge; winners use fewer words (median 3 vs 5).
Winner kinds: 11 labels, 7 numbers, 4 hook phrases, 1 question, 1 phone message. White or yellow, black stroke,
median 14.5 percent of frame height.
- Faces: 0 of 41 show a real creator. 14 of 26 winners show no face vs 3 of 15 normals (normals lean on key-art
characters). Hafu is in 4 of 10 anyway (his rule): his cards borrow one winner device and add him reacting.
- GTA VI logo: 20 of 26 vs 11 of 15, baseline.
- Marks: 5 of 26 vs 7 of 15. Winners circle ONE thing in the game world; normals circle icons, a tweet, or four
things at once (Project Vice, 0.19x).
- Both date questions ("GTA 6 PC IN 2027?!", "ONLINE 2027?") are normals: tease a weird detail, not a date.
- Plain "N details" counts appear on both sides; only "4 BILLION" stands out.
Patterns used:
1. One circled world detail + "99% OF PEOPLE MISSED": Krown Fitz, "CRAZY Details You Probably MISSED in The GTA 6 Extended Look", 38.0x.
2. Absurd number + marks: The Big J, "4 Billion Details YOU MISSED in GTA 6", 20.8x.
3. Character + in-game phone message: Misteri, "100+ CONFIRMED Gameplay Features in GTA 6", 22.8x.
4. Split of one subject: Leonida Man, "GTA 6 Trailer 1 vs Trailer 2, HUGE Changes You Missed!", 17.7x.
5. Big yellow multiplier: No Fast Travel, "GTA 6's Map Is WAY Bigger Than We Thought", 10.4x.
6. Short hook phrase, white + yellow over a GTA scene: No Fast Travel, "GTA 6 in 25 Days, The Complete Guide Before You Play", 10.2x.
Sample: 26 winners (same-video GTA 6 uploads, last 18 months, 100K+ views, 3.1x to 227x) and 15 normals (the same
channels' same-topic uploads below their channel median). Ratio = lower of views/subs and views/channel median. Stats:
niche corpus 10-25/26 (API refresh: 403 quota, api_log.txt).
Winners vs normals
- Text: 24 of 26 vs 13 of 15. Text is the baseline, not the edge; winners use fewer words (median 3 vs 5).
Winner kinds: 11 labels, 7 numbers, 4 hook phrases, 1 question, 1 phone message. White or yellow, black stroke,
median 14.5 percent of frame height.
- Faces: 0 of 41 show a real creator. 14 of 26 winners show no face vs 3 of 15 normals (normals lean on key-art
characters). Hafu is in 4 of 10 anyway (his rule): his cards borrow one winner device and add him reacting.
- GTA VI logo: 20 of 26 vs 11 of 15, baseline.
- Marks: 5 of 26 vs 7 of 15. Winners circle ONE thing in the game world; normals circle icons, a tweet, or four
things at once (Project Vice, 0.19x).
- Both date questions ("GTA 6 PC IN 2027?!", "ONLINE 2027?") are normals: tease a weird detail, not a date.
- Plain "N details" counts appear on both sides; only "4 BILLION" stands out.
Patterns used:
1. One circled world detail + "99% OF PEOPLE MISSED": Krown Fitz, "CRAZY Details You Probably MISSED in The GTA 6 Extended Look", 38.0x.
2. Absurd number + marks: The Big J, "4 Billion Details YOU MISSED in GTA 6", 20.8x.
3. Character + in-game phone message: Misteri, "100+ CONFIRMED Gameplay Features in GTA 6", 22.8x.
4. Split of one subject: Leonida Man, "GTA 6 Trailer 1 vs Trailer 2, HUGE Changes You Missed!", 17.7x.
5. Big yellow multiplier: No Fast Travel, "GTA 6's Map Is WAY Bigger Than We Thought", 10.4x.
6. Short hook phrase, white + yellow over a GTA scene: No Fast Travel, "GTA 6 in 25 Days, The Complete Guide Before You Play", 10.2x.
The 10 thumbnails
Every AI render (picked = green)
Research (29 winners vs 14 normals): having text does not separate them; WHAT it says does: winners use short verdicts or numbers, normals use information labels. All 6 winners with this exact title use one layout: GTA V logo left, GTA VI logo right, one character in the middle, and no normal uses it (cards 1 and 2 copy it with the real logos). No winner shows a creator face; you are in 4 only so you can judge. Flags: small AI flaws on the rocket launcher (cards 1 and 10 share that item); the witness in "1 WITNESS" is tiny at phone size.
What wins for this video (research)
# Research: "15 Things GTA 5 Lets You Do That GTA 6 Won't" (v2, 2026-11-05)
Sources: v1 pulls (348 videos), the 10-30 pool, niche channel medians (10-26) for normals; the YouTube key was out of
quota (api_log.txt). Tagged by eye at full size: 29 winners (3x to 523x, 100K+, since May 2025) and 14 normals (same
channels, same topic, under median). Tags: winners.json; picture: winners_sheet.jpg.
## Winners vs normals
- Text: 21 of 29 winners, 13 of 14 normals, median 3 words. The KIND decides: winners carry 11 hook verdicts ("SCARY
REAL POLICE", "GETTING RICH WILL BE HARD", "Already Cut...") and 4 shocking numbers; normals carry 6 info labels
("EVERY CONFIRMED DETAIL", "NEW MECHANICS"). White heavy sans, black stroke, second line red or yellow, 10 to 45
percent of frame height.
- Exact title (COLETHEMAN 174x, MrGTA 66.9x, BreeZe 64x, Olegator 19.4x, Igor Alves 11.8x, Glooby 5.2x): 6 of 6 use one
template, no hook words: GTA V logo left, GTA VI logo right, one GTA 6 character centred, Los Santos day behind V,
Vice City neon behind VI. 0 of 14 normals.
- GTA VI logo: 20 of 29 vs 9 of 14; both logos: 11 vs 2.
- Red arrow at a real detail: 3 of 29 (Game guru 31.9x points at a game-UI money box). Blur: 0 winners, 2 normals.
- Faces: 20 winners show a GTA character, 0 a creator; normals 1 creator (0.53x). Creator faces are unproven here, so
Hafu is in 4 of 10 (minimum 3), as the reacting creator.
## Patterns used
1. Logo split, one hero: "10 CRAZY THINGS In GTA 6 That You Can't Do In GTA 5", COLETHEMAN GTA, 174x.
2. Verdict text + threat face: "La Nueva IA Policial de GTA 6 Te Caza", Lottik, 43.4x.
3. Verdict text + real game-UI box + red arrow: "Why Getting Rich in GTA 6 Will Be HARD", Game guru, 31.9x.
4. Shocking realism verdict: "Los NPC de GTA 6 Son Perturbadoramente Reales", Lottik, 11.2x.
5. Removed verdict + reaction: "Rockstar Already Cut These GTA 6 Features", Dandy Andy, 3.8x.
6. Deadpan phrase over a gameplay shot: "I Turned GTA 5 Into GTA 6 With Only 8 Mods", Project VI, 340x.
Font: Roboto Black (winner fonts are heavy sans of the same weight).
Sources: v1 pulls (348 videos), the 10-30 pool, niche channel medians (10-26) for normals; the YouTube key was out of
quota (api_log.txt). Tagged by eye at full size: 29 winners (3x to 523x, 100K+, since May 2025) and 14 normals (same
channels, same topic, under median). Tags: winners.json; picture: winners_sheet.jpg.
## Winners vs normals
- Text: 21 of 29 winners, 13 of 14 normals, median 3 words. The KIND decides: winners carry 11 hook verdicts ("SCARY
REAL POLICE", "GETTING RICH WILL BE HARD", "Already Cut...") and 4 shocking numbers; normals carry 6 info labels
("EVERY CONFIRMED DETAIL", "NEW MECHANICS"). White heavy sans, black stroke, second line red or yellow, 10 to 45
percent of frame height.
- Exact title (COLETHEMAN 174x, MrGTA 66.9x, BreeZe 64x, Olegator 19.4x, Igor Alves 11.8x, Glooby 5.2x): 6 of 6 use one
template, no hook words: GTA V logo left, GTA VI logo right, one GTA 6 character centred, Los Santos day behind V,
Vice City neon behind VI. 0 of 14 normals.
- GTA VI logo: 20 of 29 vs 9 of 14; both logos: 11 vs 2.
- Red arrow at a real detail: 3 of 29 (Game guru 31.9x points at a game-UI money box). Blur: 0 winners, 2 normals.
- Faces: 20 winners show a GTA character, 0 a creator; normals 1 creator (0.53x). Creator faces are unproven here, so
Hafu is in 4 of 10 (minimum 3), as the reacting creator.
## Patterns used
1. Logo split, one hero: "10 CRAZY THINGS In GTA 6 That You Can't Do In GTA 5", COLETHEMAN GTA, 174x.
2. Verdict text + threat face: "La Nueva IA Policial de GTA 6 Te Caza", Lottik, 43.4x.
3. Verdict text + real game-UI box + red arrow: "Why Getting Rich in GTA 6 Will Be HARD", Game guru, 31.9x.
4. Shocking realism verdict: "Los NPC de GTA 6 Son Perturbadoramente Reales", Lottik, 11.2x.
5. Removed verdict + reaction: "Rockstar Already Cut These GTA 6 Features", Dandy Andy, 3.8x.
6. Deadpan phrase over a gameplay shot: "I Turned GTA 5 Into GTA 6 With Only 8 Mods", Project VI, 340x.
Font: Roboto Black (winner fonts are heavy sans of the same weight).
The 10 thumbnails
Every AI render (picked = green)
Research (27 winners vs 13 normals, median 10.8x vs 0.45x): text is on almost all of both; what wins is WHAT the text says, a promise, a date or a claim (never vague phrases or the game name alone). No winner shows a creator face. Four of these use no AI at all: the real 2019 fan trailer, the real 2020 comments and official Rockstar stills (this format got "AI thumbnail" comments). Weakest: "$100?" (busy shelves, AI case art) and "2019 LEAK: RICARDO?" (painted look).
What wins for this video (research)
# What wins for "How We Imagined GTA 6 vs What We Got" (v2, 2026-11-05)
winners.json tags 27 winners and 13 normals by eye. Ratio = views over the median of the 50 long
uploads around it (2026-11-04 caches of 36 GTA channels; API out of quota, see api_log.txt). 26 winners are GTA or
Rockstar, 1 is Nick930's Anthem. Median ratio 10.8x vs 0.45x.
## Counts, winners vs normals
- Text: 24 of 27 vs 13 of 13. The norm. Median 3 words; white with a thick black stroke,
yellow for years and numbers; biggest line about 12 percent of frame height.
- What the text says separates them. Winners name a promise, a date or a claim ("2011 / 2024 TRAILER / REALITY",
"2X", "ARE THESE NPCS TOO REAL?"). Normals are vague ("THE REAL STORY", "YEAH...") or only name the games.
- Same template, both sides: Onespot "TRAILER / REALITY" split 32.6x, logo-only splits 0.29 to 0.45x; Trojan comment
cards 13.9 to 59x when the comment is a claim history settled ("DELAYED TO 2008!!!!!!"), 0.65x for insults.
- Split or grid: 14 of 27 vs 5 of 13. Circle, arrow or X: 9 of 27 vs 2 of 13.
- Faces: a GTA character in 11 of 27 vs 5 of 13. A creator's face in 0 of 40, so the research gives Hafu no pattern;
his 3 cards use the one reaction-face winner (Glouton).
- GTA VI logo: 16 of 27 vs 8 of 13.
## Patterns used
1. Fan cover vs real cover, years and arrow: dLorer "COMO A GENTE IMAGINAVA GTA 6?" 10.8x
2. Dated receipt card: Trojan "How The Community Reacted To RDR2" 59.2x
3. Circled detail: Mr Snook "ALL Differences From the FIRST GTA 5 Version ..." 31.9x
4. Question over one big hero: Dandy Andy "GTA 6 NPCs Are WAY More Advanced..." 9.2x
5. Map then vs now: No Fast Travel "GTA 6's Map Is WAY Bigger..." 11.9x
6. Mystery figure: Hazard "GTA 6 LEAKER JUST DROPPED 7 MORE LEAKS" 4.5x
7. Reaction face: Glouton "Ce qu'on imaginait pour GTA 6 en 2014..." 6.2x
winners.json tags 27 winners and 13 normals by eye. Ratio = views over the median of the 50 long
uploads around it (2026-11-04 caches of 36 GTA channels; API out of quota, see api_log.txt). 26 winners are GTA or
Rockstar, 1 is Nick930's Anthem. Median ratio 10.8x vs 0.45x.
## Counts, winners vs normals
- Text: 24 of 27 vs 13 of 13. The norm. Median 3 words; white with a thick black stroke,
yellow for years and numbers; biggest line about 12 percent of frame height.
- What the text says separates them. Winners name a promise, a date or a claim ("2011 / 2024 TRAILER / REALITY",
"2X", "ARE THESE NPCS TOO REAL?"). Normals are vague ("THE REAL STORY", "YEAH...") or only name the games.
- Same template, both sides: Onespot "TRAILER / REALITY" split 32.6x, logo-only splits 0.29 to 0.45x; Trojan comment
cards 13.9 to 59x when the comment is a claim history settled ("DELAYED TO 2008!!!!!!"), 0.65x for insults.
- Split or grid: 14 of 27 vs 5 of 13. Circle, arrow or X: 9 of 27 vs 2 of 13.
- Faces: a GTA character in 11 of 27 vs 5 of 13. A creator's face in 0 of 40, so the research gives Hafu no pattern;
his 3 cards use the one reaction-face winner (Glouton).
- GTA VI logo: 16 of 27 vs 8 of 13.
## Patterns used
1. Fan cover vs real cover, years and arrow: dLorer "COMO A GENTE IMAGINAVA GTA 6?" 10.8x
2. Dated receipt card: Trojan "How The Community Reacted To RDR2" 59.2x
3. Circled detail: Mr Snook "ALL Differences From the FIRST GTA 5 Version ..." 31.9x
4. Question over one big hero: Dandy Andy "GTA 6 NPCs Are WAY More Advanced..." 9.2x
5. Map then vs now: No Fast Travel "GTA 6's Map Is WAY Bigger..." 11.9x
6. Mystery figure: Hazard "GTA 6 LEAKER JUST DROPPED 7 MORE LEAKS" 4.5x
7. Reaction face: Glouton "Ce qu'on imaginait pour GTA 6 en 2014..." 6.2x
The 10 thumbnails
Every AI render (picked = green)
Research (30 winners vs 14 normals): what separates winners is ONE image instead of a collage (collages 2 of 30 vs 5 of 14), a hook phrase (11 of 30 vs 2 of 14), an intrigue device (18 of 30 vs 5 of 14) and the VI logo (23 of 30 vs 9 of 14). A creator face is in only 1 of 30. Five of these use the REAL Edge 430 screenshots. You are in 3 only so you can judge: two of those miss the 0.72 face floor ("NEW WEAPON?" 0.55, your hand covers your nose; "DON'T SHOOT IT" 0.66), flagged, not re-rendered (one QA round, your call 11-04).
What wins for this video (research)
# gta4-everything-revealed v2: what wins (2026-11-05)
Data: 30 winners (3x+ channel median, 100K+ views, since 2025-05; 29 same-topic GTA news, screenshot and "everything"
uploads from GTA channels, plus MythPlays, the one creator-face winner) vs 14 normals (same channels, same topic, under
1x). YouTube quota was out (api_log.txt): ratios use the 11-04 upload caches.
Winners vs normals:
- Any text: 25 of 30 vs 11 of 14. The kind separates: hook phrases 11 of 30 (37%) vs 2 of 14 (14%); plain labels
("OVERANALYZING") 7 of 30 vs 5 of 14. Median 3 words, 18% of frame height, white with thick black stroke plus one
yellow, green or magenta word.
- GTA VI logo: 23 of 30 vs 9 of 14.
- One image, not a collage: collages and splits 2 of 30 (7%) vs 5 of 14 (36%). The strongest separator.
- An intrigue device: 18 of 30 (60%) vs 5 of 14 (36%): rare event 8, big number 4, circled detail 4, question 1,
strange creature 1.
- Circles, arrows, zoom boxes: 3 of 30 vs 2 of 14. They win with a "MISSED" hook on one real still (35x); "LOOK
CLOSER" on a collage flopped (0.1x).
- Faces: GTA characters 14 of 30 vs 8 of 14. Creator face 1 of 30, 0 of 14: winners here barely use one. Hafu is in
3 of 10 by rule, on the MythPlays layout.
- Same Edge news: 5 channels used the same skyline still; best "20 NEW DETAILS" 7.0x. The Edge shark under
low-contrast "100+ DETAILS" did 0.21x; a 4-panel collage 1.9x; South Park + score 0.8x.
Patterns used:
1. Number + VI logo on one real still: RoddyGames "56 GTA 6 Details ... NEW Screenshots", 113x.
2. Circled detail + "missed" hook: The Krown Fitz "CRAZY Details You Probably MISSED ...", 35x.
3. Rare event + short hook: GameSight "Everything We've NEVER Seen Before", 62x.
4. Character close-up + short hook or question: RENZI "Everything We Know About Gangs", 116x.
5. Creator shocked + arrow into the game: MythPlays "GTA 6 IS CRAZY", 4.3x (5.5M views).
Data: 30 winners (3x+ channel median, 100K+ views, since 2025-05; 29 same-topic GTA news, screenshot and "everything"
uploads from GTA channels, plus MythPlays, the one creator-face winner) vs 14 normals (same channels, same topic, under
1x). YouTube quota was out (api_log.txt): ratios use the 11-04 upload caches.
Winners vs normals:
- Any text: 25 of 30 vs 11 of 14. The kind separates: hook phrases 11 of 30 (37%) vs 2 of 14 (14%); plain labels
("OVERANALYZING") 7 of 30 vs 5 of 14. Median 3 words, 18% of frame height, white with thick black stroke plus one
yellow, green or magenta word.
- GTA VI logo: 23 of 30 vs 9 of 14.
- One image, not a collage: collages and splits 2 of 30 (7%) vs 5 of 14 (36%). The strongest separator.
- An intrigue device: 18 of 30 (60%) vs 5 of 14 (36%): rare event 8, big number 4, circled detail 4, question 1,
strange creature 1.
- Circles, arrows, zoom boxes: 3 of 30 vs 2 of 14. They win with a "MISSED" hook on one real still (35x); "LOOK
CLOSER" on a collage flopped (0.1x).
- Faces: GTA characters 14 of 30 vs 8 of 14. Creator face 1 of 30, 0 of 14: winners here barely use one. Hafu is in
3 of 10 by rule, on the MythPlays layout.
- Same Edge news: 5 channels used the same skyline still; best "20 NEW DETAILS" 7.0x. The Edge shark under
low-contrast "100+ DETAILS" did 0.21x; a 4-panel collage 1.9x; South Park + score 0.8x.
Patterns used:
1. Number + VI logo on one real still: RoddyGames "56 GTA 6 Details ... NEW Screenshots", 113x.
2. Circled detail + "missed" hook: The Krown Fitz "CRAZY Details You Probably MISSED ...", 35x.
3. Rare event + short hook: GameSight "Everything We've NEVER Seen Before", 62x.
4. Character close-up + short hook or question: RENZI "Everything We Know About Gangs", 116x.
5. Creator shocked + arrow into the game: MythPlays "GTA 6 IS CRAZY", 4.3x (5.5M views).
The 10 thumbnails
Every AI render (picked = green)
1. What is happening behind Jason?
2. Twelve new screenshots, and one is a shark under the sea: what are the other eleven?
3. The cops know what I was wearing?
4. Can Jason and Lucia actually get married?
5. GTA 6 judges me? On what?