Can AI Make Videos?

The Short Answer: Yes, and It's Already Commercial

Yes — AI can make videos, and it's not a lab experiment anymore. It's a commercial industry generating real ad campaigns, real marketing content, and real revenue right now.

The numbers back this up. According to Grand View Research, the AI video generator market is valued at USD 788.5 million in 2025 and is projected to reach USD 3,441.6 million by 2033, growing at a 20.3% CAGR. A broader AI video market report puts the category even larger, projecting growth from $4.6 billion in 2025 to $42.3 billion by 2033.

The most vivid proof point came in June 2025, when prediction-market platform Kalshi aired an AI-generated NBA Finals ad that cost roughly $2,000 and took two days to produce. It drew more than 3 million views — while a traditional spot in the same slot averaged around $7 million, according to industry research cited by ngram. That's not a demo; that's prime-time advertising.

The mainstream players have arrived, too:

• Google launched Flow, an AI filmmaking tool powered by Veo 3, Imagen, and Gemini, in May 2025.
• Adobe brought its Firefly Video Model to public beta with a "commercially safe" generation guarantee.
• OpenAI signed a licensing deal with Disney, making Disney the first major content partner for Sora.

Enterprise adoption is scaling fast. Google Cloud reports that brands like WPP and Authentic Brands Group use these models to turn a single creative hypothesis into "hundreds or even thousands of personalized, cinematic creative variations in hours."

To be clear, this isn't a story about AI replacing filmmakers — the consensus across experts is that AI works best as a creative partner that augments human teams. And the technology still has real limits: most generative clips run 5–10 seconds, and physics errors remain common, per CNET's hands-on testing.

What this means for you is simple: AI video is real, mainstream, and moving fast. If your business isn't experimenting with it yet, your competitors probably are. And as video workflows increasingly run through AI agents, having vetted, benchmarked agent skills — like the free directory SkillzBench is building — becomes part of doing this well.

The question is no longer "can AI make videos?" — it's "what should you make with it?"

What AI Video Can Actually Do Right Now

Yes — AI can make videos today, and not just novelty demos. Companies are producing explainer videos, ad creative, and localized training content at a scale that would have required full production crews just a few years ago.

The field splits into two dominant paradigms. The first is avatar and script-based generation: tools like Synthesia and Colossyan turn a typed script into a finished video featuring a lip-synced presenter — no cameras, actors, or crews required, with avatars available in 140+ languages starting around $18 per user per month. The second is text-to-video generation, where models like Runway, Google's Veo 3, and Adobe Firefly create visuals directly from a text prompt. A milestone here: as CNET's testing notes, Veo 3 was the first major-platform tool to automatically create and synchronize AI audio with its AI video — an important leap for the field.

Beyond generation itself, AI now handles adjacent parts of the workflow:

• Automated editing — audio leveling, noise reduction, and captioning in post-production
• Document-to-video conversion — turning PowerPoints, blogs, and PDFs into video drafts
• Pre-production assistance, including script outlines and storyboards

Enterprise adoption is where the economics get striking. According to Google Cloud's use-case roundup, brands like WPP and Authentic Brands Group use Veo 3 and Imagen 4 to turn a single creative hypothesis into hundreds — even thousands — of personalized, cinematic variations in hours. Kalshi's AI-generated NBA Finals ad reportedly cost $2,000 and took two days, versus an average $7 million for a traditional spot.

The constraints are real, though. Generative clips typically run 5–10 seconds — Adobe Firefly caps at 5 seconds, Veo at 8 — so longer-form content still favors avatar-based tools. Physics "hallucinations" (objects appearing from nowhere, people floating through walls) remain common, and commercially safe training data is a genuine differentiator: Firefly is one of the few models that guarantees its outputs are commercially safe, while Runway and Midjourney train partly on user content.

Because these workflows are increasingly agentic — Google describes the shift from AI as passive assistant to AI as an active part of the team, orchestrating entire workflows — the tooling around them matters. SkillzBench is building a free directory and MCP connector for exactly this kind of vetted, benchmarked agent skill, so teams can extend AI video workflows with confidence rather than guesswork.

The Honest Limits: Short Clips, Physics Glitches, and Legal Gray Zones

For all the breathless demos you've seen online, today's AI video tools work in very short bursts — and the gaps between those bursts are where most projects live or die.

The headline constraint is duration. According to hands-on testing by CNET, generative tools cap clips at roughly 5–10 seconds: Adobe Firefly produces 5-second clips, Google's Veo and Omni manage 8 seconds, and Runway tops out at 5–10 seconds. Midjourney starts at 5 seconds, extendable to 21. Anything longer requires stitching multiple generations together, which is where visual consistency starts to break down.

Then there are the physics hallucinations. Even the best generators produce objects that appear from nowhere or people who float through solid surfaces — errors that are easy to miss in a quick review and embarrassing to catch after publishing. Avatar-based tools have their own tell: as Atlassian's Loom team notes, AI presenters "may lack human nuances," and heavy reliance on stock media can make outputs feel generic.

The legal picture is murkier still. Deepfakes of real individuals remain a hard no for responsible publishers — newsroom guidance treats all AI-generated material as unvetted source material requiring full verification before use. Copyright questions around training data remain unresolved across the industry.

One dividing line matters most for business users: commercial safety. Adobe's Firefly models are among the only options that guarantee content you create is commercially safe. Runway, by contrast, trains on user prompts and generated videos, and Midjourney can use your information to train its algorithms — a real consideration if your output is client-facing or brand-owned.

A few practical guardrails before you press "generate":

• Match the tool to the task: short 5–10 second generative clips for social ads; avatar tools like Synthesia for longer explainers.
• Review every frame before publishing — physics glitches hide in plain sight.
• Check training-data policies if the video will carry commercial rights.
• Disclose AI use, always. As CNET's guidance puts it bluntly: "If you think you shouldn't make a certain AI video, don't do it."

That last point deserves emphasis. A TechSmith viewer trends survey found 75% of viewers receptive to AI-created instructional video — but that receptivity depends on honesty about how the content was made.

The same verification discipline applies to the AI agents increasingly orchestrating these video workflows. SkillzBench is building a free directory and MCP connector that pin, scan, and benchmark agent skills against the same standard — because "trust but verify" applies to your tools, not just your output.

How to Put AI Video to Work: A Practical Playbook

Knowing AI can make videos is one thing; putting it to work profitably is another. The teams seeing real results aren't chasing cinematic masterpieces — they're starting where AI video is provably strong.

Start with short-form, high-volume content. According to industry research, the top use cases are product demos and explainers (31%) and social ads (28%), and landing pages using AI explainer videos see 34% higher conversion. That's where the 5–10 second clip limits of generative tools matter least.

For longer content, switch to avatar-based tools. Platforms like Synthesia turn typed scripts into presenter-led videos in 140+ languages without cameras or crews, making them better suited to explainers that run minutes rather than seconds.

Before anything goes out under your brand, check commercial safety. Hands-on reviews note that Adobe's Firefly models are among the only options guaranteeing commercially safe output, while some competitors train on user prompts and generated videos — a real consideration for business use.

The bigger win comes from treating AI as an augmentation layer across the whole workflow, not just a clip generator:

• Pre-production: script outlines, storyboards, and shot planning
• Generation: avatars and text-to-video for the footage itself
• Post-production: audio leveling, noise reduction, and captioning

Teams that adopt this full-stack approach report producing 11x more content with the same headcount. Google Cloud describes the same shift — from AI as a passive assistant to specialized agents orchestrating entire workflows. If you want to extend that idea beyond video, SkillzBench is building a free directory and MCP connector for vetted, benchmarked agent skills that can automate the surrounding workflow — scripting, scheduling, publishing.

Two rules keep you out of trouble. First, always disclose AI use: viewer research shows 75% receptivity to AI-created instructional video, and honesty protects that trust. Second, review every output before publishing — even the best generators still produce physics hallucinations, from objects appearing from nowhere to people floating through walls.

Going Further: Making Your Video Workflow Agentic

The most interesting shift in AI video isn't better clip generation — it's that AI is moving from passive assistant to active teammate, with agents that orchestrate entire workflows rather than wait for the next prompt.

Google Cloud describes this as a move "from AI as a passive assistant to AI as an active part of the team, where specialized agents can orchestrate entire workflows." For video, that means one agent can take a rough idea through script drafting, storyboard outlines, avatar or text-to-video generation, captioning, audio leveling, and publishing — the full pipeline AI already touches at pre-production, generation, and post-production stages.

The payoff is measurable. According to industry research cited by Ngram, teams using AI across the video workflow report producing 11x more content monthly with the same headcount, and 78% of marketing teams now use AI video at least quarterly.

The missing piece is trust in the skills that power those agents. Agent skills — SKILL.md instruction packages for tools like Claude Code, Codex, Cursor, and Gemini CLI — are proliferating fast, but published research found roughly 26% carry at least one vulnerability. Handing an unvetted skill control of your publishing pipeline is a real risk.

That's the problem SkillzBench is built to solve, and everything on the platform is free:

• Discover — continuously crawled, deduplicated skills from GitHub
• Verify — every version pinned to a commit SHA, license-checked, security-scanned, and quality-judged
• Benchmark — each skill tested in a sandbox with and without it, publishing the Lift in percentage points
• Equip — one free MCP connector plus per-agent loadouts, so your agent only sees the skills you choose
• Stay current — upstream changes tracked, re-scanned, re-benchmarked, with alerts

Trust scores run 0–100 across security, quality, license, maintenance, and community signals, and Lift always appears separately from ratings. Money never touches scores or listings — SkillzBench earns revenue from advertising around its blog articles, never by charging users.

SkillzBench is pre-launch and actively being built, so the directory, MCP connector, and benchmarks aren't public yet. The waitlist and newsletter are live for readers who want to discover, verify, and benchmark vetted skills before their video agents go to work.

Skills, proven.