anthropic shipped opus 5 today which we all know is INSANE (lmao) but people are fixated on the wrong thing entirely. "fable 5 quality at half the price." that's cool and all, but still a smoke screen the thing that faceless creators should be staring at sits further down the announcement, on the automation benchmark... opus 5 finishes full tasks end to end at 1.5x the quality of fable 5 you heard that right mf anthropic is trying to gatekeep this to protect fable 5 positioning as a "superior state-of-the-art" mythos model compared to Opus "lower classes" also to justify its prices but EVERY benchmark says the contrary on its lowest effort setting, opus 5 clears more of those tasks than every other model running at maximum. I REPEAT the cheapest version of the new model beats the MOST expensive version of everything else. so your AI youtube production can easily 10x in quality if you use this model correctly and since most of your competition is SLOW asf you have a very small window to get ahead if you execute fast enough here's how to take advantage of Opus 5 for faceless YouTube better than everyone else with EVERY prompt included: *bookmark this* ---------- STAGE 1: outlier research ----------- EFFORT: low to medium. bulk pattern work. save the budget for stage 3 PROMPT: you are a faceless youtube strategist analyzing outlier performance. you are a faceless youtube strategist analyzing outlier performance. - my niche: [niche] - my channel's typical view count: [number] - channels i'm benchmarking against: [3 to 5 channel names or urls] find the videos in this niche from the last 90 days that broke 5x their channel's own baseline. not the biggest channels, the biggest outliers relative to their own average, since a 2M view video on a 3M subscriber channel tells me nothing. for each outlier, separate the reason it popped into one of four buckets and commit to one: 1. topic selection (the subject itself was underserved) 2. packaging (title and thumbnail did the work, the video is average) 3. execution (the retention held and the algorithm pushed it) 4. external (trend, news cycle, collab, paid push) be honest about bucket 4. if a video popped because of a news cycle i cannot replicate, say so and move on. then give me the three angles from this set i can adapt to my channel without copying the topic outright. for each angle, name the specific psychological reason a viewer clicks, and name which of the four buckets my version would be winning on. output as a table, then three angle paragraphs underneath. do not summarize what is working in the niche. diagnose why. ---------------- STAGE 2: packaging ------------------- EFFORT: high. ctr decides whether the script ever gets watched. PROMPT: you are packaging a faceless youtube video before a single line of the script exists. - video angle: [paste the winning angle from stage 1] - niche: [niche] - target viewer: [be specific, not "people interested in history" but "25 to 40 year old men who watch documentary content late at night and have already seen the mainstream takes on this subject"] produce 12 titles. structure the set like this: - 4 curiosity gap titles (reveal an outcome, withhold the process) - 4 specificity titles (a number, name, date, or dollar amount doing the heavy lifting) - 4 contrarian titles (attack a belief the target viewer already holds) every title stays under 60 characters so it survives mobile truncation. no colons stacked on colons. no "you won't believe." then take your three strongest and build a thumbnail concept for each. for every concept give me: - the focal subject and where it sits in the frame - the emotional read on the face or object in under 5 words - the text overlay, 3 words maximum, and why those 3 - the contrast relationship between subject and background - what a viewer understands in 0.4 seconds of peripheral vision, before they read anything do not describe a vibe. describe composition. finish by naming which title and thumbnail pair creates the tightest promise, and state in one sentence exactly what the script now has to deliver to avoid a payoff betrayal. bear in mind that this composition will be later fed to ChatGPT images 2.0 for production ----------------- STAGE 3: the script ----------------- EFFORT: max. this is the one. do not cheap out here. PROMPT: you are writing a faceless youtube script built on retention psychology, not information delivery. - title: [locked title from stage 2] - thumbnail promise: [one sentence from stage 2] - video length target: [minutes] - format: [faceless documentary / listicle / story breakdown / explainer] first, before writing anything, build the framework and show it to me: for faceless content use character, concept, stakes. - character: who or what the viewer follows through the video. - concept: the single idea the video proves. - stakes: the deep emotional driver, not the surface benefit. "get more views" is surface. "spend another year building something nobody watches" is stakes. then write the script under these constraints. THE FIRST 15 SECONDS confirm what the video is about immediately. a viewer who does not know what they clicked into by second 8 leaves. state the subject, name the common belief, introduce the contrarian angle, promise something specific. frontload the most compelling material in the first 50 words. write the opening like a trailer: show the outcome, withhold the process. THE BODY every line does one of four things: introduces new information, moves the viewer toward the goal, creates tension by showing what failure looks like, or shifts their emotional state. a line doing none of the four gets cut. cut ruthlessly. structure around many payoffs, not one at the end. each payoff runs setup, then tension, then resolution, and the resolution immediately opens the next setup. the viewer never gets a clean moment to feel finished. plant a mini hook every 30 to 45 seconds. build tension first, let them try to guess, then reveal. giving the answer before the tension kills the effect. order the points deliberately: second strongest point opens, strongest point sits in the middle where engagement peaks, third strongest closes. watch the context to action ratio. minimal context to set the scene, then show the thing, then add context if the viewer needs more. scripts die from explaining. signpost constantly. remind the viewer where they are, what is coming, and why to stay. THE GRAND PAYOFF foreshadow the main payoff three times: once in the hook, once around a third of the way in, once around two thirds. each foreshadow tells the viewer this is coming and worth waiting for. the grand payoff connects every mini payoff into one system. it is not a surprise fifth secret. if the hook promised four mistakes, the payoff shows how the four compound. END ON A PAYOFF PROMISE satisfy the current curiosity, then open anticipation for the next video. output the full script with rough timestamps in the margin. write for the ear, not the eye. short sentences. no line reads like it was typed. ------------- STAGE 4: the retention audit --------------- EFFORT: high. opus 5 verifies its own work, so make it verify this. PROMPT: audit the script you just wrote against the four retention killers. be adversarial. you wrote this, so you are biased toward it. find the problems anyway. 1. DELAY DISEASE does the viewer know what this video is about by second 8? quote the exact line where confirmation lands. if the line sits past second 15, flag it and rewrite the opening. 2. CONTEXT DUMP find every stretch longer than 20 seconds of pure explanation with no demonstration, example, or scene. quote each one. rewrite the worst three. 3. PAYOFF VOID map every curiosity gap. for each one, mark where it opens and where it closes. flag any gap open longer than 90 seconds, and flag any resolution that fails to open a new gap immediately after. quote the dead spots. 4. GRAND PAYOFF BETRAYAL count the foreshadows of the main payoff. if fewer than three, say where the missing ones go. then check the payoff against the title promise word by word and tell me honestly whether a viewer would feel delivered to or clickbaited. then run the ai tell pass. strip: - any sentence in an "it's not x, it's y" shape - three short sentences of identical length stacked in a row - the same word or sentence skeleton opening two consecutive lines - rhetorical questions used to transition between sections - hedging language of any kind give me a retention risk score out of 10 for the first 30 seconds, the middle, and the final third, with the single highest-leverage fix for each. then output the corrected script in full. do not tell me the script is strong. tell me where it breaks. ------------ STAGE 5: production handoff ------------- EFFORT: medium, computer use enabled. opus 5 beats every model on computer use at any given cost, so stop pasting things back and forth. PROMPT: you have computer use. take the finished script and run production prep end to end. 1. break the script into voiceover segments at natural breath points, 15 to 25 seconds each, and export as a numbered text file. 2. for every segment, write a b-roll direction line: what is on screen, the shot type, and the movement. specific enough to search a stock library or feed an image generator without further interpretation. 3. build the visual asset list as a spreadsheet: segment number, shot description, search terms, whether it needs stock or generation, and a priority flag for the shots carrying the hook. 4. write the youtube description with the first two lines carrying the same curiosity gap as the title, then chapters pulled from the script timestamps. 5. generate 15 tags weighted toward search terms the target viewer types, not terms describing the video. verify your own output before handing back: open the files, check the segment count matches the script, confirm no segment runs past 25 seconds of spoken audio at 150 words per minute, and fix anything off before you tell me you are done. ------------------ STAGE 6: repurpose ----------------- EFFORT: medium. same session, context still loaded, no re-explaining. PROMPT: the script is finished. pull everything else out of it without re-writing from scratch. 1. find the single strongest 40 second stretch and cut it into a short. rewrite the opening line so the short stands alone, since a viewer landing on it has no context from the long form. 2. write two more shorts from the second and third strongest moments. each one opens a loop the long form resolves. 3. write a community post asking a question the video answers, without spoiling the answer. 4. write a newsletter section, 200 words, taking the one idea from the script with the highest standalone value. no "as i said in my video." it reads as its own piece. 5. write the pinned comment: one line of extra value the script left out, so it earns the top spot instead of begging for engagement. every piece keeps the voice of the script. nothing reads like a summary of the video. ----------------------- these prompts get you a repeatable pipeline. they do not get you the pattern data though a raw model has read the whole internet. a trained system has read 8,000 scripts that already worked, across 50+ niches, carrying 200M+ views of evidence about which hooks hold and which ones leak. and i've grabbed all that EXACT data (with those exact numbers) directly from my experience of scriptwriting videos for the biggest creators in the space and packaged it ALL into facelessOS: the best AI scriptwriting system and community in the world for faceless youtube, trained on $5M+ in client revenue generated with viral scripts DM me "OS" if interested (will respond to all of you)