An AI video editor you prompt, versus one that invents footage

Prompt-to-video means two unrelated things. One invents footage. The other decides how to cut the footage you already have. The first wins demos; the second is what gets posted on a Tuesday.

State of generation

• Veo 3.1 leads on prompt adherence and generates synchronised audio, including dialogue and ambience, in one pass.
• Runway has the strongest editing workflow layered on top of generation.
• Kling 3.0 handles ambitious physical motion better than the rest.
• Luma is strongest animating a reference image, with self-correction that cuts repeated regenerations.
• Sora's consumer app shut down on 26 April 2026, with the API scheduled to stop on 24 September 2026.

Where generation still breaks

The failure modes are consistent: hands, fast movement, face continuity across shots, and audio that does not match the scene. None of that matters in a showreel and all of it matters when the subject is supposed to be you. Generated footage also carries rights and brand-safety exposure that lands on the publisher, not the model.

What a prompted editor does instead

1. You shoot one take on a phone and say what you want in a sentence.
2. The model decides where to cut, what to drop, what to illustrate and where to place emphasis.
3. Captions, music and b-roll are chosen against the meaning of what you actually said.
4. Versions per platform come out of the same decision, not a separate export session.

The honest state of agents

Agentic editing works and is early. An agent can draft a script, build a shot list, call the services it needs and assemble a rough cut. Human review still decides taste, rights, brand safety, factual claims and pacing. Autonomous assembly and publish-ready production are not yet the same thing, and anyone claiming otherwise is selling the demo.

How to choose

Where Monty sits

Monty is the prompted-editor side. One raw take in, a sentence about what you want, and out comes a finished vertical video: dead air and filler cut, captions synced to speech, music ducked under the voice, b-roll placed by meaning, audio cleaned, a version per platform, published to YouTube Shorts, Instagram Reels, TikTok and Telegram. It does not invent footage you never shot, and that limit is deliberate.