Six weeks ago I shipped a small automation tool that saves me about two hours a week.

It monitors a few data sources, sends me a formatted summary, and flags anything that looks off.

The kind of thing I’d wanted for years and never built because I assumed I couldn’t.

I built it in an afternoon, with an AI coding assistant doing most of the actual work.

I’ve been thinking about that afternoon a lot since.

I should say upfront: I’m not a developer. I can read code well enough to follow what’s happening. I’ve written basic Python. But building something that actually runs reliably, handles errors, connects to external APIs, sends formatted emails that was outside what I thought I could do.

Until it wasn’t.

The AI walked me through the whole thing. When I got confused, I described what I wanted in plain language and it rewrote the relevant section.

When something broke, I pasted the error message and it explained why. The gap between “I want this to exist” and “this exists” collapsed in a way that still feels slightly unreal.

I’ve tested a lot of AI tools over the past three years. Impressive demos and disappointing follow-throughs, in roughly equal measure.

This was different. It wasn’t a demo. I wanted a specific thing, and I got it, and it works.

But here’s what I keep coming back to: do I actually know how to build this?

I watched every line of code get written. I made decisions about what I wanted the tool to do. I tested it, found problems, directed the fixes. In some sense I understand the whole thing. In another sense, if you sat me down today and said “write this from scratch without AI,” I couldn’t.

I don’t know what to call that.

One reading is: the tool is mine, the decisions were mine, the AI was a capable assistant, same as hiring a contractor who speaks better code than I do. That reading feels true some of the time.

Another reading is: I don’t actually know how to do this. I just know how to direct something that does. That also feels true. Sometimes at the same time.

I keep thinking about a question circling through writing and creative communities over the past year what “authorship” even means when AI fills in significant parts of the work.

Most of the people asking it are writers or artists. But I think the same thing is happening in technical work now, just with less ceremony around it.

When the AI fills in the gap, what did it actually fill?

I’ve talked to a few people who went through something similar. A designer I know built her first working web app this way last year.

She called it “the most satisfying and the most confusing thing I did all year.” Those two feelings, she said, were hard to separate from each other.

That matches what I felt. Real satisfaction mixed with something underneath it she couldn’t name.

Not imposter syndrome exactly. Closer to not being sure whether she earned it, which isn’t quite the same thing.

Sam Altman said in early 2026 that he expects AI to compress decades of scientific research into years in certain fields. Dario Amodei wrote last year about a future that functions like “a country of geniuses in a data center.” These are very large claims. I have no idea if they’re right.

What I do know is that six weeks ago the gap between what I could imagine and what I could build got smaller in a concrete way.

Not because I got smarter. Because the tools available to me got better.

I’ve wondered whether this is how all technology works. Word processors didn’t make people better writers, but they removed the physical friction of revision.

Calculators didn’t improve anyone’s grasp of math; they just made arithmetic faster. Maybe this is the same thing: implementation overhead goes down, the decisions and intentions you bring stay yours.

Maybe. I keep finding this one feels different in degree, even if not in kind. A calculator can’t decide how to solve the problem. The AI, at least a little, can.

At one point I asked it to write a section I couldn’t figure out how to frame, giving it only a rough description.

What came back was good. Probably better than what I would have written alone. I used most of it. That part of the tool now exists in a form I wouldn’t have reached by myself.

Was that wrong? I don’t think so. But I notice it.

I don’t have a conclusion here. How I feel about all this genuinely shifts depending on the day.

Some days the version that makes sense is: skill is moving up the stack, the people who figure out how to work at that higher level will be fine, and worrying about whether you “really” know how to do something is just nostalgia wearing a practical disguise. Other days: I’ve started outsourcing things I used to be able to do, and I should pay attention before I notice the capability is gone.

Both feel true. Not at the same time, but on different mornings.

I’ll keep using the tool. It keeps working. I’ll probably build the next thing the same way.

Still figuring out what I think about that.

For more on how AI tools actually perform in practice, I wrote about the bigger picture here: [Our Future With AI And Why I Keep Changing My Mind About It]

Photo by Andrea De Santis on Unsplash

The first time I tried prompting Veo 3.1 with the same kind of description I used for Midjourney, I got something that looked cinematic for about two seconds and then fell apart.

Wrong camera movement.

Flat lighting.

No sense of motion direction. The model was capable of better I just had not told it what to do.

Video models in 2026 are not image generators with a play button attached. Veo 3.1, Kling 3.0, Seedance 2.0, Runway Gen-4.5 these models respond to specific language. Camera movement type. Lighting direction. Emotional tone. Speed cues. Scene framing. A prompt written for a still gets you a blurry, drifting clip. A prompt written for video gets you something that looks like it came from a real production.

This is a reference list of ai video prompts organized by style. Everything below comes from a vault I built testing across all the major 2026 models. Use what fits your project.

The full vault : 1,086 prompts across 8 styles is organized by use case and compatible with Veo 3.1, Kling 3.0, Runway Gen-4.5, Seedance 2.0, Sora 2, and WAN 2.6:

👉 AI Video Prompt Vault -1,086+ Prompts

A selection of prompts across each style is below.

What makes a good AI video prompt different from an image prompt

Image prompts describe a scene. Video prompts describe a scene that moves.

That distinction matters more than most guides acknowledge. The models above are processing four layers of information at once: what the subject looks like, how the camera is moving, how light is behaving, and what physical motion is happening in the frame. If your prompt only covers the first layer, the model fills in the rest at random.

The prompts that work consistently share a structure:

Scene setup. Who or what is in the frame. Location. Time of day.

Camera descriptor. Movement type dolly, handheld, drone, locked, tracking. Not just “cinematic” the specific movement.

Lighting cue. Direction, quality, temperature. Golden hour, rim light, neon spill, overcast diffuse.

Motion direction. What is moving in the frame and how fast. Wind in hair, water flow, a subject walking, debris in the air.

That structure is what separates prompts for ai video from prompts for everything else. Once you understand it, you can adapt any of the prompts below to your own content.

Cinematic and Documentary

These work best for scene-setting shots, establishing frames, and anything that needs to look like it came from a real production rather than an AI generator.

Volcano documentary (slow motion, aerial)

Semi-transparent volcano erupts in slow motion, revealing magma chambers in cutaway | aerial POV, dawn mist, lava sparks illuminate geology labels hovering in AR | crane-up + reverse-zoom (Vertigo effect) while layers expand holographically | golden dawn key-light, smoke catches orange rim, desaturated background to isolate reds | rumbling sub-bass, crackling lava pops

Rain and letter (handheld, dusk)

A young woman reads a handwritten letter on a rainy balcony at dusk, city lights twinkling behind her through streaked glass | cozy European apartment setting, indoor-outdoor contrast, reflections on the window | handheld camera with soft natural head bobs and light reframe zooms | ambient tungsten lighting mixing with cool blue street glow, warm/cool cinematic blend | Foley of rain tapping and gentle traffic below

Field reporter (locked wide, practical light)

Natural light. Field reporter mid-action in an open field, looking directly at the camera, tornado in the background. A reporter, in a muted dark raincoat, stands firmly in a wide grassy field. Wind pulls at his coat and hair. Behind him, a tornado swirls menacingly under an overcast sky. He speaks clearly, lips synced to audio, holding his position with calm authority.

Superhero over city (drone glide, night)

Cinematic, Christopher Nolan style — superheroes flying through New York City | night time, dramatic cinematography | drone glide | shot on ARRI Alexa 65 with Zeiss Supreme Prime 35mm, shallow depth of field, motion blur on cape | city grid 1,000 feet below, ambient light from below illuminating subjects from underneath, deep shadow zones

Fashion Film and Editorial

Fashion film is one of the categories where Veo 3.1 and Kling 3.0 perform best when given proper runway and editorial language. Slow dolly movements and fabric-in-motion are where these models show their range.

Golden hour rooftop (soft dolly, luxury):

Stunning 24-year-old woman, long wavy brunette hair, flawless radiant skin, wearing an elegant silk evening gown, standing on a rooftop lounge with a panoramic city skyline at golden hour | cinematic soft lighting, luxury aesthetic, Vogue-style high fashion photography | slow push-in dolly, subject stationary, wind catching the gown fabric | 8K resolution, ultra-realistic, depth of field pulls from background to subject

Tokyo night editorial (handheld tracking, neon):

A striking Korean model with long sleek jet-black hair and porcelain skin, posing in the heart of a futuristic Tokyo cityscape at night | edgy black leather trench coat over a crystal-embellished bodysuit, knee-high boots | bold red lips and smoky cat-eye makeup | neon signs reflecting off wet pavement | handheld tracking shot following her as she walks toward the lens, slight lens flare from neon above

Beach luxury (golden light, wind motion):

Breathtaking young woman, caramel skin, long jet-black wavy hair, wearing a designer swimsuit and sheer cover-up, standing on a white sand beach with turquoise waters | golden sunset glow, wind-blown hair with natural motion | confident expression, looking just past camera | slow wide tracking shot from left to right, ocean in background out of focus | editorial magazine quality, shot on Leica simulation

Sci-Fi and Futuristic

These are the prompts where synthetic environments and biopunk aesthetics produce the most distinctive output. The camera language here needs to match the scale of the world being described.

Bioluminescent cyborg (static medium shot, practical light):

Ultra-realistic woman who is half-cyborg, seated in a futuristic bar. Her skin is illuminated with intricate patterns of bioluminescent light casting a soft ethereal glow. Human side features striking natural beauty with long flowing hair and expressive eyes. Mechanical side reveals circuit patterns and chrome plating. Neon bar signs reflect in her chrome components. Medium shot, slight camera breathing, subject makes deliberate eye contact with lens.

Rain-soaked cyberpunk assassin (dynamic mid-air):

A deadly assassin with dark smoky eyes wearing a silver sequined dress with thigh-high slit, leaping through a rain-soaked cityscape while brandishing a katana | neon signs reflect off dress and wet pavement, surreal glow | high-speed photography style, dynamic mid-air shot, subject frozen at peak height | blue and magenta tone grade | rain falls in visible streaks behind subject

Cyberpunk singer (close studio, practical neon):

Cyberpunk female singer in a futuristic recording studio, facing camera directly | long messy bright pink hair, large black headphones, metallic futuristic jewelry | microphone in the foreground, music equipment and speakers filling the background | natural-feeling studio lighting, neon accent strips from below | slow zoom toward subject over 6 seconds, ambient electronic hum in audio layer

Nature and Scenic

Nature prompts perform well when they specify the direction of movement and the quality of light. “Beautiful landscape” produces generic output. “Low-angle drone glide at golden hour over fog” produces something specific.

Jungle drone chase (high-speed motion, canopy):

Flying through a dense jungle canopy during a high-speed drone chase, sunlight piercing through the leaves in shafts | vibrant green foliage, rustling leaves from motion, distant wildlife sounds | camera moves fluidly: weaving between trunks low to the ground, then soaring up for a bird’s-eye view | high saturation on greens, fast shutter to freeze individual leaves mid-blur, speed ramp at apex

Coral reef dive (periscope reveal, Maldives):

Periscope Dive Shot into an underwater coral reef scene, midday, Maldives | camera moves as if peeking over a coral outcrop, revealing a colorful bustling marine ecosystem | “periscope” movement introduces the world in stages — from sunlit surface water down to the vibrant coral city teeming with life | crystal-clear water, dappled caustic light on the reef floor, fish scatter as the camera descends | ambient underwater audio

Golden hour beach café (static, long lens):

Woman at a beachside café during sunset | loose linen shirt, sipping an iced latte, gazing directly into the camera with a relaxed smile | golden hour sunlight from camera-left, strong directional warmth | soft waves blur beautifully in the background, long lens compression flattens the distance | camera locked, slight heat shimmer from warm air in the long-lens frame, ambient café sounds and distant surf

AI Influencer Video

Talking-head and product-adjacent content needs specific setup language lighting position, camera height, background treatment, and the subject’s relationship to the lens. These prompts keep the character consistent across multi-shot workflows.

Studio influencer (ring light, direct to camera):

Ultra high-res beauty influencer seated at vanity table, opening skincare box | glossy lips and dewy foundation reflecting soft ring light | camera mounted at eye level, centered, subject speaks directly to lens | high-end bokeh on background, hands mid-motion hold for reaction beat | warm skin tone, sharp facial detail, product clearly visible label-forward in frame

Gaming streamer setup (neon room, authentic):

24-year-old blonde-haired influencer at a custom neon-lit gaming desk, wearing a black cropped hoodie and premium gaming headset | leaning slightly forward with an engaging smile, looking directly into the camera | dual ultra-wide monitors glow in soft blues and purples behind her | Shure SM7B microphone positioned close to her lips | shot on Leica SL2 simulation, shallow depth of field, subtle rim light tracing her hair, slight organic film grain

TikTok selfie (handheld, close-range, movement):

TikTok-style selfie video of a mixed-race influencer laughing mid-spin in an art district alleyway | wearing oversized blazer with bike shorts, Nike Air Force 1s kicking up autumn leaves | iPhone 16 Max Pro night mode look, capturing every detail | motion blur in background lights, sharp focus on subject’s face | warm vsco film filter tones, natural makeup, visible breath in cool air

I have 1,086 of these organized by style in the full vault. If the prompts above are useful, the full pack covers everything:

👉 AI Video Prompt Vault -1,086 Prompts

General Video

These are the versatile prompts that bridge between styles useful for transitions, B-roll, and content that does not fit a single category.

Close-up beauty (ring light, macro movement):

Close-up of a young TikTok creator in selfie mode, speaking directly to camera with natural expression | soft front-facing light, ultra clean skin tone and subtle makeup | slight movement blur from handheld shake, iPhone 16 Pro Max lens simulation | photorealistic, subject slightly animated, blinks naturally mid-take, ambient room sound

Broadcast news (field, movement in background):

A field reporter stands in an open urban plaza at midday, speaking directly to camera | pedestrians and city traffic move behind her in natural out-of-focus motion | bright overcast daylight, even illumination on her face, no harsh shadows | camera locked to a tripod, subject moves slightly while speaking, banner ticker graphic suggested in lower third space

Three things that will improve your output

1. Camera movement is not decoration. Stating “cinematic” without specifying the movement type tells the model nothing useful.

Stating “slow push-in dolly, subject stationary” tells it exactly what to render. Every prompt I have tested that produced consistently good output had a named camera movement.

2. Lighting direction changes the mood more than anything else. Golden hour from camera-left produces something different from overcast diffused light, even at the same location with the same subject.

Be specific about direction, temperature (warm/cool), and quality (hard/diffused). The difference in output quality is immediate.

3. Iterate on camera height and distance first. When output looks “almost right” but not quite there, the fastest fix is usually changing the camera position rather than rewriting the scene description.

Eye-level to low-angle, wide to long-lens compression these variables shift the output significantly with a small prompt edit.

(Every prompt in this article has been tested across Veo 3.1, Kling 3.0, Seedance 2.0, and Runway Gen-4.5.)

There is a Customize button below the generate controls.

Most people ignore it.

It opens a text box labeled “Additional instructions for the hosts.” What you type there runs before the model generates anything.

I have been running the same sources through both approaches for a few months. A well-written prompt under 500 characters routinely doubles the output length and changes the tone from morning show to actual analysis. Five of those prompts are below. The full set of 50 is in the pack.

👉 Get all 50 NotebookLM podcast prompts

How do NotebookLM podcast prompts work?

NotebookLM podcast prompts are copy-paste instructions for the Audio Overview Customize box. The box accepts up to 500 characters and controls how the episode is structured before it generates.

Without any instruction, the default output runs about 12 minutes at a general-audience level regardless of how technical or dense the source is.

A well-constructed prompt changes both the length and the analytical depth, routinely producing 25–35 minute episodes that cover methodology, limitations, and evidence quality rather than just the surface findings.

The 500-character limit is the central constraint: every character has to do work, which is why pre-tested prompts that land under the limit on the first paste are more useful than writing from scratch.

When you click Customize under Audio Overview, a text input opens labeled “Additional instructions for the hosts.” This is the notebooklm podcast system prompt. It runs before the model generates the episode.

With no instruction, you get about 12 minutes of conversation. The hosts cover the main ideas and say things like “that’s so fascinating” and “what I find really interesting here” a lot. They simplify. For anyone who actually read the source, this is mostly useless.

A focused 300-character instruction doubles the length. The tone shifts. The hosts stop sounding like they are recording a morning radio show.

Default vs. with a custom prompt:

Output length: ~12 minutes → 25–35 minutes
Tone: Conversational, enthusiastic → Analytical, direct
Methodology coverage: Rarely mentioned → When explicitly instructed
Filler phrases: Common (“wow,” “fascinating”) → Removable with one instruction

Audience calibration: General public only → Configurable per use case

The character limit is also why having pre-tested prompts matters. Getting something that fits under 500 characters and reliably changes the output takes several attempts. The prompts below are already tested and work on the first paste.

NotebookLM podcast prompts by category

1. For length and depth

Deep Technical Briefing

Generate a deep technical briefing, not a casual overview. Target 25–30 minutes. Assume the listener has expert-level knowledge of the field. Use precise terminology from the source material. Analyze methodology, evidence quality, and implications, not just findings. End with the strongest open question the source raises but does not answer.

2. For audience calibration

Student Exam Prep

Generate an exam preparation briefing. Emphasize key definitions, testable claims, cause-and-effect relationships, and the most commonly misunderstood concepts in the source. Phrase key points in ways that are easy to remember. Flag the 5 most exam-critical ideas.

263 / 500 characters

Instead of a general summary, this produces something built around recall: flagged concepts, cause-effect chains, the ideas most likely to come up in an exam. I use it on lecture transcripts and textbook chapters before a test. It makes the output feel like a study session rather than a podcast.

3. For source type

Research Paper Specialist

This is an academic research paper. Cover in order: research question and why it matters, methodology and sample, main findings with statistical context, limitations acknowledged by the authors, and implications for the field. Use language from the paper. Target 20 minutes.

273 / 500 characters

The instruction that matters most here is “limitations acknowledged by the authors.” The default output almost never covers limitations. It presents findings as if they are settled. This forces coverage of what the authors themselves flagged as uncertain.

4. For format and style

Remove Filler and Enthusiasm

Generate a calm, professional briefing. Remove all filler affirmations: do not say ‘great point,’ ‘absolutely,’ ‘fascinating,’ ‘wow,’ or similar. Speak directly and analytically. The tone should be like a serious podcast, not a daytime talk show. Prioritize information density over warmth.

290 / 500 characters

Consistently the most requested change in the NotebookLM community. The “wow, that’s absolutely fascinating” default gets complained about more than anything else about the feature. This removes it.

5. For specific use cases

Weekly Reading Digest

I have uploaded several articles or papers from my reading this week. Do not summarize each separately. Instead: identify the patterns and connections across them, highlight the most important single idea from the week’s reading, and surface any tensions between sources.

267 / 500 characters

Upload your week’s reading, run this once, and you get a single perspective on the whole pile rather than five separate recaps. The “tensions between sources” instruction is what makes it more than a summary. It finds where sources actually disagree rather than smoothing everything into a false consensus.

Those are five of the 50 in the pack. The rest cover advanced depth modes for dense documents, audience personas from complete beginner to policy analyst, source-specific instructions for books, meeting transcripts, legal documents, strategy papers, format controls for news broadcast and debate styles, and use-case prompts for interview prep, investment research, medical documents, and more.

👉 50 NotebookLM Podcast Prompts Pack

Four things that change what you get

Beyond the prompt itself, a few things affect the output more than people expect.

Source quality shows up in episode quality. A scanned image PDF gives worse output than a clean, text-selectable one. A rough transcript gives you less than a clean Otter.ai or Fireflies export. The messier the source, the thinner the episode.

Stacking depth and audience instructions produces better results than either alone, when the character count allows it. “Target 30 minutes” does something. “Target 30 minutes, assume an expert listener” does more.

I have found that leading with the audience instruction tends to yield more depth than leading with the length target. When you have to pick one, go with the audience.

For any notebooklm long podcast prompt, use a specific number. “Go deep” is vague. “Target 30 minutes or longer” gives the model something concrete to work toward. The difference in output length is consistent.

You can also run different prompts on the same uploaded source across separate sessions. The 500-character limit applies per generation, not per source file. Run Deep Technical Briefing for the expert-level version. Then come back and run Student Exam Prep for something you can review from. The best notebooklm podcast prompts for sources you want to return to are the audience variants. Each one produces a genuinely different listen from the same upload.

Further Reading

NotebookLM Prompts for Notes: 30 Templates for Smarter Summaries and Meeting Recaps

(Amara has tested prompt packs for NotebookLM, Suno, and a dozen other AI tools. She writes about what actually works at aitooldiscovery.com. Every prompt in this article has been run in NotebookLM.)

That is most of what Antigravity skills are, practically speaking. Work you kept re-explaining, turned into a persistent brief.

Install the file once. Every session after that, the tool already knows the format, the rules, and what you care about.

You stop briefing and start working.

Over the past few months, I built 15 of these Skill files for the recurring work that kept eating time: SEO & AEO audits, content briefs, cold email sequences, Zapier automation, Product manager toolkit, Shopify builds, marketing copy, Marketing Psychology. Then decided to share the pack rather than let it sit on a hard drive.

This is every file and what it does. I have also included the specific situation I reach for each one, because for example “SEO audit skill” means nothing without knowing when it actually earns its place.

👉 The pack is here: Antigravity Skills Pack : 15 .md files

What an Antigravity skill actually does

A skill file is a plain-text .md file with a structured brief inside. When you invoke it, that brief loads as the working context for the session. The tool stops starting from zero.

The practical difference from a saved prompt: a prompt is something you paste at the start of a conversation. A skill file is something the tool reads from disk before you type anything. For example, you type /seo-audit and the output format, the scoring criteria, the no-fabrication rule, and the section structure are already in context. You skip straight to the work.

The SEO Blog Writer skill shows what this looks like in practice. Every post it generates auto-includes four elements: a TL;DR block, a one-sentence definition of the topic, a comparison table, and a five-question FAQ. Not because they are in my brief because they are in the skill file. Those four elements are what AI search results (Perplexity, Bing Copilot, Google AI Overviews) pull from for passage-level citations. I noticed the difference in citability within two weeks of switching. They appear in every post now without me thinking about it.

That is the real payoff. The behavior you want becomes the default.

The 15 Skill files

Five areas, 15 Skills. For each one I have included the specific work I reach for it on.

SEO

01 : SEO & AEO Blog Writer

Writes long-form blog posts structured for both Google rankings and AI citation. Every post auto-includes four things: a TL;DR block, a one-sentence topic definition, a comparison table, and a five-question FAQ. Built in, not requested. Those four elements are what AI search results (chatGPT, Perplexity, Copilot, Google AI Overviews) pull from for passage-level citations. I noticed the citability difference within two weeks of switching.

02 : SEO Audit

Diagnoses crawlability, indexation, and ranking issues across a site. Give it a URL or a list of pages and it returns a prioritized fix list sorted by impact, not by type. A standard audit groups everything by category regardless of urgency. This one tells you what to fix first. I reach for it when a page is stuck and I want fast triage before deciding where to spend time.

03 : SEO Content Analyser

Full E-E-A-T content quality audit with AI citation readiness check. It tells you what a page is missing for trust signals, topical authority, and passage-level citability in AI search results. Most useful for pages sitting on page two that are not moving despite decent backlinks — there is almost always a specific gap this surfaces that a standard audit misses.

04 : Programmatic SEO

Designs pSEO strategies for building ranking pages at scale. Give it a niche and a data source and it returns a template structure, URL schema, and variable logic. It covers the data layer decisions most pSEO plans skip which is usually the part that collapses three months into execution.

Copywriting and email

05 : Copywriting

Writes conversion-focused copy for landing pages and emails. The skill confirms the brief before writing, which sounds minor but saves a lot of back and forth. A no-fabrication rule is built in — nothing in the output is invented. I use this for fresh landing page copy and for rewriting existing pages that have traffic but are not converting. The no-hallucination constraint is what makes the output usable without fact-checking every sentence.

06 : Copywriting Psychologist

Applies behavioral psychology triggers to copy you have already written: scarcity, social proof, authority, loss aversion, reframing. In audit mode it identifies which triggers are missing and either rewrites those sections or gives you a revision brief. I run this on every landing page before it goes live. It almost always finds one thing I missed — and it is usually the reason the page is not converting.

07 : Cold Email

Writes B2B cold email sequences and multi-touch follow-up series. Personalized opening lines, tested subject line formats, CTAs calibrated for replies rather than opens. The no-fabrication rule matters more here than anywhere else — one invented personalization detail kills a cold email faster than a bad subject line. Everything in the output comes from what you gave it. Nothing invented.

08 : Email Sequence

Builds nurture sequences that move subscribers from awareness to conversion. Handles stage logic, timing, and tone across welcome series, re-engagement, and post-purchase flows. You give it the audience, goal, and product context. It returns a sequenced series with the rationale for each email’s position in the flow — which makes it easy to adjust without breaking the logic.

Strategy and analytics

09 : Content Strategy

Plans topic clusters, editorial roadmaps, and content mix for traffic, authority, and lead generation. I use this when auditing a content calendar that is producing traffic but no leads. It finds the missing conversion layer quickly — usually within the first output. Give it your goals and current content mix and it returns a prioritized publishing plan with the reasoning for each decision.

10 : Marketing Psychology

Applies behavioral science and mental models to marketing decisions. Scores each recommendation by psychological leverage and implementation difficulty, so you know what to act on first rather than working through a flat list. More useful than a general marketing brief because it explains the mechanism, not just the tactic.

11 : Analytics Tracking

Designs, audits, and improves analytics tracking systems. Covers event taxonomy, tag architecture, and data layer logic. The output is a tracking plan that produces decision-ready data. Useful for new setups and for diagnosing existing ones where the numbers are technically there but feel unreliable which is most of the ones I have looked at.

Technical and e-commerce

12 : Shopify Development

Builds Shopify apps, extensions, and themes using GraphQL Admin API, Shopify CLI, Polaris UI, and Liquid. Covers new builds and targeted modifications to existing stores. For developers or technical founders working across multiple stores on the same stack the persistent context from the skill file is what stops you re-explaining the architecture every session.

13 : WordPress & WooCommerce

Full store setup, payment integration, shipping configuration, and customization. Includes WordPress 7.0 AI connectors, DataViews, and collaboration tools. Handles new builds and existing store improvements. If you manage multiple client stores on the same stack, the persistent context makes this worth installing from day one.

14 : Zapier & Make Patterns

Designs no-code automation workflows with multi-step logic, error handling, and conditional routing. I use this for client onboarding workflows and content pipelines. The error handling it builds in by default has saved hours I would have spent debugging months later. Something always eventually breaks. This skill builds around that assumption from the start.

Growth and operations

15 : Product Manager Toolkit

Essential PM frameworks from discovery to delivery: PRDs, prioritization scoring, sprint planning, and stakeholder communication templates. Useful if you do product work without a dedicated PM, or if you are a PM who wants to hand off the document scaffolding and stay on the actual decisions. I use the PRD and prioritization scoring sections most — they are the parts that always take longer than they should without a structure to pull from.

👉 Get all 15 skills

Where the files work

The same .md file installs across six tools. Only the folder path changes.

| Platform | Install path |

| Antigravity | .agent/skills/ |

| Claude Code | ~/.claude/commands/ |

| Cursor | .cursor/commands/ |

| Codex CLI | ~/.codex/skills/ |

| Gemini CLI | ~/.gemini/skills/ |

| Windsurf | .windsurf/commands/ |

For a global install available across every project, not just one folder, copy the files to the platform’s home directory path instead. The README in the pack has the exact path for each tool.

How you invoke a skill:

- Claude Code, Cursor, Windsurf: type /skill-name

- Antigravity and Gemini CLI: mention the skill name in chat

- Codex CLI: type $skill-name

Install in 2 minutes

1. Download and unzip the pack

2. Copy the .md files from the skills/ folder into your tool’s commands folder (path in the table above)

3. Open your AI tool and type the slash command or mention the skill name

No configuration. No API keys. No code. The files are plain text -open any in a text editor to read or edit the instructions directly.

Further reading

Three other articles worth reading if you are in this space:

Viggle AI Prompts: Complete List for Dance, Lipsync and Character Animation — if you do character animation or short-form video content.

NotebookLM Prompts for Notes: 30 Templates for Smarter Summaries and Meeting Recaps — if your work involves documents, meeting transcripts, or research. NotebookLM pairs well alongside Claude Code for any workflow that mixes document analysis with task execution.

Suno AI Lyric Prompts: Complete Template Guide With [Section] Tags and Vocal Cues — if you use Suno and want proper lyric structure, not just style prompts

The first time I tried to get Suno AI to follow my lyrics properly, I pasted in a verse and a chorus with no tags and no structure.

Suno generated something, but it ignored half my lines, repeated sections I had not written, and arranged the song in a way that made no sense.

The problem was not my lyrics.

The problem was that I had not told Suno how to read them.

Suno has two input boxes: the Style box and the Lyrics box. Almost every Suno prompt guide covers the Style box (BPM, genre, instruments, mood). That side of prompting is well covered. The Lyrics box is where the gap is. Without the right formatting, Suno treats your text as raw material and rearranges it however it wants.

The full pack of 50 Suno lyric prompt templates-each with [Section] tags, vocal cues, and a matched style prompt is here:

👉 50 Suno AI Lyric Prompts

And the Five complete templates are given below.

How Suno actually reads your lyrics

The Lyrics box is not a text field. It is a structure field. What you put in controls how Suno assigns melody, arrangement, and song sections.

Two things drive the output quality. The first is section tags, which tell Suno where each part of the song starts and ends. The second is vocal cues, which go inside the lyrics in parentheses and guide how Suno delivers each line.

Without section tags, Suno reads your text as one block. The melody loops without logic. Verse and chorus carry the same energy. Arrangement becomes unpredictable. With tags, Suno knows where the verse stops, where the chorus begins, and how much lift to give each section.

Line count matters too. Suno gives the most melodic weight to the first line of each tagged section. Choruses work best at 2 to 4 lines. Anything longer spreads the hook too thin and the melody flattens. Verses hold up to 8 lines. Put your strongest line first.

The full [Section] tag list

These go at the start of each new section, on their own line, in square brackets. No punctuation after them:

[Verse 1]

[Verse 2]

[Pre-Chorus]

[Chorus]

[Bridge]

[Outro]

[Intro]

[Hook]

[Break]

[Interlude]

[Post-Chorus]

[Refrain]

You do not need all of them. For most songs, Verse, Pre-Chorus, Chorus, Verse, Bridge, Outro covers everything.

Vocal cues

These go inside the lyrics on their own line in parentheses, directly before the lines they apply to:

(whispered)

(belted)

(falsetto)

(harmonized)

(ad-lib)

(spoken word)

(spoken, soft)

(building intensity)

(stripped back)

(key change)

(half-time feel)

(half-spoken)

Two or three per song is enough. The bridge is usually where the shift matters most. A (whispered) bridge after a big chorus creates contrast without touching the Style box.

5 Suno lyric prompt templates

One per major theme. Each template has two parts: the Style Prompt for Suno’s Style box and the Lyrics for Suno’s Lyrics box. Copy both, paste each into the correct box, generate. (use your own lyrics as needed, these are examples)

Template 1 : Quiet Regret

Theme: Heartbreak | Genre: Pop Ballad | Mood: Melancholic

Style Prompt -paste into Suno’s Style box:

Piano-driven pop ballad, raw intimate female vocals, subtle string pads, builds from sparse verse to soaring chorus, reverb on vocals in chorus, stripped acoustic bridge, BPM: 72, Key: G Minor

Lyrics -paste into Suno’s Lyrics box:

[Verse 1]

I still fold your sweater, leave it on the chair

Still brew two cups before I catch myself

The calendar says it’s been a year now

But some mornings I forget

[Pre-Chorus]

(building intensity)

I reach for the phone and stop

I know how this ends but something keeps me holding on

[Chorus]

I’m still learning how to miss you quietly

Not in the loud way, just the small things

Just the ordinary things

I’m still learning how to let you go

[Verse 2]

I drove past the diner on a Tuesday

The booth we had is someone else’s now

The strangest part is life kept moving

Even when I swore it couldn’t

[Bridge]

(whispered)

Maybe some goodbyes take longer than the love did

Maybe that’s just how this works

Maybe I’ll be fine

Maybe I already am

[Outro]

(stripped back, piano only)

Still learning

Still learning

Still here

Template 2 : New Love Butterflies

Theme: Love | Genre: Upbeat Pop | Mood: Playful / Electric

Style Prompt -paste into Suno’s Style box:

Bright upbeat pop, playful female vocals with ad-libs, pulsing synth layers, danceable groove, euphoric pre-chorus build, radio-ready production, BPM: 122, Key: C Major

Lyrics -paste into Suno’s Lyrics box:

[Verse 1]

I’ve rewritten this text about eight times

Deleted it and typed it back the same

You replied in under thirty seconds

Which is absolutely destroying me

[Pre-Chorus]

(building intensity)

And I don’t know what this is yet

But my heart does it’s been doing the math

[Chorus]

I think I like you more than I planned to

More than the version of me that plays it cool

I think I like you in the embarrassing way

The can’t-sleep, can’t-eat, can’t-think-straight way

Yeah I think I like you

[Verse 2]

My friends have noticed I’m ridiculous

I keep smiling at my phone for nothing

You said my name like it was casual

It wasn’t casual nothing about this is

[Bridge]

(harmonized, bright)

Don’t tell me to slow down

I’ve been slow for years

This is the first time running felt like the right direction

[Outro]

(playful)

I think I like you

I think I like you

Yeah — yeah I do

Template 3 : Late Night Hustle

Theme: Motivation | Genre: Boom Bap Hip-Hop | Mood: Determined

Style Prompt -paste into Suno’s Style box:

Classic boom bap hip-hop, confident storytelling rap delivery, jazzy piano sample, vinyl crackle, punchy snare, deep 808 bass, BPM: 88, Key: Eb Major

Lyrics -paste into Suno’s Lyrics box:

[Verse 1]

Midnight again and the city’s finally quiet

I don’t chase sleep when I still got work to do

Five years of nos stacked up like a blueprint

Every rejection taught me what wall to go through

[Pre-Chorus]

(half-time feel)

They said wait your turn

I said after what wait, exactly

[Hook]

This is the part where I stop explaining

This is the part where the work starts speaking

Nobody clocks the nights you gave up everything

They only see the morning that it changes

[Verse 2]

Third cup notebook half-destroyed

Twelve tabs open and one direction

You can’t negotiate with what you’re built for

I tried to walk away, it walked me right back

[Bridge]

(spoken word)

This version of me took years to make.

I’m not walking away from it for anyone.

All those people who left -they left before the chapter they were supposed to witness.

That’s not my fault. That’s their timing problem.

[Outro]

(confident, controlled)

Watch this part

This is the part that pays off

Watch this part

Template 4 : Childhood Home

Theme: Nostalgia | Genre: Acoustic Folk | Mood: Warm / Tender

Style Prompt -paste into Suno’s Style box:

Acoustic folk, warm storytelling male vocals, fingerpicked guitar, gentle intimate atmosphere, unhurried and honest, BPM: 74, Key: C Major

Lyrics -paste into Suno’s Lyrics box:

[Verse 1]

I drove past it without meaning to

Turns out memory lives in the body, not the mind

The grass is different -somebody repainted the shutters

But the front step sags the same

[Pre-Chorus]

(soft, half-spoken)

You can’t go back-I know that

But something in my chest didn’t get the message

[Chorus]

I grew up in that yard

I learned what safe felt like in that room

I didn’t know I was collecting it back then

All that ordinary warmth I took for granted

I grew up in that yard and I miss it like a person

[Verse 2]

The kitchen window used to catch the afternoon

I did my homework in that exact light

Someone else does their homework there now

I hope they know it -I hope they feel it

[Bridge]

(whispered)

My mother called it home for thirty years

The word felt small for what she meant

Now I understand the size of it

Now I understand the exact size

[Outro]

I grew up there

Right there

I grew up there

Template 5 : Carefree Weekend

Theme: Party | Genre: Dance Pop | Mood: Euphoric / Free

Style Prompt -paste into Suno’s Style box:

Euphoric dance pop, bright female vocals with harmonies, pulsing synth bass, driving beat, radio-ready summer production, BPM: 126, Key: C Major

Lyrics -paste into Suno’s Lyrics box:

[Verse 1]

Phones off -nobody needs to know where we are

This is the kind of night we don’t document

Just live inside it until it’s over

And then talk about it for the next ten years

[Pre-Chorus]

(building, grinning)

Didn’t plan it that’s why it works

Every good night starts exactly like this

[Chorus]

This is it this is the night that gets remembered

This is us when nobody’s asking anything

No agenda just the music and the moment

And the version of ourselves that only shows up here

This is the night this is the one

[Verse 2]

My friend looks over and we don’t have to say it

The knowing thing — the same time same thing

Years of friendship looks like that sometimes

A look that means a whole conversation

[Bridge]

(euphoric)

Keep going — just keep going

This is the only hour that exists right now

[Outro]

(fading out, joyful)

This night

This exact night

This one

Those are five of the 50 templates in the pack. The rest go deeper into each theme. Heartbreak has templates like Deleting Someone and Finding Their Things. Love covers Long-Distance Love, Late Night Intimacy, and Growing Old Together. Motivation includes First Generation Pride and Proving Them Wrong. Each one is built the same way as the five above, so once you have used any of these, the rest work immediately.

👉 50 Suno AI Lyric Prompts

Four things that will improve your output

Match the lyric mood to the style prompt mood. A melancholy lyric with an upbeat style prompt produces something incoherent. Suno reads both boxes at the same time. If they conflict, the output tries to satisfy both and fails at both. Check the mood of your style prompt before you hit generate.

Keep choruses short. I have tested this enough to say it clearly: anything over four lines loses melodic focus. Suno spreads the hook across too many notes and it flattens out. Two or three lines is where the hook actually stays with you.

Put your strongest line first in every section. Suno gives the most melodic weight to the opening line of each tagged section. If your best lyric is buried in line three, it will not land. Move it up before you generate.

Run it more than once. The same template produces different results on each run. The third or fourth generation is usually noticeably better than the first. Keep the strongest one and move on.

(I write about AI tools, prompts, and practical workflows at aitooldiscovery.com. Every template in this article has been tested in Suno AI.)

The first time I used NotebookLM for meeting notes, I typed something like: “summarize this for me.” It did.

Four generic bullet points I could have written myself in two minutes.

The problem was not NotebookLM. The problem was I was treating it like a search bar. It is a structured research assistant.

It follows instructions precisely, pulls only from what you have uploaded, and shapes the output to match exactly what you ask for. Once I started writing prompts with real structure in them, the results got useful.

This is the list of notebooklm prompts for notes I use most. Eight are in full below. The remaining 17 are in the complete pack. Use what fits.

Get all 100+ copy-paste prompts organized by use case : notes, studying, slides, and infographics.

👉 Get the NotebookLM Prompt Pack

What NotebookLM note prompts actually do

NotebookLM prompts for notes are different from asking a regular chatbot to summarize something. The key difference: it works only from what you have uploaded. No internet. No training data bleed. Every output comes from your specific sources, whether that is a meeting transcript, a research paper, a book chapter, or your own notes.

This means you are not getting a generic summary. You are getting a structured output built from your exact material. The prompt is just the instruction for how to organize it.

The prompts below tell NotebookLM what format to use, what sections to include, and how long each one should be. You paste the prompt into the notebook chat box after uploading your source. That is it.

NotebookLM note prompts by use case

For meetings and documents

These three prompts work on any uploaded meeting transcript, call recording transcript, briefing document, or internal report.

Meeting Summary Generator

Summarize my uploaded meeting transcript or notes into a structured format: (1) Decisions made — numbered list. (2) Action items — who does what by when. (3) Open questions — unresolved items needing follow-up. (4) Next steps — concrete next actions. Keep the total to one page.

The one I use most. The four-part structure gets everything a team needs from a meeting recap and nothing they do not.

Action Items Extractor

Scan my sources for every task, recommendation, or action item mentioned — explicitly or implicitly. For each item: (1) What needs to be done, (2) Who is responsible (if mentioned), (3) Deadline or urgency (if mentioned), (4) Which source it came from.

The “explicitly or implicitly” part is what matters. A document that says “we should probably look into this” gets captured, not just the items marked action required.

Executive Brief

Write a 1-page executive brief from my sources for a senior decision-maker who will not read the full material. Use this structure: Situation (1 sentence) → Background (2 sentences) → Key Findings (4 bullet points) → Recommended Action (1–2 sentences) → Key Risk (1 sentence). No jargon.

Good for dense reports when you need someone senior to understand the core without reading the whole thing.

For reading and research notes

These prompts work on books, academic papers, articles, or any material you have uploaded for learning or research.

Chapter-by-Chapter Summary

Summarize each chapter or major section of my source separately. For each: section title → main argument in one sentence → 3 key points → one standout quote with page reference. End with a cross-section synthesis paragraph showing how all sections connect.

The synthesis paragraph at the end is the part worth keeping. It shows you the through-line you would probably miss just scanning the individual sections.

TL;DR Generator

Write a TL;DR for each uploaded source. Format: [Source name] — [2-sentence summary of what it argues and why it matters]. Then write one master TL;DR (3 sentences maximum) that summarizes the combined insight across all uploaded sources.

Useful when you have uploaded three or four sources and want orientation before reading any of them properly.

Source Conflict Detector

Identify where my sources directly disagree or contradict each other. For each conflict: quote the conflicting claims from each source, explain the nature of the disagreement, and evaluate which claim is better supported by evidence.

I would not have thought to use this one on my own. If you are working with multiple sources on the same topic, it finds where they actually contradict each other rather than smoothing everything into a false consensus.

For daily note-taking workflows

Daily Notes Digest

I have uploaded today’s notes or reading. Summarize them into: (1) 3 things I learned or encountered, (2) 2 connections to concepts in my other uploaded sources, (3) 1 question worth exploring further, (4) 1 concrete action or next step I should take.

Upload whatever you read or wrote that day and run this before you close the tab. Item two, the connections to other uploaded sources, is the part that surprises me most consistently.

One-Pager Summary

Compress my sources into a one-page summary using this exact layout: Title → 3-sentence overview → 5 bullet key points (max 15 words each) → 1 key statistic or quote → 1 conclusion or call to action. The entire output should fit on one A4 page when pasted into a document.

The word limits and structure constraints are why this works. Without them, NotebookLM sprawls.

Those are 8 of the 25 notes templates in the complete pack. The full set adds book notes, literature review outlines, interview note formatters, research memos, decision logs, quote extractors, recurring themes extractors, and 10 more formats.

👉 100+ NotebookLM Prompt Pack : notes, studying, slides, and infographics

Four things that will improve your note outputs

The prompt is the instruction. Your uploaded source is the raw material. Both matter.

Upload clean sources. NotebookLM does much better with PDFs that have selectable text than with scanned images or rough audio transcripts. A Fireflies or Otter.ai transcript gives cleaner output than an audio file directly. The format of your source shows up in the format of your output.

Be specific about format in your prompts.

The ones in this article work because they spell out exactly what they want: numbered sections, word limits, required headings.

When you write your own prompts, describe the output format at the same level of detail.

This is the same principle that makes prompts work across completely different AI tools : I apply the same structural specificity when building prompt lists for video tools like Viggle AI, and also for the podcast prompts in notebookLM where the output format is just as important as the instruction itself.

Pick two or three prompts and use the same ones for the same source types. A month in, all your meeting notes look the same, all your book notes look the same, and they are much easier to navigate. Consistency compounds faster than you expect.

The best first use is usually your existing notes. Upload a folder from a project you already finished and run the Recurring Themes Extractor or Synthesis Summary on it. You will find patterns across months of material that you thought you had already fully processed.

(I write about AI tools, prompts, and practical workflows at aitooldiscovery.com. Every prompt in this article has been tested in NotebookLM.)

I was testing a new AI tool about six months ago.

I do this most weeks.

I asked it something specific, something niche, something I was fairly confident it would get wrong.

It got it right. I sat there for a moment and thought: okay. So that line moved again.

It wasn’t dramatic. I didn’t feel awe. I felt something quieter and harder to name. Like the ground had shifted a fraction of an inch, just enough to notice.

That feeling comes back to me a lot lately when I read what the people building these systems actually think is coming.

What the optimists say

Dario Amodei, the CEO of Anthropic, published an essay in late 2024 called machines of loving grace. It is worth reading in full if you haven’t.

His core image is this: imagine having “a country of geniuses in a data center.” AI systems smarter than a Nobel Prize winner across most fields, compressing decades of biological research into a few years, potentially solving diseases that have resisted science for generations.

He writes it not as prediction but as genuine possibility, which is its own kind of statement from someone in his position.

Sam Altman, the CEO of OpenAI, said in early 2026 that His timeline: small scientific discoveries happening this year, robot deployment at meaningful scale by 2027, superintelligence within a few years after that.

Demis Hassabis at Google DeepMind has the first automated AI research lab in 2026 and AGI exceeding human-level performance by 2035.

I find myself reading all three and taking them seriously. Not uncritically. There is an obvious thing worth holding in mind: every one of these people is building the thing they are predicting.

That doesn’t make them wrong. But it does add a layer of context to the confidence.

What the critics say

Yuval Noah Harari spoke at Davos earlier this year and said something that has stayed with me.

he said “It is an agent. It can learn and change by itself.” His concern isn’t robots or job losses, at least not primarily. It’s language. AI’s ability to generate and shape language gives it power over human communication and institutions in ways that accumulate quietly, before most people notice what has changed. He framed AI as something the existing system was never built to accommodate, the way a city built around cars was not built for the volume of traffic that eventually arrived.

Max Tegmark at MIT has written and spoken consistently about the speed problem. Not that AI is inherently dangerous, but that the pace of development is running well ahead of any
governance structure capable of managing it. The two are moving on very different timescales, and that gap is widening. You can read his thinking on this alongside Harari’s.

I find myself nodding at both camps. That’s uncomfortable. I’d rather have a clean position.

Most technologies eventually get boring. They become infrastructure. The internet was going to change everything, and it did, and now it’s mostly just there, humming in the background while people argue about other things. Maybe AI follows the same arc. Or maybe it doesn’t, because it’s different in kind and not just degree. I genuinely don’t knon which of those is true.

What I actually notice

I’ve been writing about AI tools for long enough that my early notes are sometimes already outdated by the time I look back at them. Something I wrote about a specific tool three months ago has occasionally already been superseded by a new version or a new competitor that didn’t exist yet. The pace isn’t slowing.

Some tools I use every day now without thinking about them. That’s different from two years ago, when each one felt like a deliberate choice. The deliberateness has faded for some of them. They’ve become part of how I work the way a particular word processor or a specific notebook format becomes part of how you think.

The thing I find nobody talks about much: using AI daily makes you more aware of what it can’t do, not less. You stop being impressed by the easy stuff. You start noticing the gaps.

The places where the answer is confident and wrong. The places where something that should be simple comes out sideways. Living with these tools changes what you expect from them, and not always upward.

What I keep wondering is whether there will be a version of this that feels stable. Or whether the whole point is that it never stops moving, that every six months the ground shifts a little and you recalibrate again. I don’t have an answer to that.

The question I keep coming back to Everyone writes the piece about jobs. I’m not going to write that piece here.

The question I find myself sitting with is smaller and more personal. I write about AI tools. I use AI to help with parts of my work. At what point does the thing I publish stop being mine in the way it used to be? Not legally. Not in terms of effort. But in some other sense that I don’t have clean language for yet.

Dario Amodei wrote a second essay this year, that touches on some of this. It’s less discussed than the first one, worth looking at. It doesn’t resolve the question. Nothing does right now.

I think the people who are most honest about this question, who sit with it rather than reaching for a clean answer, are going to be the ones who figure out something real about what creative and intellectual work means going forward. I don’t know what that looks like yet. I’m trying to stay honest about not knowing.

I’ll keep changing my mind about this. I think that’s the honest thing to do.

(Amara writes about AI tools and workflows at https://aitooldiscovery.com.)

Stories you may be interested in :

Notebooklm podcast prompts guide

AI Video prompts guide

Viggle prompts are not like Midjourney or Sora prompts.

You are not describing a scene or a visual style.

You are describing a character and separately pairing that character with a motion. The second I understood that distinction, my results went from broken to actually good.

This is the full list I have built from testing. It covers character descriptions you can paste directly into Viggle’s text field, the built-in motion presets worth knowing about, and combo templates that pair the two together for specific content goals.

Use what applies to your content and ignore the rest. (Use Ctrl-F for search)

What Viggle prompts actually are

Viggle animates a still image using motion from a reference video or a built-in preset. There are two inputs you control:

1. A character : either an uploaded image or a text description of who you want animated

2. A motion : either a video you upload or a preset name you type

The prompts in this article are for the character text field. They describe appearance: outfit, hair, expression, background. The motion is separate. That is why most free Viggle prompt lists you find online are useless, they describe full scenes instead of isolated character appearances, and Viggle does not respond to that.

The three modes you should know:

Mix : puts your character into an existing motion video. Best for meme content and dance clips.

Move : applies motion to a still photo of a real person. Best if you are animating yourself or a client.

Ideate : text-to-animation with no image needed. Describe both the character and the motion in one field. Useful for quick tests.

👉 I have 160 character prompts in total across 8 categories. If you want the full list in a PDF you can actually search through, I organized everything into a pack here: 200+ Viggle AI Prompt Pack

It also includes a full motion preset pairing guide and 30 combo templates.

Or start with a free sample: Free 30-prompt starter pack

Character description prompts by category

These go directly into Viggle’s character description field. Use a clean white or solid-color background in any reference image you pair with them for the cleanest output.

Anime and manga

1. Anime girl in neon cyberpunk outfit, short pink hair, glowing blue eyes, white background

2. Cute chibi anime character, big eyes, cat ears, pink dress, transparent background

3. Waifu anime girl with long silver hair, school uniform, blushing expression

4. Kawaii anime fox girl, fluffy tail, maid outfit, smiling, white background

5. Cyber anime hacker girl, round goggles, oversized hoodie, neon forearm tattoos

6. Android girl anime, visible mechanical joints, chrome finish, LED strip eyes

7. Dark fantasy anime demon girl, horns, red eyes, black leather armor, white bg

8. Idol anime girl, stage costume, sparkles, microphone in hand, beaming smile

9. Magical girl anime with wand, frilly pastel dress, starry wide eyes

10. Samurai anime warrior, topknot hair, traditional gi, serious expression

11. Shonen hero in orange gi, blonde spiky hair, muscular build, studio lighting

12. Anime witch, wide-brim hat, robe, spell book, glowing hand

Realistic humans

13. Young athletic man, short black hair, fitted white t-shirt, neutral gray background

14. Fitness model woman, high ponytail, sports bra and leggings, clean gym background

15. Plus-size woman in bold floral dress, natural hair, radiant smile

16. Tech startup founder, signature hoodie and jeans, relaxed but confident lean

17. Female CEO in structured charcoal blazer, sleek high ponytail, confident boardroom stance

18. Senior lawyer man, navy suit, briefcase, composed professional expression

19. Ballet dancer woman, bun, leotard, graceful neutral pose, white background

20. Young child boy, gap-tooth smile, cartoon t-shirt, arms wide open

21. South Asian woman, traditional salwar kameez, bangles, dignified expression

22. Muscular boxer man, shaved head, tank top, intense focused gaze

Cartoon and illustrated characters

23. Looney Tunes style rabbit, massive ears, buckteeth grin, white background

24. Pixar-style round robot, shiny chrome panels, expressive LED eyes

25. Bug-eyed cartoon monster, purple fur, goofy lolling tongue, rainbow background

26. Emoji face character, bright yellow circle, big eyes, simple expression

27. Classic superhero in primary colors, cape, strong chin, hands on hips

28. Rubber hose animation style 1930s character, pie-cut eyes, bendy limbs

29. Large purple round mascot, happy bounce-ready pose, white bg

30. Funko Pop style figure, big block head, tiny body, painted eyes

Meme characters

31. Shiba Inu in Doge meme style, wow expression, colorful comic sans text, white bg

32. Sad frog, green, beige hoodie, drooping eyes, slumped posture

33. Bald everyman face, simple features, neutral expression, gray background

34. Sunglasses wolf in leather jacket, cool composure, sigma alpha posture

35. Dog sitting calmly surrounded by flames, coffee mug in paw, relaxed expression

36. Surprised Pikachu face expression, mouth open wide, eyes wide, yellow round

37. Blank-faced NPC in corporate polo, vacant stare, stiff arms, fluorescent lighting

38. Gigachad ultra-masculine realistic face, sharp jaw, composed stare

Fantasy and sci-fi

39. High elf archer woman, long pointed ears, green ranger cloak, bow drawn

40. Vampire aristocrat, black cape, pale gaunt face, red glowing eyes, fangs

41. Cybernetic bounty hunter, half-metal face revealing circuits, hooded trench coat

42. Steampunk inventor girl, large brass goggles, corset, gears and coils attached

43. Orc warlord, large tusks, war paint, double-headed axe, muscular green skin

44. Space marine in powered exo-armor, full helmet visor, laser rifle at ready

Built-in motion presets worth knowing

Viggle has a library of built-in presets you can use without uploading a reference video. Type the preset name into the motion field or select it from the dropdown. These are the ones that consistently produce clean output:

Dance presets

- $hip_hop_dance : works with athletic builds and streetwear characters

- $viral_hugo_dance : best for cute anime characters, very high TikTok performance

- $get_griddy : sports characters and meme characters respond well to this

- $robot_dance_walk : good for android characters, NPC meme characters

- $smooth_criminal : anime villain types, suave characters

$dazzling_dance_entrance : princess, idol, fantasy characters

Meme and viral presets

- $rickroll_never_gonna_give_you_up : works with almost any character, meme energy

- $severance_dance_party_meme : professional or office characters specifically

- $peter_parker_evil_dance : villain characters and anti-heroes

- $fortnite_dance_fixes_awkwardness : NPC and meme characters

Sports and action presets

- $high_rise_parkour : athletic, sci-fi, or hero characters

- $get_griddy : sports celebration, athlete characters

- $bruce_lee_nunchaku : martial arts or warrior character types

K-pop and idol presets

- $like_jennie_live : idol anime, cool female characters

- $jhope_mona_lisa : K-pop style characters

- $le_sserafim_hot : sleek idol characters

The full preset reference table : 30 presets with the character type and platform each works best for is in the paid pack. The table format makes it much faster to find the right preset for whatever character you are working with.

Combo templates (5 of 30)

A combo is a character prompt plus a motion pairing plus tips for the output. These are the ones I go back to most often.

Kawaii Loop (for TikTok, highest rewatch rate)

- Character: Cute chibi anime character, big eyes, cat ears, pink dress, transparent background

- Motion: $viral_hugo_dance

- Tips: Enable loop output, white background in character image, 24 FPS. The loop is the whole point — watch it muted first to check the seam before adding audio.

- Result: Seamless endless loop. This is the format TikTok rewards with replays more than almost anything else.

Professional Rickroll (for LinkedIn and office humor)

- Character: Senior lawyer man, navy suit, briefcase, composed professional expression

- Motion: $rickroll_never_gonna_give_you_up

- Tips: Office background added in editor after generation, let the character’s formal energy clash with the dance. Do not smooth it out.

- Result: Office meme content. Works on LinkedIn because it subverts the platform’s usual tone.

Cyberpunk Moonwalk (for Instagram art accounts)

- Character: Anime girl in neon cyberpunk outfit, short pink hair, glowing blue eyes, white background

- Motion: Moonwalk reference clip (search “Michael Jackson moonwalk 4K”)

- Tips: Generate first, then add a neon city background in CapCut or VN. Do not include a background in the character prompt — Viggle blends it incorrectly.

- Result: Looping art video with high save rates. The white background in the prompt is what makes the edge quality clean.

Sigma Wolf Strut (for motivation and fitness Reels)

- Character: Sigma male wolf, sunglasses, leather jacket

- Motion: Slow confident walk reference (search “sigma walk video” on YouTube, pick one that shows full body, 8–12 seconds)

- Tips: City street background in editor, slow-motion setting in Viggle if available

- Result: Motivational gym content. The character type matches the meme format perfectly and the combination gets strong comment engagement.

NPC Wojak (for gaming and relatable meme content)

- Character: Bald everyman face, simple features, neutral expression, gray background

- Motion: $robot_dance_walk

- Tips: Add NPC text overlay in editor (“Thanks for visiting,” “Okay,” “I will stand here now”)

- Result: Gaming culture meme. The NPC format is still getting strong engagement in 2026 and this character description produces a cleaner result than using a real photo.

The full pack has 25 more of these. Grab the complete 30-combo pack here

Four things that will improve your output immediately

Image quality first. Use 1024x1024px minimum for any character image you upload. Front-facing, upright pose. Viggle does badly with profile angles as a starting image : it can generate them, but it takes more iterations.

White or solid background in the character image. Adding a background to your source image causes Viggle to blend edges incorrectly. Generate on white, add the background afterward in editing. This is the single most common mistake I see in Viggle content that looks slightly off.

Reference video length matters. Keep motion reference videos between 5 and 12 seconds. Longer clips cause the AI to pick inconsistent frames from different parts of the motion. Shorter than 5 seconds and the motion does not have enough data.

Iterate, do not discard. Viggle’s output is probabilistic. The first generation of a character is rarely the best. Run the same prompt 3–5 times and pick the cleanest result. The prompt is the setup the iteration is where you actually find the winner.

The prompts in this article are about a quarter of what I have tested and documented.

The full pack includes 160 character description prompts across 8 categories (anime, realistic humans, cartoon, fantasy, sci-fi, meme, business, sports, historical), the complete preset reference table with 30 presets and which character types they pair with, all 30 combo templates, and 10 production tips.

If you want to test the format before buying, there is also a free 30-prompt starter pack with 8 combo templates and no email required.

I publish weekly breakdowns of AI tools and prompts at AI Tool Discovery. If you found this useful, the newsletter covers new tools every week without the noise.

(Amara writes about AI tools, prompts, and practical workflows at aitooldiscovery.com. Every prompt in this article has been tested in Viggle AI.)