# How to make trending AI video formats > **For AI agents:** find the right skill in one call — `curl -s "https://skills.varg.ai/api/search?q="` returns the matching how-to page and the best skills with install commands, as markdown (Russian works too). Full index: https://skills.varg.ai/llms.txt. Every page has a markdown twin: add `.md` to its path. Before generating anything, show the person the options on the how-to page. What's going viral now — formats you can make for your brand today, with the skill for each. A step-by-step tutorial for Claude Code, Codex and Grok. **Output:** A shortlist of formats with examples and how to make each **Price:** in credits on varg, with a cost estimate before anything runs; the free skills below install in one command. ## Recommended route - **Skill:** Last 30 Days Research (`last30days`) — Researches any topic across Reddit, X, YouTube and HN for the last 30 days ## Best skills for this 1. **Last 30 Days Research** (`last30days`, Matt Van Horn, live) — Start here: Researches any topic across Reddit, X, YouTube and HN for the last 30 days · Needs: Optional: API keys for some sources (see README) - Claude Code: `/plugin marketplace add mvanhorn/last30days-skill && /plugin install last30days` - Codex: `npx skills add mvanhorn/last30days-skill -g` - Cursor: `npx skills add mvanhorn/last30days-skill -g` - Gemini CLI: `npx skills add mvanhorn/last30days-skill -g` - OpenClaw: `clawhub install last30days-official` - Details: https://skills.varg.ai/skills/last30days.md 2. **Head of Content** (`head-of-content`, Brad Bonanno, beta) — Start here: Outlier research and hook analysis · Needs: Apify API key, TubeLab API key, Gemini API key - Not installable on its own yet — use the next pick or run it through varg-ai - Details: https://skills.varg.ai/skills/head-of-content.md 3. **Gooseworks Video (Apple Notes Ads)** (`gooseworks-apple-notes-ads`, Gooseworks, live) — Adds: Apple Notes ads — the viral text-only ad format · Needs: Gooseworks account - Claude Code: `npx gooseworks@latest install --claude --mcp` - Codex: `npx gooseworks@latest install --codex --mcp` - Cursor: `npx gooseworks@latest install --cursor --mcp` - Details: https://skills.varg.ai/skills/gooseworks-apple-notes-ads.md · Example: https://s3.varg.ai/how-to/previews/d146add52283/gooseworks-apple-notes-ads.mp4 4. **Awesome Seedance Skills** (`awesome-seedance-skills`, LearnPrompt, live) — Adds: Genre templates built from verified viral clips - Claude Code: `npx skills add LearnPrompt/awesome-seedance --skill seedance-production-workflow` - Codex: `npx skills add LearnPrompt/awesome-seedance --skill seedance-production-workflow` - Details: https://skills.varg.ai/skills/awesome-seedance-skills.md 5. **FLUX 3 Archival Formats** (`flux-3-archival-formats`, Black Forest Labs, live) — Adds: Old-camera looks for the lost-home-video format · Needs: BFL API key (BFL_API_KEY) - Claude Code: `/plugin marketplace add black-forest-labs/skills && /plugin install flux-3-video@black-forest-labs` - Codex: `npx skills add black-forest-labs/skills --skill flux-3-archival-formats -a codex` - Cursor: `npx skills add black-forest-labs/skills --skill flux-3-archival-formats -a cursor` - Gemini CLI: `npx skills add black-forest-labs/skills --skill flux-3-archival-formats -a gemini-cli` - Details: https://skills.varg.ai/skills/flux-3-archival-formats.md · Example: https://s3.varg.ai/how-to/previews/0f57a7e83417/flux-3-archival-formats.mp4 6. **iMessage Video** (`varg-imessage-video`, varg, beta) — Adds: Text-message story videos - Not installable on its own yet — use the next pick or run it through varg-ai - Details: https://skills.varg.ai/skills/varg-imessage-video.md · Example: https://s3.varg.ai/how-to/previews/2fc5d178d91b/varg-imessage-video.mp4 ## Ask the person first Put these choices in front of the person before spending anything. Paste the example link next to each option: people pick a look by seeing it. **Which format?** - AI gameplay parody — [Fake AAA Gameplay Trailer](https://skills.varg.ai/skills/want-gameplay-concept-trailer.md) · example: https://s3.varg.ai/how-to/previews/31d5b877e9a6/want-gameplay-concept-trailer.mp4 - Lost home video (DV camcorder) — [FLUX 3 Archival Formats](https://skills.varg.ai/skills/flux-3-archival-formats.md) · example: https://s3.varg.ai/how-to/previews/0f57a7e83417/flux-3-archival-formats.mp4 - Caught-on-phone stunt · real example: https://x.com/techhalla/status/2093364330985136572 - Show opener built in code · real example: https://x.com/Wakuwakukucx/status/2088804332783435958 - Frame-break pop-out — [Frame-Break Pop-Out Shot](https://skills.varg.ai/skills/want-frame-break-pop-out.md) · example: https://s3.varg.ai/how-to/previews/7b094ad21157/want-frame-break-pop-out.mp4 - Apple Notes ad — [Notes-App Ad](https://skills.varg.ai/skills/want-notes-app-ad.md) · example: https://s3.varg.ai/how-to/previews/868762346d1a/want-notes-app-ad.mp4 ## How to make trending AI video formats: step by step 1. Find this month's outliers with last30days in discovery mode; keep posts with many bookmarks per like 2. Open the post and its first reply, and copy the prompt's block structure, not its text 3. Change the setting, the hero and the twist; put your product in the hero object or the background 4. Lock the first frame with a still, then write one 15-30 s take second by second with the event already happening at 0 s (for code formats: narration with word timestamps) 5. Approve the still or a contact sheet before spending, then generate 6. Label realistic AI and post your own prompt in the first reply ## What works now (researched 2026-10-27) Aug–Oct 2026 belongs to AI video that passes as a real thing: a real video game, an old home video, a phone clip of an absurd accident, a real TV opener. Each is made in one 15–30 s take, or built in code by an agent. Glossy "cinematic" clips captioned "this AI video is a masterpiece", model-vs-model comparisons and bought engagement (11k likes, 7 bookmarks) stopped spreading. 1. **Name the capture device and its flaws, not a mood: "early-2000s DV camcorder, friend filming, autofocus hunting, accidental zooms, no music, cuts to black mid-motion". Then give the world ordinary detail and one small mishap.** This prompt skeleton drove the biggest realism trend of the quarter. @Sheldon056's 30 s Seoul home video got 5.35M views and 11,064 bookmarks. At least 8 creators reused the same skeleton within weeks (3.67M, 843k, 753k views…). ([@Sheldon056 on X, full prompt in post, 2026-08-23](https://x.com/Sheldon056/status/2091396663718117706)) 2. **Write one continuous take with a second-by-second timeline, and start with the event already happening at 0 s. Seedance 2.5 renders 30 s in a single generation, so plan beats such as [0–3s HOOK], [3–5s] and so on, instead of stitching clips.** TechHalla's "try this at home" (64.4M views, 7,681 bookmarks) is one 30 s take. Its first beat reads "Phone already rolling… cola ALREADY jetting from the hole". The same structure gave him 15.4M views two weeks earlier. ([@techhalla on X, GPT Image 2.5 + Seedance 2.5 prompt reply, 2026-09-17](https://x.com/techhalla/status/2100526486113472703)) 3. **Post the full prompt or workflow in the post or the first reply. People save what they can copy.** In our sample of 40 viral AI-video posts from July–October 2026, the 22 that shared a prompt or workflow had a median of 0.95 bookmarks per like. The 18 that did not had 0.43. That is about twice the saves at the same reach. ([varg sample of 40 X posts, Jul–Oct 2026 (e.g. @AIwithkhan, prompt in post: 2,514 bookmarks / 4,656 likes)](https://x.com/AIwithkhan/status/2091724049684783146)) 4. **Ride a format within days, but change the setting, the character and the twist each time. Never post the same prompt with a new face.** "Grandma GTA" spread from 9 to 13 September (an Osaka mod got 1.13M views two days in, then an Edo edition and a racing spin-off). YouTube's 16 July 2026 clarification demonetizes "generic, repetitive, or template-based content". One creator's same-prompt DV series fell from 714k–843k views per post in late August to 201k–484k by October. ([TechCrunch, YouTube clarifies inauthentic-content rules (2026-07-20)](https://techcrunch.com/2026/07/20/youtube-clarifies-policies-around-ai-slop-and-upsetting-videos/)) 5. **Label realistic AI, and win on the idea rather than on fooling people. Keep real people and other companies' brands out of fake-footage formats.** TikTok's guidelines (released 25 August, effective 24 September 2026) require a label on AI content that "shows realistic-looking scenes or people". Meta reads C2PA since June 2026, and YouTube auto-labels since 27 May 2026. A 2026 study cited by Billo found AI labels cut engagement by 24–46 points, so the concept has to carry the video. ([TikTok Community Guidelines, Integrity & Authenticity (Aug 2026)](https://www.tiktok.com/community-guidelines/en/integrity-authenticity)) **Stopped working:** - Caption-farm reposts ("This AI video is absolutely masterpiece 🔥") and creator-program posts with bought likes: 11k likes with 7 bookmarks reads as fake and teaches nothing. - Fake-footage stunts built on real people or other companies' brands ("Nike's legal team woke up to a nightmare…"), and school-uniform or child protagonists in chaos formats. Realistic AI needs a TikTok label, and these get removed or called out. - Daily reposts of the same prompt with a new face. Reach decays fast, and YouTube's July 2026 rules demonetize template-based channels. ## Prompts that work Templates from the creators of the best current results and from the model guides. Fill in the {CAPS}. ### Lost home video take (DV camcorder) (seedance_2_5) One 30 s Seedance take of an ordinary day filmed on an old DV camera, with one small mishap and a cut to black; a product can sit in the background. Adapted from [@Sheldon056 on X, full Seedance 2.5 prompt in post (5.35M views), with @AIwithkhan's mishap version and awesome-seedance's 41-case DV skill](https://x.com/Sheldon056/status/2091396663718117706). ```text Create a 30-second ultra-realistic personal home video of an ordinary {SEASON} day in the life of {SUBJECT}, filmed by a friend on an early-2000s consumer DV camcorder. Spontaneous, imperfect, observed, not performed. MAIN SUBJECT: {AGE}, {HAIR}, {FULL OUTFIT}, {ONE ACCESSORY}. Keep face, hair, outfit and body identical from start to end. LOCATION: {LIVED-IN NEIGHBORHOOD} with {5-8 ORDINARY DETAILS}. No landmarks, ads or recognizable brands. {PRODUCT} appears only as an everyday background object. CAMERA: handheld by a friend walking nearby. Natural shake, drifting framing, autofocus hunting, exposure shifts between sun and shade, accidental zooms, soft detail, muted colors, off white balance, compression noise. The operator reacts late and sometimes cuts off the subject. No stabilization, gimbal, drone, slow motion or cinematic grade. 00:00-00:05 {OPENING MOMENT}; notices the camera, small nod. 00:05-00:12 {WALK OR ERRAND}. 00:12-00:18 {SMALL MISHAP}; tries to fix it twice, fails. 00:18-00:24 {REACTION}: laughs, carries on; {CALLBACK TO AN EARLIER DETAIL}. 00:24-00:29 Turns to camera: "{ONE SHORT LINE}". Walks away. At 00:29 the recording cuts to black mid-motion. No fade. PHYSICAL REALISM: natural hands, feet on the ground, no extra fingers, no floating or vanishing props; {PROP} stays where it lands. AUDIO: location sound only ({3-5 SOUNDS}) and camera-handling noise. No music, no narration. FEEL: a forgotten recording, not a commercial. 16:9. ``` ### Raw phone stunt, one 30 s take (seedance_2_5) The caught-on-phone stunt: lock the first frame with a still (gpt_image_2_5, dashcam or phone JPEG look), then this Seedance prompt plays the whole gag in one take. Adapted from [@techhalla on X, "RAW HOME PHONE VIDEO" Seedance 2.5 prompt in the first reply (15.4M-view post); still + timeline structure from his 64.4M-view follow-up](https://x.com/techhalla/status/2093364333518414064). ```text RAW PHONE VIDEO. One continuous 30-second vertical 3:4 clip filmed by a bystander on a current smartphone from {VANTAGE POINT}. Amateur handheld, micro-shake, clumsy digital pinch-zooms in and out, no cuts. Auto exposure, slightly crushed highlights, ordinary digital noise. No film look, no cinematic grade. Feels like a neighbor filmed it. @Image1 = locked first frame: {SET} with the event ALREADY happening. Set and framing in @Image1 are law. LOCKED CAST: {HERO: one adult, look, outfit}. Nobody else. HERO OBJECT: {PRODUCT}, starts {STATE}, stays in {HAND} until {MOMENT}. TIMELINE 0-3s [HOOK] Phone already rolling, {EVENT} already happening; the first zoom punches in. 3-10s [Handheld wide] {HERO} enters from the right, calm, {SETUP ACTION}. One contact only. 10-20s [Lazy zoom] {ESCALATION WITH HERO OBJECT}; hold the key action on screen for 2 full seconds. 20-27s [Widening, more shake] {IMPOSSIBLE PHYSICAL PAYOFF}; the phone struggles to track. 27-30s [PAYOFF] {AFTERMATH}; the phone holds on the empty spot, one tiny final zoom. AUDIO: diegetic only: {3-5 SOUNDS}, phone mic clipping. No score, no dialogue. QUALITY: real weight, liquid and debris physics, wet skin and fabric, coherent objects, identity and set locked to @Image1. ``` ### Open-world game parody take (seedance_2_5) The grandma-game format: an unlikely hero on an everyday errand, shown as third-person gameplay with a pinned HUD that changes every beat. Adapted from [@mi7_crypto on X, shared prompt for the Edo edition of the grandma-game wave (Seedance 2.5, text to video), plus awesome-seedance's game-HUD rules](https://x.com/mi7_crypto/status/2098577397717307413). ```text 30-second photoreal clip shown as open-world crime-game gameplay capture, 16:9, one continuous third-person take. Real {CITY OR ERA} streets, but clearly a GAME, not a film or cartoon. Fast, comic, lawless. PLAYER: {UNLIKELY HERO}, @Image1 for face only, {OUTFIT IN TEXT}, {TEMPERAMENT}. Appears once, never duplicated. Sidekick NPC: {SIDEKICK}. Exactly {N} other characters. CAMERA RIG: 1.5 m behind the player, slightly camera-right, FOV breathing 30-60 degrees. Light screen shake on every hit, red flash on damage. HUD, fixed positions, layout never moves, only values change: circular minimap bottom-left, mission line top-center, WANTED stars top-right, health bar and item slot bottom. Show game text exactly as written, 2-4 caps words. 0-5s {HOME}: {SIDEKICK} demands {EVERYDAY ITEM}. HUD: "MISSION: {GET ITEM}". 5-11s Street: the player shoves past a passer-by. HUD: "WANTED 1 STAR". 11-18s {SHOP} brawl, props fly. Popup: "{ITEM} GET". 18-24s Steals {VEHICLE OR MOUNT}, the sidekick clings on. "WANTED 2 STARS". 24-30s Chase through {STREET}. "WANTED 3 STARS". Exactly one hard cut to black at 29.5s. AUDIO: engine, footsteps, crowd, short lines in {LANGUAGE}, game SFX on pickups. NEGATIVE: no real game names or logos, no real people, no gore, no children, HUD never drifts. ``` ## Trending formats to remake To remake one: download the example with the `media-download` skill, take a frame a second and the transcript, write a beat list (time, shot, camera, on-screen text, sound), keep what the format needs, swap in the person's product or character (never the creator's face, voice, logo or footage), show the beat list and stills before spending, then make it with the skill named below. 1. **AI gameplay parody (everyday life as an open-world game)**: A photoreal third-person "game" with minimap, health bar, mission text and WANTED stars, starring an unlikely hero (a granny, an office worker, your mascot) doing an everyday errand that turns into chaos. Why it spreads: This is the biggest copy wave of the quarter. The original "grandma game" got 6.84M views and 7,709 bookmarks on the creator's post, and a repost of the same clip got 83.8M views and 142,646 bookmarks. 11+ creators made their own mods within four days (Osaka, Edo, a mobility-scooter racing game), and the creator posted that many look-alikes were circulating. Keep: Keep the third-person camera behind the hero, the HUD pinned in fixed spots (minimap, mission line, wanted stars) and changing every beat, and an everyday errand that escalates. Swap the hero, the city or era and the cheat item. Never name or copy a real game. No dedicated skill yet ([wanted](https://skills.varg.ai/skills/want-gameplay-concept-trailer.md)): build it with `varg-ai`. - A shared prompt for this format: https://x.com/mi7_crypto/status/2098577397717307413 - Example to remake: [@Ai_Labokoikoi on X, 2026-09-10, 6.8M views, 8K saves](https://x.com/Ai_Labokoikoi/status/2097905327111114825): The original, from the creator's own account. Look at the HUD grammar: circular minimap, red mission line in subtitles, health bar, game-style camera behind the granny. Made on Pollo AI with Seedance 2.5 (watermark). No prompt; the sequel is on YouTube. - Example to remake: [@SSSS_CRYPTOMAN on X, 2026-09-11, 1.1M views, 1K saves](https://x.com/SSSS_CRYPTOMAN/status/2098336640997744921): A copy posted two days later as an "Osaka MOD", where candy is the cheat item. It shows how to localize the format rather than clone it. Pollo AI, Seedance 2.5. 2. **Lost home video (early-2000s camcorder realism)**: 30 s of an ordinary day filmed by a friend on an old DV camera: one handheld take, natural sound only, one small mishap, an abrupt cut to black. A brand can place its product as a background detail. Why it spreads: It is the most copied realism prompt of the quarter: 5.35M views and 11,064 bookmarks for @Sheldon056, 3.67M for @RishuaVR, 753k and 4,218 bookmarks for @kingofdairyque, plus 8+ more creators. Bookmarks often outnumber likes because people save the prompt. Keep: Keep the camera defect list, one ordinary day with one small mishap, natural sound with no music, and the cut to black mid-motion. Swap the person, the neighborhood and the mishap. A product only appears as a background object. Make it with [FLUX 3 Archival Formats](https://skills.varg.ai/skills/flux-3-archival-formats.md) (`flux-3-archival-formats`). - A shared prompt for this format: https://x.com/Sheldon056/status/2091396663718117706 - Example to remake: [@Sheldon056 on X, 2026-08-23, 5.3M views, 11K saves](https://x.com/Sheldon056/status/2091396663718117706): A young man's summer day in a Seoul alley: rooftop drink, missed basketball shot, sudden rain. The full timestamped Seedance 2.5 prompt is in the post. Copy its CAMERA block and its "no music, cut to black at 00:29" ending. - Example to remake: [@AIwithkhan on X, 2026-08-24, 843K views, 3K saves](https://x.com/AIwithkhan/status/2091724049684783146): The same look with a story beat: a sandal strap breaks on a milk run, and the punchline is "I only went out for milk." The full prompt is in the post. It shows how one small mishap gives the clip an ending. 3. **Caught-on-phone stunt (one-take fake phone footage)**: A vertical phone or dashcam clip filmed "by a bystander". An everyday scene escalates into an impossible physical gag in one continuous take with clumsy pinch-zooms. A brand version puts your own product at the center of the stunt. Why it spreads: TechHalla's two clips got 64.4M views (7,681 bookmarks) and 15.4M (6,164). @mightyking's raw-phone marina jump got 424k views and 2,840 bookmarks. Tanker-spill and inflatable-sneaker copies spread across repost accounts in September. Keep: Keep the bystander phone (pinch-zooms, auto exposure), the event already happening at 0 s, and one 30 s take that escalates to an impossible physical payoff. Lock the first frame with a still. Swap the setting and make your product the hero object. No real brands or people. No dedicated skill yet: build it with `varg-ai`. - A shared prompt for this format: https://x.com/techhalla/status/2093364333518414064 - Example to remake: [@techhalla on X, 2026-08-28, 15.4M views, 6K saves](https://x.com/techhalla/status/2093364330985136572): A man in a kiddie pool on a balcony, filmed by a neighbor, until the balcony gives way. The "RAW HOME PHONE VIDEO" prompt is in the first reply. The post itself is framed as a hook for a tool launch. 4. **Agent-built show opener or explainer (motion graphics in code)**: A coding agent (Claude Code or Codex with Remotion or HyperFrames) builds a TV-style title sequence or a Kurzgesagt-style "what if" explainer frame by frame in code, with voice and music synced, and no video model. Why it spreads: A fake TV-show opener made with Claude Code × Remotion got 11.7M views and 6,186 bookmarks. A Claude-drawn fruit-fly animation got 1.2M views and 3,904 bookmarks. A free Kurzgesagt-style prompt got 5,122 bookmarks. Workflow threads for the stack have 14k–18k bookmarks. Keep: Keep a camera move or transformation on every narrated line, events fired on word timestamps, one palette and one typeface, and a contact sheet before render. Swap the topic, mascot and palette. Never copy a named studio's characters or look. Make it with [HyperFrames](https://skills.varg.ai/skills/hyperframes.md) (`hyperframes`). - A shared prompt for this format: https://x.com/finnardley/status/2114305886420086817 - Example to remake: [@Wakuwakukucx on X, 2026-08-16, 11.7M views, 6K saves](https://x.com/Wakuwakukucx/status/2088804332783435958): A 15 s opener for a made-up show, "What if great figures wrote résumés": a fountain pen writes Oda Nobunaga's name on a résumé form, then the show logo lands. Built with Claude Code × Remotion. No prompt shared. - Example to remake: [@finnardley on X, 2026-10-25, 236K views, 5K saves](https://x.com/finnardley/status/2114305886420086817): "What if the Sun vanished?" in Kurzgesagt style, drawn entirely in canvas and WebGL. The full Claude Code prompt is in the post: palette hex codes, a camera move on every narrated word, an ElevenLabs word-timed voice, a −14 LUFS master, and a contact sheet before render. ## Open-source repos - [LearnPrompt/awesome-seedance](https://github.com/LearnPrompt/awesome-seedance) (2K★): Seedance 2.5/2.0 prompt library traced to original posts, with one agent skill per viral genre (DV home video, game HUD, POV take, meme comedy). How: 840+ verified cases are distilled into genre SKILL.md files, each a fixed block structure plus cases.md; the agent picks one anchor case, fills every block, and returns one paste-ready prompt with a 3-line QA checklist. Install as a skill: https://skills.varg.ai/skills/awesome-seedance-skills.md - [mvanhorn/last30days-skill](https://github.com/mvanhorn/last30days-skill) (66K★): Agent skill that finds what people are actually engaging with this month across Reddit, X, YouTube, TikTok, HN and more. How: Searches every source in parallel, scores items by real engagement (upvotes, likes, views), and an agent judge writes a cited brief; discovery mode ("what's exploding in ...?") returns 5-10 velocity-ranked topics with a follow-up command. Install as a skill: https://skills.varg.ai/skills/last30days.md - [heygen-com/hyperframes](https://github.com/heygen-com/hyperframes) (74K★): HTML-to-video framework built for coding agents, for the show-opener and explainer formats made in code. How: The agent writes each scene as seekable HTML/CSS/GSAP with timed data attributes, previews it, and renders a deterministic MP4; voice, captions and music are added as tracks. Install as a skill: https://skills.varg.ai/skills/hyperframes.md ## Set up varg varg is one API key for the models these skills drive — Seedance 2.5, Kling 4, Veo 3.1, Wan 3, Nano Banana 2.1, GPT Image 2.5, ElevenLabs v4 and 200+ more — billed in credits, with a cost estimate before anything runs. - Claude Code: `claude plugin marketplace add vargHQ/skills && claude plugin install varg@varg` (skill + hosted MCP) - Codex, Cursor, Gemini CLI, OpenClaw: `npx skills add vargHQ/skills --skill varg-ai -a -y` - Grok and any agent that reads SKILL.md: `npx skills add vargHQ/skills --skill varg-ai`, or point it at the skill's SKILL.md - Claude.ai, ChatGPT, any MCP client: connect `https://mcp.varg.ai/mcp` - API key: https://varg.ai/dashboard/api · Model specs: https://varg.ai/models/llms.txt ## Related tutorials - [How to remake a reference video with AI](https://skills.varg.ai/remake-a-reference-video.md): Take a reel or ad you like, break it down shot by shot, and remake the format for your brand.