kurzgesagt is DEAD
i just re created kurzgesagt's entire channel in 10 minutes with claude opus
they have 0 competition because every video takes a full team of animators , voice actors and 1200+ hours to make
this video is pretty similar to kurzgesagt's channel but i used a free tts from claude so the audio sounds ass
elevenlabs fixes that , the animation may not be as smooth as kurzgesagt's but consumers wont be able to tell a diffrence
Imagine running ads like this with 100% retention rate
comment " Anthropic Mogs " if u want the code + prompt
leaked the prompt
nitter.gq/OnlySainted/status/211…
As promised here is the entire prompt for FREE
open claude code, paste the prompt, pick an idea, and your video is fully automated
thank me later ↓
<inputs>
Ask me for: one topic as a "what if" question (default: "What if Earth stopped spinning?"), the length (default 30 seconds), and whether I have an ElevenLabs key (if not, use a free local voice).
</inputs>
<direction>
A Kurzgesagt-style animated explainer at 1920x1080, 30fps, drawn entirely in code: one 2D canvas plus one WebGL canvas for planets.
Space: a vertical gradient #081d2c to #102348, five huge parallax nebula blobs (teal rgb 24,150,160 at 20%, blue 40,95,195 at 16%, coral 220,95,110 at 10%, indigo 60,70,175 at 15%, a second teal at 14%), three star layers at parallax depths .12/.35/.6 that twinkle, 40% of the near stars drawn as 4-point sparkles, a few tinted warm or cyan.
The Sun: radius 300px, sitting on the left, Earth orbiting it at 1250px so a 0.3x zoom shows the whole orbit. A radial gradient #fff7dc to #ffc857 to #ff8f4a with an offset highlight, additive halo rings at 1.23/1.48/1.76/2.1x the radius, 18 faint rotating ray wedges, 28 drifting glowing granules clipped inside the disc.
Earth: radius 200px, an orthographic sphere in a fragment shader, 4x supersampled, from a 2048x1024 map painted from longitude/latitude polygons (continents, major islands, inland seas cut back out): land #5fc47a, forest #48ad6b, desert #f5c75f, rainforest #3a9d62, ice #f1f8ff, ocean gradient #4dc7eb (shallow, from a blurred coast mask) to #1a66b8, white capsule clouds. Tilt .26 rad. Day/night uses smoothstep(-.12, .24, N·L) with the light direction recomputed from Sun to Earth every frame. Navy night side with ~380 warm city lights scattered around real cities, a coral band on the terminator, a cyan fresnel rim, ocean specular, an atmosphere halo that is stronger on the lit side.
The shader also takes these uniforms so the story can transform the planet in place: spin angle, pitch, wind (smears the clouds east), ocean (0 to 1, water drains to above 40° latitude and the exposed seabed is a cool grey-sand #a89e9e to #d6ccb8, never the desert gold), heat (lit side bleaches warm gold with a shimmer), frost (night side frosts from the dark pole of the terminator inward), equator glow, city-light pop.
A small astronaut mascot (~180px) built from rounded rectangles: suit #f3f7fb, orange #ff8f4a boots, a visor gradient #1f5a73 to #0a1c2c with glowing cyan eyes that look at whatever matters, a blinking antenna light, an idle bob, squash and stretch on every pose. Props for the funny beat: sunglasses that drop onto the visor, a puffy orange coat that springs on.
Type: Fredoka 600/700 only, embedded as base64. Captions in caps with 3-6px letter spacing in teal #8fd8e6, gold #ffd98a or white, sitting in rounded pills of rgb(6,20,40) at 62%. Every line of text rises 1.3x its own height through a clip on an ease-out-quint. Live counters count, never cut (1,670 KM/H, WIND SPEED, 24 HRS to 365 DAYS) and flip to gold when they hit the number that matters.
Finish: a vignette to rgba(4,4,20,.45) and a 6% film grain re-randomized 24 times per second.
Banned: outlines, flares or loops on the Sun's edge, static infographic strips, full stops in on-screen text, text that fades in whole, AI images or AI video, Kurzgesagt's own characters, bird or music.
</direction>
<script>
Default narration (70 words, lands at ~27s at ~157 wpm, leaving room for the title):
What if Earth suddenly stopped spinning? At the equator, the ground races east at nearly seventeen hundred kilometres an hour. Including you. The ground stops. You don't. Neither does the air. It becomes a supersonic wind. Then the oceans slide toward the poles, leaving one giant continent around the middle. And a day now lasts a whole year. Six months of sunlight. Six months of night. Pack sunglasses. And a coat.
Facts it rests on: a sudden stop, equatorial surface speed ~1,670 km/h, speed of sound at sea level ~1,235 km/h, without spin the equatorial bulge relaxes and the oceans migrate poleward, with no rotation one solar day equals one orbit.
</script>
<structure>
30 seconds, every line of narration gets a camera move or a transformation, nothing ever holds still. Times below are from the default voice, every event fires on its word, not on these numbers.
Tight on the spinning Earth at 2.25x (0s) → on "stopped" the globe brakes in 0.18s with a squash and stretch jolt, camera shake, a white flash and a shockwave ring (1.2s) → a 0.75s REWIND on a separate story clock that spins Earth back up (1.6s) → pull back to 1x revealing the Sun and the astronaut (2.4s) → on "equator" a glowing equator line, on "east" a gold arrow sweeps east along it (4.1s) → on "seventeen" a 1,670 KM/H counter slams in and counts up (4.8s) → dive to 2.3x on the equator with speed streaks (7.3s) → on "you" a YOU pin drops and the astronaut stands next to it, "1,670 KM/H" beside it (7.8s) → on "stops" the surface freezes with a thud, flash and ring, "GROUND 0 KM/H" (9.0s) → on "don't" the pin and the astronaut are flung east, "STILL MOVING 1,670 KM/H" pinned where they stood (10.2s) → pull back, on "air" white wind streaks wrap the globe, the clouds smear and a WIND SPEED counter climbs (11.6s) → on "supersonic" it passes 1,235, flips gold, a SUPERSONIC label rides the streaks (13.0s) → dive for "oceans", arrows flow to the poles while the water drains and land rises as a band, "ONE CONTINENT" rides it (15.0-18.3s) → one pull-back to a 0.3x orbit view, dotted orbit ring drawn in, Earth carried halfway round with the same face locked to the Sun, a ONE DAY LASTS counter goes 24 HRS to 365 DAYS on "year" (19.3-21.6s) → dive back in, on "sunlight" the lit side bleaches hot gold (22.4s) → pan to the dark side, on "night" frost spreads and the city lights pop (24.2s) → push to the astronaut, sunglasses drop on "sunglasses", the puffy coat springs on with "coat" (25.4-26.9s) → callback: the equator arrow returns frozen with "0 KM/H", then the title rises in, WHAT IF EARTH / STOPPED SPINNING (27.5s).
For any other topic, keep the same shape: the event happens on its word, pull back to show scale, dive in to make it personal ("you"), pull back out, the consequence with a live counter, an escalation, a funny beat with the mascot, a callback and the title.
</structure>
<build>
1. One HTML file. Every frame is a pure function of time: window.seek(t) draws frame t. No timers, no CSS animation, no randomness that isn't seeded.
2. Camera moves on easeInOutCubic, zoom is interpolated in log space, every pop is a damped spring (damping .42-.5, frequency 12-16). Add a slow drift and a 0.25%/s zoom creep so the frame never freezes.
3. Voice: ElevenLabs eleven_v4 with a designed voice (calm, warm, curious male, late 30s, crisp, light neutral European English), stability .5, similarity .75, generated with timestamps. Without a key, use kokoro-onnx locally (bm_fable at speed 1.25 reads ~157 wpm) and take word times from the model's own duration predictor: expose the encoder's /encoder/Clip_output_0 as an extra graph output, it gives frames per phoneme, 600 samples per frame at 24kHz. Fall back to faster-whisper only if that isn't possible. Generate sentence by sentence, trim each to its first and last word, 0.24s between sentences, a 0.36s beat before the big reveal. Every animation event fires on its word.
4. Keep two clocks: ambient time for twinkle, bob, grain and clouds, and a separate story clock for spin and events, so the rewind stays consistent.
5. The spin is integrated, not set: angle = sum of angular velocity, and every stop is a constant-deceleration brake (d - d²/2T), so there is never a jump.
6. Render with Playwright, GPU if there is one, otherwise headless Chromium with --use-angle=swiftshader. Seek every frame at 30fps and blend 2 subframes (shutter 0.6) for motion blur. Draw the globe once per frame and reuse it for the second subframe. Pipe PNGs straight into ffmpeg as a lossless segment, split the frames across one worker per CPU core, then concat. Encode with libx264 at CRF 16 for the master and CRF 20 for delivery, -tune animation, yuv420p, +faststart.
7. Sound: a calm space music bed whose chords resolve on the title and swell 5dB under it, plus one sound effect per event kind (brake, impact, rewind, pull-back whoosh, dive, counter tick, pin tap, fling, wind, sonic boom, reveal, light hit, frost, sunglasses, coat), each peak-aligned to its frame. Use royalty-free downloads if you can reach them, otherwise synthesise them in numpy. Balance by K-weighted loudness, not by peaks: bed at -21 LU under the voice, effects peaking at -9 LU under it, and a sidechain duck (40ms attack, 300ms release) that drops the bed 7dB and the effects up to 4dB under every word. Master with gain plus a 5ms lookahead limiter to -14 LUFS and a -2.9dB ceiling, so the file still measures under -2.2dB true peak after AAC encoding.
</build>
<gotchas>
Make the WebGL globe canvas at least 2 x radius x zoom x 1.18 pixels or the sphere clips flat in close-ups. Never draw prominences on the Sun's edge, they read as white hooks. Never pad the voice to fit the visuals, tighten the read instead. Rotate the clouds the same direction as the land or they look like they're slipping. Keep labels off the mascot's flight path, anchor them to where things were, not where they're going. Make the exposed seabed a different colour from the deserts or the continent beat reads as more desert. A voice that sounds fine solo gets buried if the bed is only 7dB down by RMS, measure in LUFS. AAC adds 0.5dB of true peak, leave headroom. An ffmpeg concat list resolves paths relative to the list file. Run render workers as children of a process you wait on, detached ones get killed. Show me a contact sheet of every beat, and the voice track, before rendering.
</gotchas>
<start>
Ask me the inputs, write the narration with the beat each word triggers, show it to me, then build and render the video.
</start>
As promised here is the entire prompt for FREE
open claude code, paste the prompt, pick an idea, and your video is fully automated
thank me later ↓
<inputs>
Ask me for: one topic as a "what if" question (default: "What if Earth stopped spinning?"), the length (default 30 seconds), and whether I have an ElevenLabs key (if not, use a free local voice).
</inputs>
<direction>
A Kurzgesagt-style animated explainer at 1920x1080, 30fps, drawn entirely in code: one 2D canvas plus one WebGL canvas for planets.
Space: a vertical gradient #081d2c to #102348, five huge parallax nebula blobs (teal rgb 24,150,160 at 20%, blue 40,95,195 at 16%, coral 220,95,110 at 10%, indigo 60,70,175 at 15%, a second teal at 14%), three star layers at parallax depths .12/.35/.6 that twinkle, 40% of the near stars drawn as 4-point sparkles, a few tinted warm or cyan.
The Sun: radius 300px, sitting on the left, Earth orbiting it at 1250px so a 0.3x zoom shows the whole orbit. A radial gradient #fff7dc to #ffc857 to #ff8f4a with an offset highlight, additive halo rings at 1.23/1.48/1.76/2.1x the radius, 18 faint rotating ray wedges, 28 drifting glowing granules clipped inside the disc.
Earth: radius 200px, an orthographic sphere in a fragment shader, 4x supersampled, from a 2048x1024 map painted from longitude/latitude polygons (continents, major islands, inland seas cut back out): land #5fc47a, forest #48ad6b, desert #f5c75f, rainforest #3a9d62, ice #f1f8ff, ocean gradient #4dc7eb (shallow, from a blurred coast mask) to #1a66b8, white capsule clouds. Tilt .26 rad. Day/night uses smoothstep(-.12, .24, N·L) with the light direction recomputed from Sun to Earth every frame. Navy night side with ~380 warm city lights scattered around real cities, a coral band on the terminator, a cyan fresnel rim, ocean specular, an atmosphere halo that is stronger on the lit side.
The shader also takes these uniforms so the story can transform the planet in place: spin angle, pitch, wind (smears the clouds east), ocean (0 to 1, water drains to above 40° latitude and the exposed seabed is a cool grey-sand #a89e9e to #d6ccb8, never the desert gold), heat (lit side bleaches warm gold with a shimmer), frost (night side frosts from the dark pole of the terminator inward), equator glow, city-light pop.
A small astronaut mascot (~180px) built from rounded rectangles: suit #f3f7fb, orange #ff8f4a boots, a visor gradient #1f5a73 to #0a1c2c with glowing cyan eyes that look at whatever matters, a blinking antenna light, an idle bob, squash and stretch on every pose. Props for the funny beat: sunglasses that drop onto the visor, a puffy orange coat that springs on.
Type: Fredoka 600/700 only, embedded as base64. Captions in caps with 3-6px letter spacing in teal #8fd8e6, gold #ffd98a or white, sitting in rounded pills of rgb(6,20,40) at 62%. Every line of text rises 1.3x its own height through a clip on an ease-out-quint. Live counters count, never cut (1,670 KM/H, WIND SPEED, 24 HRS to 365 DAYS) and flip to gold when they hit the number that matters.
Finish: a vignette to rgba(4,4,20,.45) and a 6% film grain re-randomized 24 times per second.
Banned: outlines, flares or loops on the Sun's edge, static infographic strips, full stops in on-screen text, text that fades in whole, AI images or AI video, Kurzgesagt's own characters, bird or music.
</direction>
<script>
Default narration (70 words, lands at ~27s at ~157 wpm, leaving room for the title):
What if Earth suddenly stopped spinning? At the equator, the ground races east at nearly seventeen hundred kilometres an hour. Including you. The ground stops. You don't. Neither does the air. It becomes a supersonic wind. Then the oceans slide toward the poles, leaving one giant continent around the middle. And a day now lasts a whole year. Six months of sunlight. Six months of night. Pack sunglasses. And a coat.
Facts it rests on: a sudden stop, equatorial surface speed ~1,670 km/h, speed of sound at sea level ~1,235 km/h, without spin the equatorial bulge relaxes and the oceans migrate poleward, with no rotation one solar day equals one orbit.
</script>
<structure>
30 seconds, every line of narration gets a camera move or a transformation, nothing ever holds still. Times below are from the default voice, every event fires on its word, not on these numbers.
Tight on the spinning Earth at 2.25x (0s) → on "stopped" the globe brakes in 0.18s with a squash and stretch jolt, camera shake, a white flash and a shockwave ring (1.2s) → a 0.75s REWIND on a separate story clock that spins Earth back up (1.6s) → pull back to 1x revealing the Sun and the astronaut (2.4s) → on "equator" a glowing equator line, on "east" a gold arrow sweeps east along it (4.1s) → on "seventeen" a 1,670 KM/H counter slams in and counts up (4.8s) → dive to 2.3x on the equator with speed streaks (7.3s) → on "you" a YOU pin drops and the astronaut stands next to it, "1,670 KM/H" beside it (7.8s) → on "stops" the surface freezes with a thud, flash and ring, "GROUND 0 KM/H" (9.0s) → on "don't" the pin and the astronaut are flung east, "STILL MOVING 1,670 KM/H" pinned where they stood (10.2s) → pull back, on "air" white wind streaks wrap the globe, the clouds smear and a WIND SPEED counter climbs (11.6s) → on "supersonic" it passes 1,235, flips gold, a SUPERSONIC label rides the streaks (13.0s) → dive for "oceans", arrows flow to the poles while the water drains and land rises as a band, "ONE CONTINENT" rides it (15.0-18.3s) → one pull-back to a 0.3x orbit view, dotted orbit ring drawn in, Earth carried halfway round with the same face locked to the Sun, a ONE DAY LASTS counter goes 24 HRS to 365 DAYS on "year" (19.3-21.6s) → dive back in, on "sunlight" the lit side bleaches hot gold (22.4s) → pan to the dark side, on "night" frost spreads and the city lights pop (24.2s) → push to the astronaut, sunglasses drop on "sunglasses", the puffy coat springs on with "coat" (25.4-26.9s) → callback: the equator arrow returns frozen with "0 KM/H", then the title rises in, WHAT IF EARTH / STOPPED SPINNING (27.5s).
For any other topic, keep the same shape: the event happens on its word, pull back to show scale, dive in to make it personal ("you"), pull back out, the consequence with a live counter, an escalation, a funny beat with the mascot, a callback and the title.
</structure>
<build>
1. One HTML file. Every frame is a pure function of time: window.seek(t) draws frame t. No timers, no CSS animation, no randomness that isn't seeded.
2. Camera moves on easeInOutCubic, zoom is interpolated in log space, every pop is a damped spring (damping .42-.5, frequency 12-16). Add a slow drift and a 0.25%/s zoom creep so the frame never freezes.
3. Voice: ElevenLabs eleven_v4 with a designed voice (calm, warm, curious male, late 30s, crisp, light neutral European English), stability .5, similarity .75, generated with timestamps. Without a key, use kokoro-onnx locally (bm_fable at speed 1.25 reads ~157 wpm) and take word times from the model's own duration predictor: expose the encoder's /encoder/Clip_output_0 as an extra graph output, it gives frames per phoneme, 600 samples per frame at 24kHz. Fall back to faster-whisper only if that isn't possible. Generate sentence by sentence, trim each to its first and last word, 0.24s between sentences, a 0.36s beat before the big reveal. Every animation event fires on its word.
4. Keep two clocks: ambient time for twinkle, bob, grain and clouds, and a separate story clock for spin and events, so the rewind stays consistent.
5. The spin is integrated, not set: angle = sum of angular velocity, and every stop is a constant-deceleration brake (d - d²/2T), so there is never a jump.
6. Render with Playwright, GPU if there is one, otherwise headless Chromium with --use-angle=swiftshader. Seek every frame at 30fps and blend 2 subframes (shutter 0.6) for motion blur. Draw the globe once per frame and reuse it for the second subframe. Pipe PNGs straight into ffmpeg as a lossless segment, split the frames across one worker per CPU core, then concat. Encode with libx264 at CRF 16 for the master and CRF 20 for delivery, -tune animation, yuv420p, +faststart.
7. Sound: a calm space music bed whose chords resolve on the title and swell 5dB under it, plus one sound effect per event kind (brake, impact, rewind, pull-back whoosh, dive, counter tick, pin tap, fling, wind, sonic boom, reveal, light hit, frost, sunglasses, coat), each peak-aligned to its frame. Use royalty-free downloads if you can reach them, otherwise synthesise them in numpy. Balance by K-weighted loudness, not by peaks: bed at -21 LU under the voice, effects peaking at -9 LU under it, and a sidechain duck (40ms attack, 300ms release) that drops the bed 7dB and the effects up to 4dB under every word. Master with gain plus a 5ms lookahead limiter to -14 LUFS and a -2.9dB ceiling, so the file still measures under -2.2dB true peak after AAC encoding.
</build>
<gotchas>
Make the WebGL globe canvas at least 2 x radius x zoom x 1.18 pixels or the sphere clips flat in close-ups. Never draw prominences on the Sun's edge, they read as white hooks. Never pad the voice to fit the visuals, tighten the read instead. Rotate the clouds the same direction as the land or they look like they're slipping. Keep labels off the mascot's flight path, anchor them to where things were, not where they're going. Make the exposed seabed a different colour from the deserts or the continent beat reads as more desert. A voice that sounds fine solo gets buried if the bed is only 7dB down by RMS, measure in LUFS. AAC adds 0.5dB of true peak, leave headroom. An ffmpeg concat list resolves paths relative to the list file. Run render workers as children of a process you wait on, detached ones get killed. Show me a contact sheet of every beat, and the voice track, before rendering.
</gotchas>
<start>
Ask me the inputs, write the narration with the beat each word triggers, show it to me, then build and render the video.
</start>
kurzgesagt is DEAD
i just re created kurzgesagt's entire channel in 10 minutes with claude opus
they have 0 competition because every video takes a full team of animators , voice actors and 1200+ hours to make
this video is pretty similar to kurzgesagt's channel but i used a free tts from claude so the audio sounds ass
elevenlabs fixes that , the animation may not be as smooth as kurzgesagt's but consumers wont be able to tell a diffrence
Imagine running ads like this with 100% retention rate
comment " Anthropic Mogs " if u want the code + prompt
saint retweeted
kurzgesagt is DEAD
i just re created kurzgesagt's entire channel in 10 minutes with claude opus
they have 0 competition because every video takes a full team of animators , voice actors and 1200+ hours to make
this video is pretty similar to kurzgesagt's channel but i used a free tts from claude so the audio sounds ass
elevenlabs fixes that , the animation may not be as smooth as kurzgesagt's but consumers wont be able to tell a diffrence
Imagine running ads like this with 100% retention rate
comment " Anthropic Mogs " if u want the code + prompt
saint retweeted
holy f*ck this ai elephant page hasnt had a reel under 1M in 18 days
-> 12 reels since oct 7
-> lowest one 1.1M
-> half of them over 9M
-> biggest one 48.7M
every caption ends by telling you who to send it to
"Send this to the one who claps loudest for you"
that one got 17.9M
people send it to the person it reminds them of and the page lands in front of someone new every time
if you run a character page end every caption with who to send it to
holy f*ck this ai elephant page hasnt had a reel under 1M in 18 days
-> 12 reels since oct 7
-> lowest one 1.1M
-> half of them over 9M
-> biggest one 48.7M
every caption ends by telling you who to send it to
"Send this to the one who claps loudest for you"
that one got 17.9M
people send it to the person it reminds them of and the page lands in front of someone new every time
if you run a character page end every caption with who to send it to
saint retweeted
i just recreated this entire channel in 10 minutes with claude opus
they have very low competition rn cuz its a really new channel
i turned it into a claude prompt that im giving out for free
open claude code, paste the prompt, pick an idea, and your video is fully automated
thank me later ↓
<inputs>
Ask me for: one premise as an "If your ___ ever ___… DO NOT ___" line (default: "If your Wi-Fi ever shows a network with your name on it… DO NOT CONNECT"), the length (default about 5:20, roughly 700 spoken words), one accent colour (default warm amber #FFB547), and whether I'll upload the narration as MP3s or you should generate it (ElevenLabs key, or a free local voice if not).
</inputs>
<direction>
An analog-horror "survival rules" explainer at 1920x1080, 30fps, drawn entirely in code: one HTML page with one full-frame SVG for the scene and HTML layers for HUD and captions.
Background: flat near-black #060606, a vignette to rgba(0,0,0,.5) from 52% radius, and a 2% film grain re-randomized every frame. Nothing else: no scanlines, no flicker, no colour grade. It must read as deep black with soft glowing shapes.
Pictograms: flat off-white #ececec with rounded corners and no outlines on filled shapes. A person (head r46, rounded-shoulder body 132x122, no face), a phone outline (12px stroke, radius 14% of width), the Wi-Fi symbol (dot r15, arcs r62/118/174 at 22px stroke, 90° each), houses, a router (340x110 body, two antennas, 4 status lights r11), a padlock, a door, a bed, a lying figure, a floor plan, 4 signal bars.
Glow: one SVG filter on the scene group, a 5px blur at 70% alpha plus a 22px blur at 38% alpha merged under the crisp source. A soft halo, never bloom.
Accent: only on the anomaly. The network row labelled YOUR NAME, its Wi-Fi arcs, filled signal bars, and a 4-corner bracket frame (12px stroke, 50px legs) that snaps onto whatever is "the copy". Red #E5322D only for one stamp and one slash.
HUD in grey #6f6f6f monospace 20px, .2em tracking: a case number top-left, a running timecode HH:MM:SS:FF top-right, seven 24px numbered boxes top-centre, filled with the accent once a rule has passed and outlined white on the current rule.
Rule cards: "RULE" in #d9d9d9 Montserrat ExtraBold 190px, the number centred with dominant-baseline central in a 190px accent rounded box (radius 26), a 600x7 accent line, then the rule in white caps at 52px, .08em tracking. Rule 7 gets a tilted red REDACTED stamp instead of text.
Type: Montserrat 600-900 only, plus one monospace for the HUD. Captions in caps, Bold 50px, .035em tracking, one sentence at a time, white with a 4px/14px soft glow, key words (CONNECT, DISCONNECTED, FAMILIAR, YOU) in the accent, centred 110px from the bottom.
Finish: RGB split (red and cyan drop-shadows offset 4-18px), 9 horizontal tear bars and an x-jitter only on glitch moments.
Banned: grain over 3%, scanlines, brightness flicker, hard cuts, text running under icons, stepped animation of any kind, gradients on icons, AI images or AI video, any real channel's characters, names, logos or catchphrases.
</direction>
<structure>
About 5:20, every sentence gets its own shot or transformation, nothing ever holds still. The one threat, stated by 1:00 and protected by every rule: the network is learning you so it can become you, and because two devices can't share one name, one of you gets disconnected.
A phone Wi-Fi list, rows sliding in 0.35s apart, the YOUR NAME row pulsing accent (0s) → on "rename" a text cursor blinks in the HOME_WIFI row (5s) → on "walk" a house cross-section, a figure crossing room to room with its Wi-Fi bars rising (7s) → on "seven rules" seven boxes pop in on springs, four light on "four", box 7 gets brackets on "seventh" (14-18s) → a title card (22s) → the list again, NEIGHBOUR, PRINTER and FBI_SURVEILLANCE_VAN highlighting on their words (26-34s) → houses broadcasting rings (36s) → on "yours" a new YOUR NAME row slides in (44s) → on "no lock" an open padlock (47s) → on "full bars" the arcs fill smoothly, a figure appears under the symbol on "close" (49-51s) → on "learns a phone" particles reverse direction between phone and router, a message, key, keyboard and moon icon land on their words (55-65s) → a COPY bar eases 60 to 94%, hits 100% with brackets on "finishes" (67-72s) → two identical figures under the same name, one flickers out on "disconnected" (75-81s) → "ONLINE 0 DAYS" vs "ONLINE 214 DAYS" (85s) → RULE 1 card (86s) → on "copy of you" a router morphs into a bracketed figure (106s) → on "familiar" the list shows HOME_WIFI · AUTO-JOIN beside YOUR NAME · NOT SAVED (110s) → RULE 2, a thumb hovers over the Wi-Fi toggle, the copy figure jitters harder on each "guessing" (119-140s) → RULE 3, a floor plan, bars gliding 2, 3, full as the dot moves, brackets at the bedroom door (142-150s) → on "two of you" a second figure fades in beside the door (158s) → a case file with redaction bars and a DISCONNECTED stamp, two identical HANNAH · ROOM 12 rows, two phones, two figures, one fades on "disconnected" (167-189s) → RULE 4, the plug slides out, a countdown from 30, router lights fade one by one, YOUR NAME still glowing (194-212s) → on the cut-off word a slow push in on a bracketed figure, then a 1.4s static burst, RGB split and tear bars (214s) → SIGNAL RESTORED in monospace (216s) → RULE 5, waves bounce wall to sofa to you, a dashed copy follows the figure a beat behind, a breathing waveform reaching MATCH 100% (220-250s) → RULE 6, a device list filling row by row, UNKNOWN · SAME MODEL · 02:14 in the accent, a finger hovers REMOVE and stops, brackets jumping between two figures on "real" (252-277s) → RULE 7 with the REDACTED stamp, a ghost of the RULE 1 card, the row changing to CONNECTED (279-300s) → on "disconnected" one figure fades with a glitch, brackets flicker between the two on "which" (305-312s) → the last word huge in the accent, a hum swell, cut to black on a boom (318s).
For any other premise keep the same shape: the threat stated early, one rule per section each with a real fact and a twist, a case file, a narrator line cut off mid-word, a rule 7 that pays off rule 1, and a one-word ending.
</structure>
<build>
1. One HTML file. Every frame is a pure function of time: window.renderAt(t) draws frame t. No timers, no CSS animation, no requestAnimationFrame.
2. Scenes are keyed to sentences. Each one crossfades in over 0.4s with the previous scene still drawn underneath, scaling 0.975 to 1 on easeOutCubic. The camera is one continuous drift for the whole video (zoom 1.02 ± 0.015 on a slow sine, pan ± 8px) and never resets per scene. Pops are damped springs (damping .45, frequency 14). Bars, lights and progress are fractional and eased.
3. Voice: a calm, emotionless public-safety announcer, 40s, medium-low, neutral American, ElevenLabs Multilingual v2 at speed .90, stability .70, similarity .75, style 0, one MP3 per section with break tags (0.4-1.5s) written in, and the cut-off line and its recovery as separate clips. About 140-150 words per minute. Fire every event on its word from timestamps; without them, align each sentence to the pauses in the audio with dynamic programming against the known script.
4. Sound, all synthesized in code: a drone bed about 17dB under the voice (46Hz and 69Hz sines with slow vibrato plus filtered brown noise on a 0.06Hz LFO, faded in over 3s), a falling 95 to 38Hz boom plus a soft 1320Hz tone on every rule card, a 1.4s gated-noise static burst with a 120Hz buzz between the cut-off line and its recovery, 0.35s bit-crushed glitch hits on the twist, a rising 55Hz hum under the last line, a final boom on the cut to black. Master with a limiter to -14 LUFS, -2dB true peak.
5. Render with Playwright headless Chromium: screenshot every frame at 30fps split across workers, then encode with libx264 at CRF 18 for the master and a capped copy under 400MB for sharing.
6. Also build 9:16 Shorts from the same scenes: a vertical mode that crops the scene viewBox to 1060x920, places captions at 70px in the lower middle, re-lays out the HUD, and ends on a "full video on this channel" card.
</build>
<gotchas>
Long network names must shrink to fit their row, never run under the padlock or Wi-Fi icons. Centre numbers in boxes with dominant-baseline central, not by eye. Grain over 3% is the first thing that looks cheap and also triples the file size. If the entrance zoom resets the camera on each scene it reads as a jump. Keep all scene content above the caption band. Captions shorter than half a second mean the alignment is off. Never pkill with a pattern that matches your own command line. Show me a contact sheet of one frame per scene, plus one mid-crossfade frame, before rendering.
</gotchas>
<start>
Ask me the inputs, write the narration split into voice sections with the beat each word triggers, show it to me, then build and render the video.
</start>
i just recreated this entire channel in 10 minutes with claude opus
they have very low competition rn cuz its a really new channel
i turned it into a claude prompt that im giving out for free
open claude code, paste the prompt, pick an idea, and your video is fully automated
thank me later ↓
<inputs>
Ask me for: one premise as an "If your ___ ever ___… DO NOT ___" line (default: "If your Wi-Fi ever shows a network with your name on it… DO NOT CONNECT"), the length (default about 5:20, roughly 700 spoken words), one accent colour (default warm amber #FFB547), and whether I'll upload the narration as MP3s or you should generate it (ElevenLabs key, or a free local voice if not).
</inputs>
<direction>
An analog-horror "survival rules" explainer at 1920x1080, 30fps, drawn entirely in code: one HTML page with one full-frame SVG for the scene and HTML layers for HUD and captions.
Background: flat near-black #060606, a vignette to rgba(0,0,0,.5) from 52% radius, and a 2% film grain re-randomized every frame. Nothing else: no scanlines, no flicker, no colour grade. It must read as deep black with soft glowing shapes.
Pictograms: flat off-white #ececec with rounded corners and no outlines on filled shapes. A person (head r46, rounded-shoulder body 132x122, no face), a phone outline (12px stroke, radius 14% of width), the Wi-Fi symbol (dot r15, arcs r62/118/174 at 22px stroke, 90° each), houses, a router (340x110 body, two antennas, 4 status lights r11), a padlock, a door, a bed, a lying figure, a floor plan, 4 signal bars.
Glow: one SVG filter on the scene group, a 5px blur at 70% alpha plus a 22px blur at 38% alpha merged under the crisp source. A soft halo, never bloom.
Accent: only on the anomaly. The network row labelled YOUR NAME, its Wi-Fi arcs, filled signal bars, and a 4-corner bracket frame (12px stroke, 50px legs) that snaps onto whatever is "the copy". Red #E5322D only for one stamp and one slash.
HUD in grey #6f6f6f monospace 20px, .2em tracking: a case number top-left, a running timecode HH:MM:SS:FF top-right, seven 24px numbered boxes top-centre, filled with the accent once a rule has passed and outlined white on the current rule.
Rule cards: "RULE" in #d9d9d9 Montserrat ExtraBold 190px, the number centred with dominant-baseline central in a 190px accent rounded box (radius 26), a 600x7 accent line, then the rule in white caps at 52px, .08em tracking. Rule 7 gets a tilted red REDACTED stamp instead of text.
Type: Montserrat 600-900 only, plus one monospace for the HUD. Captions in caps, Bold 50px, .035em tracking, one sentence at a time, white with a 4px/14px soft glow, key words (CONNECT, DISCONNECTED, FAMILIAR, YOU) in the accent, centred 110px from the bottom.
Finish: RGB split (red and cyan drop-shadows offset 4-18px), 9 horizontal tear bars and an x-jitter only on glitch moments.
Banned: grain over 3%, scanlines, brightness flicker, hard cuts, text running under icons, stepped animation of any kind, gradients on icons, AI images or AI video, any real channel's characters, names, logos or catchphrases.
</direction>
<structure>
About 5:20, every sentence gets its own shot or transformation, nothing ever holds still. The one threat, stated by 1:00 and protected by every rule: the network is learning you so it can become you, and because two devices can't share one name, one of you gets disconnected.
A phone Wi-Fi list, rows sliding in 0.35s apart, the YOUR NAME row pulsing accent (0s) → on "rename" a text cursor blinks in the HOME_WIFI row (5s) → on "walk" a house cross-section, a figure crossing room to room with its Wi-Fi bars rising (7s) → on "seven rules" seven boxes pop in on springs, four light on "four", box 7 gets brackets on "seventh" (14-18s) → a title card (22s) → the list again, NEIGHBOUR, PRINTER and FBI_SURVEILLANCE_VAN highlighting on their words (26-34s) → houses broadcasting rings (36s) → on "yours" a new YOUR NAME row slides in (44s) → on "no lock" an open padlock (47s) → on "full bars" the arcs fill smoothly, a figure appears under the symbol on "close" (49-51s) → on "learns a phone" particles reverse direction between phone and router, a message, key, keyboard and moon icon land on their words (55-65s) → a COPY bar eases 60 to 94%, hits 100% with brackets on "finishes" (67-72s) → two identical figures under the same name, one flickers out on "disconnected" (75-81s) → "ONLINE 0 DAYS" vs "ONLINE 214 DAYS" (85s) → RULE 1 card (86s) → on "copy of you" a router morphs into a bracketed figure (106s) → on "familiar" the list shows HOME_WIFI · AUTO-JOIN beside YOUR NAME · NOT SAVED (110s) → RULE 2, a thumb hovers over the Wi-Fi toggle, the copy figure jitters harder on each "guessing" (119-140s) → RULE 3, a floor plan, bars gliding 2, 3, full as the dot moves, brackets at the bedroom door (142-150s) → on "two of you" a second figure fades in beside the door (158s) → a case file with redaction bars and a DISCONNECTED stamp, two identical HANNAH · ROOM 12 rows, two phones, two figures, one fades on "disconnected" (167-189s) → RULE 4, the plug slides out, a countdown from 30, router lights fade one by one, YOUR NAME still glowing (194-212s) → on the cut-off word a slow push in on a bracketed figure, then a 1.4s static burst, RGB split and tear bars (214s) → SIGNAL RESTORED in monospace (216s) → RULE 5, waves bounce wall to sofa to you, a dashed copy follows the figure a beat behind, a breathing waveform reaching MATCH 100% (220-250s) → RULE 6, a device list filling row by row, UNKNOWN · SAME MODEL · 02:14 in the accent, a finger hovers REMOVE and stops, brackets jumping between two figures on "real" (252-277s) → RULE 7 with the REDACTED stamp, a ghost of the RULE 1 card, the row changing to CONNECTED (279-300s) → on "disconnected" one figure fades with a glitch, brackets flicker between the two on "which" (305-312s) → the last word huge in the accent, a hum swell, cut to black on a boom (318s).
For any other premise keep the same shape: the threat stated early, one rule per section each with a real fact and a twist, a case file, a narrator line cut off mid-word, a rule 7 that pays off rule 1, and a one-word ending.
</structure>
<build>
1. One HTML file. Every frame is a pure function of time: window.renderAt(t) draws frame t. No timers, no CSS animation, no requestAnimationFrame.
2. Scenes are keyed to sentences. Each one crossfades in over 0.4s with the previous scene still drawn underneath, scaling 0.975 to 1 on easeOutCubic. The camera is one continuous drift for the whole video (zoom 1.02 ± 0.015 on a slow sine, pan ± 8px) and never resets per scene. Pops are damped springs (damping .45, frequency 14). Bars, lights and progress are fractional and eased.
3. Voice: a calm, emotionless public-safety announcer, 40s, medium-low, neutral American, ElevenLabs Multilingual v2 at speed .90, stability .70, similarity .75, style 0, one MP3 per section with break tags (0.4-1.5s) written in, and the cut-off line and its recovery as separate clips. About 140-150 words per minute. Fire every event on its word from timestamps; without them, align each sentence to the pauses in the audio with dynamic programming against the known script.
4. Sound, all synthesized in code: a drone bed about 17dB under the voice (46Hz and 69Hz sines with slow vibrato plus filtered brown noise on a 0.06Hz LFO, faded in over 3s), a falling 95 to 38Hz boom plus a soft 1320Hz tone on every rule card, a 1.4s gated-noise static burst with a 120Hz buzz between the cut-off line and its recovery, 0.35s bit-crushed glitch hits on the twist, a rising 55Hz hum under the last line, a final boom on the cut to black. Master with a limiter to -14 LUFS, -2dB true peak.
5. Render with Playwright headless Chromium: screenshot every frame at 30fps split across workers, then encode with libx264 at CRF 18 for the master and a capped copy under 400MB for sharing.
6. Also build 9:16 Shorts from the same scenes: a vertical mode that crops the scene viewBox to 1060x920, places captions at 70px in the lower middle, re-lays out the HUD, and ends on a "full video on this channel" card.
</build>
<gotchas>
Long network names must shrink to fit their row, never run under the padlock or Wi-Fi icons. Centre numbers in boxes with dominant-baseline central, not by eye. Grain over 3% is the first thing that looks cheap and also triples the file size. If the entrance zoom resets the camera on each scene it reads as a jump. Keep all scene content above the caption band. Captions shorter than half a second mean the alignment is off. Never pkill with a pattern that matches your own command line. Show me a contact sheet of one frame per scene, plus one mid-crossfade frame, before rendering.
</gotchas>
<start>
Ask me the inputs, write the narration split into voice sections with the beat each word triggers, show it to me, then build and render the video.
</start>
saint retweeted
F*CK it here is exactly how i made these animations using ai fully automated for $20
Watch the video below first and then come back to read the post (Free sauce in comments)
i saw these stickman vector animations going insanely viral i did a bit of research and turns out this is fully automated using Claude code
Everything from the visuals to the voiceover and the sound effects is all 100% code
i trained claude with a few animations and gave it the prompt i sent below in the comments
it gave me back a rough version of this
after a few revisions this is what it made
it took me 22 hours to make all of this the breakdown is below
-> Bought a new claude account
-> Gave it an hours worth of content i scraped from ig
-> it analyzed everything
-> came up with a script using claude
-> created character design sheets using claude
-> created the background using claude
-> added the voice over using claude
-> added the sound effects using claude
this is a very early version of the animations i made
ive already made way higher quality stuff for brands to run as ads organically
for the higher quality production i use
-> Eleven labs for voice over
-> Free sfx
-> Capcut to stitch everything together and add sfx
-> GPT plus to make character sheets and background images + title cards
-> claude to code the whole video
i leaked the entire masterprompt for this project its the first link in the comments
F*CK it here is exactly how i made these animations using ai fully automated for $20
Watch the video below first and then come back to read the post (Free sauce in comments)
i saw these stickman vector animations going insanely viral i did a bit of research and turns out this is fully automated using Claude code
Everything from the visuals to the voiceover and the sound effects is all 100% code
i trained claude with a few animations and gave it the prompt i sent below in the comments
it gave me back a rough version of this
after a few revisions this is what it made
it took me 22 hours to make all of this the breakdown is below
-> Bought a new claude account
-> Gave it an hours worth of content i scraped from ig
-> it analyzed everything
-> came up with a script using claude
-> created character design sheets using claude
-> created the background using claude
-> added the voice over using claude
-> added the sound effects using claude
this is a very early version of the animations i made
ive already made way higher quality stuff for brands to run as ads organically
for the higher quality production i use
-> Eleven labs for voice over
-> Free sfx
-> Capcut to stitch everything together and add sfx
-> GPT plus to make character sheets and background images + title cards
-> claude to code the whole video
i leaked the entire masterprompt for this project its the first link in the comments
There are new youtube channels making what if videos
and making Hundreds of thousands of $$$
and its completely automated using ai
i oneshotted this in 2 hours with claude
if you want a full guide on how to create 3d videos and 2d videos like this lmk and maybe ill make a guide
ps: im working on a big project in the 2d video style follow along to see what that is.
saint retweeted
imagine building an app to $90K MRR
just to be overshadowed by a copy
this is exactly what happened to stalkie :
stalkie launched in january
their rival Leaked came out in july with exactly the same app but ...
Leaked went F*CKING INSANE on marketing
their main growth strategy was 100% organic :
-> AI UGC
-> Tiktok slideshows
-> Creator campaigns
and just 2 months later Leaked crossed $1M MRR while the original was still at $90K MRR
because Leaked focused more on distribuiting and marketing
this is the part app founders should care about :
you dont need to have a completely unique idea for an app to scale it you can build what works and outmarket everyone else
ive been using the same method across multiple apps so i am LEAKING the whole playbook FOR FREE
it has :
1. working hooks ive personally tested
2. my entire thought process on how i scale apps
3. A full 14 day guide on how id scale an app to $10k MRR
4. AI UGC formats
5. tiktok slideshow structures
6. how we test creatives
7. the entire organic growth playbook
Leaking it for free here :
brainextends-organic-playboo…
