Claude Fable 5 Prompts for YouTube: Scripts, Hooks, Titles, Thumbnails & Shorts
45+ battle-tested Claude Fable 5 prompts for YouTube — long-form scripts, hooks, titles, thumbnail copy, image prompts, and Shorts. The full creator stack.
P
PromptsRushJune 10, 2026
•19 min read
On this page
12 sections
Claude Fable 5 Prompts for YouTube: Scripts, Hooks, Titles, Thumbnails & Shorts
Read
0%
2 views
Claude Fable 5 is the first frontier model that writes YouTube scripts like a working creator rather than a model imitating one. The prose has rhythm, hooks land, scripts breathe on camera. We have been running it as the default writing model in our content stack for the last month, and the gap to what we were getting from Opus 4.7 and GPT-5.5 on this specific job is real.
Short version. Fable 5 is the model you reach for when the writing has to perform on camera. Below: 45+ prompts we actually use — long-form scripts, cold-open hooks, titles, thumbnail copy, thumbnail image prompts, and Shorts. Lift any of them, replace the placeholders, and ship.
Each section has hero prompts with a copy button. The rest are numbered and ready to use as-is.
Why Fable 5 Is the Right Model for YouTube
Three properties of Fable 5 matter for YouTube work specifically:
Voice retention across long context. Paste in 20 of your old scripts and Fable 5 holds your voice across a fresh 1,500-word script. No other frontier model does this as cleanly.
Hook quality. Fable 5 understands the difference between an opening that sounds like a hook and one that earns the next eight seconds. Most models still write the first version.
Sentence rhythm for spoken delivery. Scripts that read fine on paper often die on camera. Fable 5's default rhythm is closer to how a human actually talks.
You can run all of these in Claude.ai directly, the Anthropic API, or as system prompts inside Claude Projects. For repeat use, save them as Projects with your channel's past scripts in the project knowledge — that turns every script generation into a one-message ask.
Pro tip: Before any prompt below, paste in 3-5 transcripts from your best-performing videos. Fable 5 will infer your voice without you having to describe it. This is the single biggest unlock.
Category 1 — Long-Form Scripts (Prompts 1–10)
For 8-25 minute videos. These prompts handle the full arc from cold open to mid-roll to CTA.
Prompt 1 — Full 12-Minute Script in Your Voice
Ready to use
I have pasted three of my top-performing video transcripts above. Identify my voice — sentence rhythm, openings, vocabulary, the connective phrases I lean on, what I avoid. Then write a fresh 12-minute video script on {{topic}} in that voice. Structure: 8-second cold open hook, 30-second value promise, 4-5 main beats with B-roll suggestions in [brackets], a single mid-roll re-engagement at the 6-minute mark, and a CTA. Match my cadence, not just my word choice. Open with the conclusion of the video, then earn it back.
Write a 10-minute video script using the story-spine structure: once there was, every day, until one day, because of that (x3), until finally, and ever since. Topic: {{topic}}. The story is the spine — the educational payoff hangs off it. Maximum 1,400 words. First line is a single sentence under 10 words. Include B-roll cues in brackets at every scene change. End with a question to the audience.
Educational deep-dive script. Write a 15-minute script that teaches {{concept}} from first principles. Structure: hook (why this matters in 30 seconds), three foundational ideas, one counter-intuitive insight, an application example, a common mistake to avoid, and a one-sentence summary. No filler. Audience: intermediate, not beginner.
Tutorial / how-to script. Write a step-by-step tutorial script for {{task}}. Constraints: state the finished result in the first 15 seconds, list what they need before starting, walk through 5-8 steps with clear verbal cues for screen recording moments, and end with the most common mistake people make. Length: 8-10 minutes.
Listicle script that does not feel like a listicle. Write a script for "{{N}} {{things}} for {{outcome}}" but structure it so the items build on each other rather than just stacking. Use callbacks. The last item should reference the first. Length: 9 minutes.
Reaction / commentary script. I am reacting to {{source content}}. Write a 7-minute script that adds substantive commentary, not just "wow that's crazy." Structure: my hot take in the first 15 seconds, three specific moments to call out with my analysis, and a final verdict. Voice: opinionated, willing to be wrong.
Interview prep script. I am interviewing {{guest}} about {{topic}}. Write me 15 questions ordered for narrative arc — start with their origin, move through the controversial middle, end with their forward-looking view. Include 3 follow-up questions for the ones most likely to need a deeper second pass.
Documentary-style script. Write a 14-minute documentary-style script on {{topic}}. Voice: third-person observational, occasionally first-person when offering my own take. Include four "chapter" beats with on-screen text cues. Make the pacing breathe — short sentences for tension, longer sentences for reflection.
Mid-roll re-engagement insert. My video is on {{topic}}. At the 5-minute mark I need a 25-second re-engagement that resets attention without breaking flow. It cannot say "before we continue." It must reference something the viewer just learned and tease something that comes next.
CTA writer that does not sound like a CTA. Write the final 30 seconds of a video on {{topic}}. The CTA is to {{ask}}. It must feel like a continuation of the video, not a tacked-on outro. Single ask. No "smash that subscribe button." Treat the viewer like an adult.
Category 2 — Hooks & Cold Opens (Prompts 11–20)
The first 8 seconds decide whether the rest exists. These prompts produce hooks that earn the retention curve, not hooks that look good in a screenshot.
Prompt 11 — 20 Hook Variations for One Video
Ready to use
Generate 20 cold-open hooks for a video on {{topic}}, each under 12 words. Mix five hook types: bold contrarian claim, specific number with no context, question that implies a forbidden answer, mid-action sentence (in medias res), pattern interrupt. Mark the three strongest with reasoning in one line each. Skip generic openers — anything that could apply to ten other videos.
My target audience for this video is {{persona}}. Their dominant pain about {{topic}} is {{specific pain}}. Write three cold-open hooks that name the pain so precisely the viewer feels caught. Each under 15 words. The hook must promise the rest of the video will resolve the pain, without saying "in this video I will."
The contrarian hook. Write a 10-second hook that takes the opposite position from the current consensus on {{topic}}. The contrarian view must be defensible, not just edgy. Sentence one is the claim. Sentence two earns it.
The mid-action hook. Write a hook that drops the viewer into the middle of an event already in progress. Topic: {{topic}}. The viewer should feel like they missed the first 30 seconds. The hook then turns to camera and explains.
The "everyone is wrong about" hook. Write 5 variations of a hook starting "Everyone tells you {{X}}. They are wrong, and here is why." Topic: {{topic}}. Each variation must end on a different reason.
The numbered-stakes hook. Write a hook that opens with a specific surprising number related to {{topic}}. The number must be verifiable. Cite the source in a [bracketed] note.
The pattern-interrupt visual hook. Write a hook for a YouTube video on {{topic}} where the first frame is a visual the audience does not expect. Describe the on-screen visual in [brackets] and the line I deliver on camera. They must contradict each other in a way that sets up the next 7 seconds.
The personal-stakes hook. Write a hook for {{topic}} that opens with a single sentence about something at stake in my own life. It must feel real. No manufactured drama. Length: under 15 seconds.
The "I tried X for Y days" hook. Write 5 hook variations using the "I tried {{X}} for {{Y}} days and {{unexpected result}}" template. Make the unexpected result actually unexpected — not "and it changed my life."
The cold open from a viewer comment. I have pasted three comments from my last video. Write a hook that opens by directly engaging with one of those comments — agreeing, disagreeing, or refining. The hook must work for new viewers who have not seen the prior video.
Re-cut hook from an existing transcript. Here is the transcript of a video that underperformed. Find three moments inside it that could have been the cold open and explain why each one is stronger than what I actually used. Suggest a re-cut for next time.
Category 3 — Titles That Win the Click (Prompts 21–30)
The title and thumbnail are the only thing 99% of potential viewers ever see. These prompts are tuned for click-through rate without dishonest clickbait.
Prompt 21 — 25 Titles Across Five Frames
Ready to use
Generate 25 title candidates for a video on {{topic}}. Distribute across five frames, five titles each: (1) curiosity gap, (2) specific number + claim, (3) contrarian / breaks consensus, (4) personal stakes / first-person, (5) practical outcome / how-to. All under 70 characters. Front-load the keyword that matches search intent. Mark the top three by predicted CTR with one-line reasoning.
I have pasted the titles of my last 30 uploads with their CTR. Find the patterns that correlated with high CTR on my specific channel. Write 10 new title candidates for {{topic}} that lean into those patterns while staying honest about the video's actual content. No clickbait that overpromises — list the promises each title makes and confirm I can deliver them.
The bracketed-context title. Generate 10 titles using the "{{Main claim}} ({{specific context or qualifier}})" template. The bracket should add specificity, not redundancy. Topic: {{topic}}.
Listicle title formats. Generate 8 listicle titles for {{topic}}. Mix odd numbers ({{N}} = 7, 9, 11) with even numbers (10, 20). Test "best", "underrated", "actually work" variants. None should use "ultimate" or "complete guide."
Question titles that earn the click. Generate 6 question titles for {{topic}} where the question implies an answer the viewer wants to verify. The question must be answerable in the video. Skip "Is X dead?" — too 2022.
Personal-result titles. Generate 5 titles structured as "I {{did X}} and {{specific outcome}}." Topic: {{topic}}. The outcome must be specific (a number, a date, a named consequence) — not "and you won't believe what happened."
Comparison titles. Generate 6 titles comparing {{A}} vs {{B}} in the context of {{topic}}. Test "vs", "or", "is better than", and "I switched from X to Y" variants.
Title rewriter for retention. I am uploading a video on {{topic}}. The current title is "{{title}}". The current thumbnail is {{description}}. Score the title on three axes: clarity, curiosity, honest delivery. Rewrite if any axis scores below 7.
Series title structure. I am starting a video series on {{theme}}. Design a title structure that brands the series without sounding repetitive. Provide 8 example titles in the structure for hypothetical episodes.
Title from a hook. Here is the cold-open hook I have written. Reverse-engineer 6 titles that match the hook's promise. The title should not give away the hook, but it should not contradict it either.
Search-intent title rewriter. The keyword I am targeting is "{{primary keyword}}". Rewrite my current title "{{title}}" to lead with the search-intent phrase without sounding like SEO sludge. Provide 5 variants.
The thumbnail is half the click decision. Fable 5 writes the on-thumbnail copy; an image model renders the visual. We use both together.
Prompt 31 — On-Thumbnail Text Block
Ready to use
Generate 10 candidate text blocks for the thumbnail of a video on {{topic}}. Each under 5 words. Mix: contrarian claim, big number, single emotionally-charged word, specific named subject, and "vs" comparisons. Mark which work paired with a face on the left vs face on the right. Skip generic copy that does not say what the video is about.
Photorealistic YouTube thumbnail image. Subject: a creator in their late 20s, expressive face, exaggerated reaction of {{emotion}}. Composition: subject on the left two-thirds, large negative space top-right for bold text overlay. Lighting: punchy three-point lighting with a colored rim light in {{accent color}}. Background: softly blurred relevant scene of {{topic}}. High contrast, saturated colors, 16:9 aspect ratio. Style: matches the visual language of {{reference creator}} but distinct.
Thumbnail concept brief. Write a 3-option thumbnail concept brief for a video on {{topic}}. For each option: the visual idea, the text overlay, the emotion the thumbnail communicates, and which audience segment it skews toward.
Thumbnail image prompt — product / tutorial channel. Generate an image prompt for a thumbnail showing {{product / tool}} with my hands or face in shot. Specify lighting, composition, what is in focus, what the text-overlay area looks like, and the color palette. Output is the image prompt only.
Thumbnail image prompt — comparison video. Generate an image prompt for a thumbnail comparing {{A}} vs {{B}}. Split composition: A on the left in a cool color tone, B on the right in a warm color tone. A bold "VS" element down the middle. Specify how the text overlay slots in.
Thumbnail image prompt — story / documentary. Generate a cinematic image prompt for a documentary-style video on {{topic}}. Subject is in shot but the environment carries the mood. Specify lens, lighting time-of-day, atmosphere. The thumbnail should feel like a still from a film.
Thumbnail image prompt — Shorts vertical. Generate a 9:16 vertical thumbnail image prompt for a Shorts video on {{topic}}. Subject fills most of the frame. Bold negative space for top and bottom text. High contrast, saturated, optimized to read at phone size.
Thumbnail A/B brief from a winning thumbnail. Here is the description of my best-performing thumbnail from the last 6 months. Identify what made it work — composition, contrast, emotion, text density. Write a brief for the next thumbnail that applies the same pattern to {{new topic}}.
Category 5 — Shorts & Vertical (Prompts 39–48)
YouTube Shorts is its own format with its own physics. Hooks faster, loops tighter, payoff in 40 seconds. These prompts are tuned for vertical-first.
Prompt 39 — 45-Second Short That Loops
Ready to use
Write a 45-second vertical YouTube Short script on {{topic}}. Structure: 2-second visual hook, 5-second verbal hook stating the payoff, 30 seconds of build-up with two reversals, 8-second resolution. The final line must connect back to the opening visual so the Short loops naturally if it auto-plays. Include on-screen text cues in brackets.
Write a 60-second story-time Short on {{topic}}. First line is a hook under 8 words. The story has one twist and one specific detail that proves it actually happened. End with a one-line takeaway, not a moral. Voice: conversational, slightly breathless, like I am telling it to a friend at the gym.
One-tip Short. Write a 30-second Short delivering exactly one tip about {{topic}}. The hook states the result. The middle shows the move. The end is the audience question. No padding.
Shorts hook bank. Generate 15 hook lines for vertical Shorts on {{topic}}. Each under 8 words. Mix: shocking-fact, question, hot-take, "stop doing X", and "the easiest way to" templates. Mark the three strongest.
Carousel-to-Short converter. Here is the script of an Instagram carousel. Convert it into a 45-second Short. Keep the strongest line as the hook. Cut the slides that do not earn their seconds. Add a visual cue for every beat.
Reaction Short. Write a 30-second reaction Short to {{external content}}. First 3 seconds: my reaction face cue + a single word. Middle 22 seconds: the substantive take. Last 5 seconds: a follow-up question that drives comments.
Educational micro-Short. Teach one concept from {{topic}} in 40 seconds. Use the explain-it-like-I-am-12 frame. Single example. End with a takeaway that connects to a longer video on my channel.
Series Shorts plan. Plan a 7-day Shorts series on {{theme}}. For each day: hook line, payoff in one sentence, the on-screen text overlay. The series should have a connective tissue — a recurring visual or phrase — that rewards viewers who watch all seven.
Repurpose a long-form moment. Here is the transcript of my last 12-minute video. Find the three moments inside it that would work as standalone Shorts. For each: the cut points, the recommended hook line, and the new on-screen text overlay.
Shorts-to-long-form bridge. I just posted a Short on {{topic}}. Write a 90-second long-form video opening that picks up where the Short ended, signalling continuity for viewers who came from the Short while still welcoming new viewers.
The Full YouTube Pipeline With Fable 5
Single prompts are useful. Chains are what actually save creators time. Here is the production pipeline we run end to end.
Idea stage. Run Prompt 22 on your last 30 uploads to identify the patterns that correlate with high CTR on your specific channel.
Hook stage. Use Prompt 11 to generate 20 hooks for your top idea. Pick three.
Title stage. Run Prompt 21 to produce 25 titles. Cross-reference with the hooks you picked. Shortlist five title-hook pairs.
Script stage. Feed the shortlisted hook-title pairs into Prompt 1 to draft full scripts for the top two. Pick the stronger.
Thumbnail stage. Use Prompt 31 for the text overlay and Prompt 32 (or 34, 35, 36) for the image prompt. Generate the image in OpenArt or your tool of choice.
Talking-head delivery. If you are not appearing on camera personally, Hedra can deliver the script in a character avatar — strong for faceless channels.
Shorts repurposing. Run Prompt 47 against the finished long-form to surface three Shorts cuts before the video even launches.
The full pipeline takes ~90 minutes from idea to shoot-ready package. Without Fable 5 in the loop, the same pipeline used to take a working day.
Patterns That Make These Prompts Work
If you read the 48 prompts above carefully, they share the same set of moves. Steal these for your own prompts:
Specific over generic. "Persona = {{specific persona}}" beats "for my audience." Fable 5 rewards specificity in the input.
Bracketed cues are gold. Asking for B-roll suggestions, on-screen text, or scene changes in [brackets] keeps the script directly usable for editing.
Forced constraints kill filler. "Under 12 words", "no generic openers", "skip 'in this video I will'" — these constraints remove the model's default lazy moves.
Past-performance context is the unlock. Almost every prompt above gets better if you paste in three of your top transcripts first. Fable 5 will infer your voice without you describing it.
One ask per prompt. When you want a script, ask for a script. When you want titles, ask for titles. Combined prompts get combined-quality answers.
Reverse-engineering existing wins. Several prompts above start with "here is a thumbnail / hook / title that worked — replicate the pattern." This is the most underrated technique for sustained channel growth.
Pro tip: Save your three highest-performing transcripts as the standing knowledge inside a Claude Project named "Voice." Every script prompt from there forward is one message away.
Where to Run These
Claude.ai web app. Most natural surface for these prompts. Save them as Custom Instructions in a Project per category (Scripts, Hooks, Titles, Thumbnails, Shorts) with your past transcripts in the project knowledge.
Anthropic API. If you are building a creator tool. Use the system prompt for the standing voice instructions and reserve the user message for the topic of the day.
Claude Code or terminal workflow. If you are operating a content pipeline programmatically, you can chain these prompts in a shell script and pipe the output between stages.
Multi-model aggregators like Genspark. Useful when you want to A/B Fable 5 against GPT-5.5 or Gemini 3.5 Flash on the same script — see our full comparison for when each model wins.
For Faceless Channels — The Avatar Layer
If you do not appear on camera personally, Fable 5 still anchors the workflow but you bring in a character delivery layer on top. The stack we recommend for faceless channels:
Script written in Fable 5 using the prompts above.
Voice generated in ElevenLabs with a cloned or branded voice.
Character delivery in Hedra from a single portrait reference.
B-roll generated in your image / video tool of choice.
Fable 5 is the first model where the script feels native to the channel rather than a draft you have to rewrite into your voice. That is the bar for usable creator AI, and Fable 5 is the first model that meets it consistently.
The 48 prompts above are not magic. They are the result of one specific habit — when a default prompt gave us a default script, we asked what constraint would have forced something better, then saved the constraint into the prompt. Run any of these for thirty days and you will have your own list of fifty, tuned to your channel. This is the head start.
All AI Models — model catalogue with current pricing and capabilities.
❓
Frequently Asked Questions
10 questions answered
Mostly yes. The structure is what carries the prompt — Sonnet 4.6 and Opus 4.7 will produce competent output. But Fable 5 was specifically tuned for prose quality, voice retention, and hook-writing, which is exactly the workload these prompts target. On long scripts and creative hooks, the gap is noticeable.
Paste 3-5 transcripts from your best-performing videos at the start of the prompt. Fable 5 will infer your voice from the examples — sentence rhythm, openings, vocabulary, the connective phrases you lean on. This works better than describing your voice in words. Save the transcripts inside a Claude Project to skip the paste step forever.
Yes — and you should. Drop your past transcripts and any channel style notes into the project knowledge, save the prompt as a custom instruction, and every future script generation is one message away. We run one Project per category (Scripts, Hooks, Titles, Thumbnails, Shorts).
If you have never used Fable 5 for YouTube, start with Prompt 22 (Title A/B From Existing Channel). It teaches you what patterns work on your specific channel before you generate anything new. From there, work through Prompt 11 (Hook Variations) and Prompt 1 (Full Script). That is the smallest viable pipeline.
No — Fable 5 outputs text and audio, not images. The thumbnail prompts above (Prompts 32, 34, 35, 36, 37) generate the image prompt which you then feed into an image model. We use OpenArt for this; Midjourney, Flux, and Imagen also work. Fable 5 writes the on-thumbnail text copy itself.
Yes. Anthropic's usage policy permits commercial use of outputs across the standard Claude tiers, including Fable 5. The scripts, titles, and thumbnails you generate are yours to publish on your channel and monetise normally. Standard caveats apply: do not generate misleading or deceptive content, do not impersonate real people without consent, do not copy existing scripts verbatim.
Not for the script being AI-written. YouTube's content guidelines target content that is mass-produced, low-effort, or misleading, not AI assistance in the script-writing process. A creator-led video where the script is AI-drafted and the creator delivers, refines, and adds personal experience is fine. A channel pumping out 30 fully-automated AI videos per week with synthetic narration and stock B-roll is what gets demonetised.
Shorts hooks need to land faster than long-form hooks — under 3 seconds, not 8. The key constraint: the hook must be promising the payoff of the entire 45-second video, not just the next 5 seconds. Prompt 42 (Shorts hook bank) is tuned specifically for this. Generate 15 candidates and shoot the strongest 3 to see which actually retains.
Yes — the “Full YouTube Pipeline With Fable 5” section shows the chain. Each stage takes the output of the previous one as input. The cleanest way to operate it is inside Claude Projects, where each Project holds one stage's knowledge and the user message is the variable input. For full automation, run the chain through the Anthropic API.
The prompts above all work — just lean harder on the educational and tutorial templates (Prompts 3, 4, 5, 45). The voice-retention property of Fable 5 still helps because technical channels often have specific vocabulary, pacing, and a particular relationship to jargon. Paste in your past transcripts and Fable 5 will respect them.