agentleFS
vargHQvargHQ/skills/ai-parody-video/SKILL.md

ai-parody-video

Skill34 starsChanged 8 months agoMITView on GitHub

Plan and make a short AI parody video: a known movie scene, TV format or news shot with one wrong element, 10 to 45 seconds, built to be shared. Gives the joke structure, six formats measured on clips with 60 thousand to 9 million views (scene recast with new dialogue, original-vs-remake, one-prompt sitcom sketch, fake live news shot, character monologue, silent slapstick), which production route each one takes, prompt templates for the sketch and the news shot, subtitle and caption rules, and a plain list of what is allowed with real people, known characters, studio footage and cloned voices, with platform AI labels. Use when someone asks for a parody, spoof, meme video, movie mashup, crossover, "X in the world of Y", a celebrity or politician comedy clip, a fake trailer, or wants videos like the viral AI parodies on X, TikTok or Shorts. Then use scene-recast-video for remakes of an existing clip and multi-angle-monologue for speeches.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the request is for something in the right-hand column, say which line it crosses and offer the nearest version that works.

    line 49
  2. Distinctive:

    Use the beat the audience is waiting for: Every famous scene has one (the turn, the catchphrase, the look to camera).

    line 63
  3. Distinctive:

    Write it as a table: before any generation:

    line 65
  4. Very specific:

    Check: before posting: the face is the same person in every shot.

    line 151
  5. Distinctive:

    Series: when one lands, quote it with the next character within two days and ask what scene to do next.

    line 153

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI parody video: a scene everyone knows, one thing wrong, a payoff

The parodies that travel are not random "X as Y" pictures in motion. They borrow a scene the viewer already knows, so recognition is instant, put one topical wrong element into it, and save a second turn for the last shot. This skill is built from nine clips by @Arminn_Ai (X, August to November 2026) that we downloaded and measured shot by shot, his posted prompts and replies, and the tutorials listed at the end.

Before you start: what is allowed

Decide this before the joke, because it changes the idea. This is practice, not legal advice.

You want to useDoDo not
A real public figureComment on their public role, in a scene nobody could take for real, labelled as AI and as parody.Private people, minors, anything sexual or intimate, a fake statement or event that could pass as news, a health or crime claim, an endorsement, an ad, anything about voting near an election.
Their voiceDescribe a voice type ("slow, hoarse, drawn-out vowels") or perform an impression.Clone a real person's voice from recordings without their consent.
A known film or game characterParody that says something about the original; your own design "in the style of"; a look built from parts (coat, sunglasses, room).A "new trailer" or "leaked scene" that imitates official titles, logos or metadata.
Studio footage as the motion referenceSeconds, not minutes; changed cast and meaning; motion-only plate (see scene-recast-video).Re-uploads with a filter; selling it; putting a brand on it.
  • Filters are part of the answer. Video providers block many real faces, named franchise characters and known footage. BytePlus returns "The input image (or video) may contain a real person"; on varg even the relaxed Seedance routes keep real-person identity blocked. Higgsfield states that Seedance runs "its own separate IP check" it cannot override, and its terms ban bypassing moderation. When a job is refused, change the idea: an invented character who plays the type (the CEO in the leather jacket, the mentor in the long coat), a consenting person through the provider's authorisation route, or a stylised look (puppet, clay, cartoon). Do not look for inputs that slip a refused face or scene through.
  • Label it. YouTube requires disclosure when realistic content makes "a real person appear to say or do something they didn't", with no parody exception; Meta and TikTok add or require an AI label; the posts studied here carry X's "AI-generated" declaration. Write "parody" or "made with AI" in the caption as well. YouTube terminated two fake-trailer channels with over 2 million subscribers in December 2025 for misleading metadata.
  • The law moves. In the US the federal NO FAKES Act was not law as of 2026-09-30; state laws apply (Tennessee, New York, Washington, where the 2026 act reaches some non-commercial uses). Parody is generally protected speech; using a likeness to sell is the clear danger. Studios sent cease-and-desist letters over Seedance output in February 2026, which is why the filters exist. Sources at the end.

If the request is for something in the right-hand column, say which line it crosses and offer the nearest version that works.

1. Ask, then go

  • The world: which scene, show or format everyone in the audience knows. One.
  • The wrong element: who or what walks into it, and why this week (a headline, a launch, a match, a meme).
  • The payoff: what the last shot shows that the first shot did not promise.
  • Sound: dialogue, a news voice, or music only.
  • Where it runs: X and YouTube 16:9; TikTok, Reels and Shorts 9:16; a comparison post is stacked.

2. Build the joke

  1. Setup in the first two seconds, by picture. The known room, costume and camera style do the explaining. No title card.
  2. The swap is the premise, not the joke. "A in the world of B" earns one second of attention. The joke is what A's real-world habit does inside B's rules: the mentor's speech keeps its rhythm but is about this week's headline.
  3. Use the beat the audience is waiting for. Every famous scene has one (the turn, the catchphrase, the look to camera). Keep it, and bend it.
  4. A second turn at the end. The caption can point at it ("Wait till the end").
  5. Write it as a table before any generation:
TimePictureLine or soundJob
0–3 sWide: the known place, both characters walkingFirst line starts on frame 1Recognition
3–30 sA new picture for every phrase, 2 to 3 s eachThe speech in the borrowed rhythm, one topical noun per lineThe joke, repeated with rising stakes
30–38 sThe beat everyone knowsThe line everyone knows, bentPayoff
38–42 sOne new shotTwo wordsSecond turn

video-script sizes lines to the clock (about 2.5 words per second for dialogue).

3. Six formats, measured

Shot counts are from scene detection on the downloaded files; views as of 2026-11-12.

FormatMeasured exampleWhy it worksRoute
Scene recast with new dialogue42 s, 17 shots (2.5 s average), 16:9, one voice speaking from frame 1 to the end, two-line subtitles, the scene's famous beat at 32 s, a new last shot. 6.3M views, 80K likes, 23K bookmarksThe picture cuts on every phrase, so the speech never sits on one facescene-recast-video, or stills per shot then video (kling-4-keyframe-film), then one voice track over the cut
Original vs remake, stacked30 s, source on top, remake below, 1440x1620. 594K views; follow-ups with other characters 100K to 1.2MProof and joke in one frame; invites "do X next"scene-recast-video section 7
One-prompt sitcom sketch15 s, 6 shots, 6 lines, the two-word punchline at 6.8 s, a button line at the end. Prompt postedOne room, one joke, a look to cameraTemplate below, one generation with native audio
Fake live news shot17 s, one continuous helicopter take, a LIVE bug and a lower third, an anchor talking, the impossible event at 10 s (59% in), a stunned last line. 9.1M viewsThe news frame says "real", the event says "impossible"Template below
Character monologueAbout 60 s, four angles of one moment, a voice with a new tone per part. Steps and prompts postedPerformance carries it; the angle changes hide the stillnessmulti-angle-monologue
Silent slapstick10 to 20 s, no dialogue, music, one physical gag, a reaction. 75K viewsNo voice to get wrong, works muted and in any languageStills, then image-to-video with varg-video-prompting

Length: 10 to 30 s for a single gag, up to 45 s only when a speech carries it. The same account's tool demos and prompt threads get bookmarks but a small fraction of the reach; the clip goes out alone, the method goes in a reply.

4. Template: the one-prompt sketch

For a model with native dialogue and multi-shot prompts (minimax_h3_max, seedance_2_5, kling_v4). Structure from the posted prompt: place and camera style, who is where, dialogue with camera beats between the lines, the button, a style line.

<Show format> style. <Room>, <time of day>, <light>, <three set details>. <Camera: handheld
documentary camera, slightly unsteady, with sudden reframes and quick zooms toward whoever is
reacting>.

<Character A: where, doing what, mood>. <The others: where, mood>. <Character B: at the far
end, the one prop, completely disengaged>.

DIALOGUE:

A (loud, proud): "<setup line with the topical noun>."

Beat. Nobody reacts. Camera pushes in slowly on B, who does not look up.

A: "Which is why <authority> is giving <the thing B wants> to whoever..."

Instant snap zoom to B. <One physical action>. Dead serious.

B: "<two or three words>."

A (waving it off): "...anyway. Moving on. <Next topic>."

Snap zoom back to B, who <undoes the action> as if nothing happened.

Cut to <C>, who slowly turns and looks directly into the camera lens with a flat, knowing
expression. Hold for two seconds.

STYLE: single-camera sitcom shot like a documentary crew is present, natural performances, no
laugh track, muted colour palette, slightly soft consumer-camera look. No subtitles or
on-screen text.

Describe characters by role, age, build and clothes. If the cast must match specific faces, make stills first and use them as references.

Tested on varg on 2026-11-12: a three-shot, three-line sketch on minimax_h3_max ("audio": true, 12 s, 720p, text only) came back in one run for $0.96 with all three shots, each line spoken by the right character and a sigh it added itself. Two things it got wrong: the lines were over by second 9 and the last shot held three seconds too long, and it put a pointed hat on a wizard the prompt never gave one. Write what a character does not wear, and ask for 2 s less than the dialogue seems to need.

5. Template: the fake live news shot

Live local TV news helicopter footage, one continuous take, long telephoto lens from high
above, slight stabilised drift and small zoom corrections, late-afternoon sun, heat haze.
<The ordinary event for 9 seconds: who is chased, how many, the road, the barrier ahead>.
At 10 seconds <the one impossible thing, in physical detail: what moves, what it throws up,
where it lands>. The camera operator loses the subject for half a second and finds it again.
A news anchor speaks over it, calm and routine: "<two sentences of routine commentary>."
On the event: "Whoa... <reaction>." Last line, quiet: "<years on the job>. I've never seen
anything like that."
Broadcast look: slightly soft, compressed. No text or logos in the picture.

Add the red LIVE bug and the lower third in the edit (social-overlays or ffmpeg drawbox and drawtext), where the letters stay sharp. Two rules that keep it a joke: the event must be physically impossible, and no real channel's name, logo or presenter. A believable fake of real news is not parody.

6. Finish and post

  • Subtitles for anything spoken: white, two lines at most, lower third, each on screen as long as its phrase (caption-styles). They are the joke with the sound off.
  • Sound: one voice track over the whole cut beats per-shot voices that change from clip to clip (video-sound).
  • Check before posting: the face is the same person in every shot; nobody's lips move on another character's line; no leftover source subtitles or logos; the payoff is readable on a phone at arm's length (media-judge).
  • Caption: one line of setup and a hook, no tool list. Turn on the platform's AI label.
  • Series: when one lands, quote it with the next character within two days and ask what scene to do next. After the first remake, five of the six follow-ups over the next three weeks passed 100K views.

7. What it costs

Plan before spending (estimate_cost): a 12 s one-prompt sketch cost $0.96 a run; a 10 s recast cost $3.00 on MiniMax H3 Max Recast and $2.48 on Seedance 2.5 at 480p (scene-recast-video); a 60 s monologue is 6 lip-synced parts at about $0.90 each (multi-angle-monologue). Measured on varg on 2026-11-12. The creator's own numbers for his tool: a 30 s render in about 5 minutes, "easier shots can land in one try, harder ones usually need a few prompt passes".

Sources

Related skills

scene-recast-video, multi-angle-monologue, video-script, varg-character-design, varg-character-absurd, kling-4-keyframe-film, varg-video-prompting, caption-styles, social-overlays, video-sound, media-download, media-judge, varg-ai.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.