What is the best AI to create funny videos?

Why 'Funny AI Video' Is a Hard Problem (and Why One Tool Isn't Enough)

Ask ten AI tools to "make a funny video" and you'll get ten competent, forgettable results. The uncomfortable truth is that no AI tool on the market is purpose-built for comedy — the best you can do today is combine tools that are each good at one layer of the job.

The evidence on AI humor is genuinely surprising. In a University of Southern California study, roughly 70% of participants rated ChatGPT-written jokes as funnier than human-written ones, with only 25% favoring the humans. In a rewriting test against The Onion, ChatGPT tied professional comedy writers, and the single top-rated headline was the AI's own: "Local Man Discovers New Emotion, Still Can't Describe It Properly."

But working comedians tell a more complicated story. Fringe performer Anesti Danelis found ChatGPT useful for ideation and even structuring a show, yet concluded that "human creativity can't be replicated or replaced", while fellow comedian James Roque argues the best comedy is "deeply human and vulnerable" — something AI can't produce. Audiences, Roque adds, "can sniff out when something isn't authentic," which is why transparency about AI use matters.

The resolution of that contradiction comes down to one word: delivery. USC researcher Drew Gorenz points out that "delivery is such a key part of humor production" — a joke read as plain text consistently lands worse than one performed well. That's exactly the layer video adds, and it's why a written gag that scores modestly on a page can become genuinely funny with the right timing, expression, and pacing on screen.

So the practical answer isn't one tool — it's a two-layer workflow:

• A large language model like ChatGPT for the comedic writing itself, where the evidence is strongest.
• A text-to-video generator for visuals and delivery — text-to-video already holds 46.25% of the AI video market, meaning most funny videos start as a written script.
• A human in the loop for punchlines, authenticity, and final judgment, since comedians report AI is stronger at ideation than at landing jokes.

The good news is that the video layer has matured fast. Zapier's hands-on testing names Google Veo 3.1 the best all-around video generator, capable of instant lip-synced dialogue — precisely the delivery layer humor needs. And the timing is right: social media is the fastest-growing AI video application, expanding at a 23.5% CAGR, fueled by demand for short-form content on TikTok and Instagram.

Agent skills can automate the surrounding workflow — script drafting, prompt generation, repurposing clips for social — and SkillzBench is building a free directory where you can browse and test those skills, with a free MCP connector to hook them into your agent of choice.

The Two-Layer Solution: An LLM for the Jokes, a Video Generator for the Delivery

The funniest AI video doesn't come from a single tool — it comes from a partnership between two very different AIs. One writes the joke; the other performs it. That's the workflow the research consistently points toward, and it maps neatly onto how the market already works.

Text-to-video dominates AI video creation, holding 46.25% of the market as the leading input method, according to market research from Fortune Business Insights. In other words, most funny AI videos begin life as a written prompt or script. That makes the writing layer — the layer that actually generates the humor — critical.

Here the evidence is surprisingly strong. A USC study covered by CBS News found that roughly 70% of participants rated ChatGPT-written jokes funnier than human-written ones, and ChatGPT tied professional comedy writers when rewriting satirical headlines. USC researcher Drew Gorenz offers the key caveat, though: "Delivery is such a key part of humor production." A joke read as text lands worse than a joke performed.

That's exactly the gap a modern video generator fills. Industry analysis of 2025's text-to-video models notes they have crossed the uncanny valley, producing natural movement, lip-synced dialogue, and "appropriate emotional responses to script tone" — meaning the generated actor now reacts to whether your script is deadpan, manic, or absurd. Google Veo 3.1, which Zapier's hands-on testing calls "the best AI video generation all-arounder," generates video and audio together with instant lip-synced dialogue, directly solving the delivery problem.

The practical two-layer workflow looks like this:

• Draft the comedic script with an LLM like ChatGPT, iterating on premises, punchlines, and tone.
• Feed that script into a text-to-video generator such as Veo 3.1 or Runway Gen-4.5 for the performance.
• Use LTX Studio's shot-by-shot storyboarding when gag pacing demands precise timing control.
• Keep a human editor in the loop — comedians report AI excels at ideation but not final polish.

The production time savings are real: Fortune Business Insights reports generative AI cuts video production time roughly in half. And if you want to automate the surrounding steps — script drafting, prompt generation, clip repurposing — agent skills can handle much of the busywork. SkillzBench is building a free directory and MCP connector for exactly that, so you can browse, benchmark, and equip proven skills as part of your pipeline.

The Contenders: Veo, Runway, LTX Studio and How to Choose

Choosing a video generator is only half the equation for comedy — the other half is delivery, and USC researcher Drew Gorenz notes that jokes read as text consistently land worse than performed ones. That makes your video tool choice genuinely consequential: it supplies the performance layer that turns a decent gag into a laugh-out-loud clip.

After roughly a month of hands-on testing, Zapier's team crowned Google Veo 3.1 "the best AI video generation all-arounder on the market." It combines strong prompt adherence with video-plus-audio generation and instant lip-synced dialogue — exactly what comedic timing needs. The main caveat: it struggles with complex scenes involving lots of movement or characters. Pricing is friendly to experimenters, with 50 free credits per day, a $4.99/month tier, $19.99/month Pro, and Ultra from $99.99/month.

Runway Gen-4.5 takes a different angle, scoring highest in blind preference leaderboards against Google and OpenAI models. Its Aleph model edits existing footage via text prompts, and Act Two transfers human acting performances onto AI characters — useful when your physical delivery is the joke. A free plan offers 125 one-time credits, with Standard at $12/month.

For creators who think in punchlines per shot, LTX Studio offers shot-by-shot storyboarding control over characters, camera angles, and scene composition — the closest thing to a director's chair for gag pacing. It starts with a free trial of 800 credits, then Lite at $12/month. Adobe Firefly rounds out the field with commercially safe outputs, priced at $9.99/month after limited free generations.

Quick comparison by use case:

• Best free starting point: Veo 3.1's 50 daily credits
• Best visual quality: Runway Gen-4.5's blind-preference wins
• Best comedic timing control: LTX Studio's storyboarding
• Best for client or brand work: Adobe Firefly's commercially safe outputs

Whichever tool you pick, the surrounding workflow — script drafting, prompt refinement, repurposing clips for TikTok and Instagram — can be automated with agent skills. SkillzBench is building a free directory and MCP connector for exactly that, so you can browse, verify, and benchmark skills before committing them to your pipeline. Because when the video generator handles the delivery, everything else should run itself.

A Step-by-Step Workflow for Your First Funny AI Video

The gap between a funny idea and a funny video is workflow. Here's how to close it in five steps, grounded in what the research actually shows works.

Step 1: Draft the script with an LLM. A USC study found roughly 70% of participants rated ChatGPT's jokes funnier than human-written ones, so start there. Prompt for ten punchline variations, not one — comedian Viv Ford found ChatGPT's self-assessments unreliable: jokes it calls "funny" flop, while ones it flags as "offensive" often land.

Step 2: Storyboard the gags. Comedy lives in timing, so map each beat shot by shot. LTX Studio offers exactly this kind of shot-by-shot storyboarding control over camera angles and scene composition, which suits gag pacing.

Step 3: Generate with lip-synced audio. USC researcher Drew Gorenz notes that "delivery is such a key part of humor production" — text-only jokes underperform. This is where video earns its keep:

• Google Veo 3.1 generates video and audio together, with instant lip-synced dialogue from a prompt.
• Its free tier offers 50 credits per day, enough to iterate on timing without paying.
• Avoid complex multi-character scenes — Zapier's testing flags this as Veo's weak spot.

Step 4: Keep a human in the loop for punchline polish. Working comedians are blunt about AI's limits. Anesti Danelis, whose Fringe show was ~20% pure AI, found it valuable for ideation but concluded human creativity "can't be replicated or replaced." James Roque adds that audiences "can sniff out when something isn't authentic." Edit the final punchlines yourself, and be transparent about AI use — comedian Bethany Radford emphasizes audiences care about that honesty. Danelis also warns against AI as a crutch, or you risk "a generation of comedians saying the same repetitive, distilled stuff."

Step 5: Publish to short-form social. Social media is the fastest-growing AI video segment, at a 23.5% CAGR, driven by demand for short-form, high-volume output on TikTok and Instagram. Since viewers decide within 3 seconds whether to keep watching, front-load your strongest visual gag.

To streamline the surrounding work — script drafting, prompt iteration, repurposing clips per platform — agent skills can automate much of this pipeline. SkillzBench is building a free directory and MCP connector for discovering, verifying, and benchmarking those skills, so you can test which ones genuinely lift your output before trusting them with your workflow.

Automate the Rest: Agent Skills for Scripting, Prompting and Repurposing

The funniest video in the world dies if nobody rewrites it for TikTok. That's why the real workflow isn't just "pick a generator" — it's everything around the clip: drafting the script, engineering the prompt, and cutting one long video into platform-ready shorts.

Here's where agent skills earn their keep. These are instruction packages (SKILL.md files) that teach coding agents like Claude Code, Codex, Cursor and Gemini CLI how to handle repeatable tasks. For a comedy pipeline, a well-chosen skill can:

• Draft joke scripts in your voice, using ideation prompts like the ones comedian Anesti Danelis used to build his Fringe show — which he described as roughly 20% pure AI and 80% human-AI mix, per BBC reporting
• Generate and refine text-to-video prompts, which matters because text-to-video is the dominant input method, holding 46.25% of the market
• Repurpose one master video into TikTok, Reels and YouTube Shorts formats — critical since social media is the fastest-growing AI video segment at a 23.5% CAGR
• Batch-edit captions, hooks and thumbnails so your first three seconds land, given that viewers decide whether to keep watching within three seconds

The catch: not every skill actually helps. Published research found that roughly 26% of published agent skills carry at least one vulnerability, and quality varies wildly across the thousands of instruction packages floating around GitHub. Equipping a skill blind is a bit like hiring a writer without reading their material — except this writer can execute code on your machine.

That's the problem SkillzBench is built to solve. Its free directory continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, and runs license checks, security scans and quality judgments before anything gets listed. Every skill is then benchmarked in a sandbox — tested with the same model, with and without the skill — so you can see whether it actually improves output before you trust it with your workflow.

The platform is currently in pre-launch, but the waitlist is live. Once public, a free MCP connector will let you attach verified skills to any agent and toggle exactly which tools each agent can see. For video creators, that means a pipeline where script drafting, prompt generation and repurposing run through tested, verified skills rather than whatever turned up first in a search.

Comedians already warn that AI works best for structure and ideation, not final punchlines — as Danelis put it, human creativity "can't be replicated or replaced." Agent skills don't change that. They just remove the tedious scaffolding, so your human judgment goes where it matters: the joke itself.