Video tools, generated and edited

Two different products under one word

Video editing with AI assistance. Cutting, transcribing, captioning, removing filler words, reframing for a vertical screen, colour matching, separating a voice from background noise. This is where nearly all the practical value is, and it is mature.

Video generation. Producing footage from a text description or an image. Improving quickly, expensive per second, and useful for a narrower range of jobs than the demonstrations suggest.

Most people who say they want AI video want the first one.

The free path for editing is professional

DaVinci Resolve is free, used on released feature films, and includes AI-assisted transcription, subtitle generation, object removal, voice isolation and face refinement in the free version. The paid Studio version adds some features; the free one is not a trial.

Shotcut and Kdenlive are free and open source, lighter on hardware, and enough for most straightforward editing.

CapCut is free, phone-first, and good at the specific job of short social video with captions.

Against these, a subscription to a paid editor is buying a workflow, collaboration features, or a specific effect. That can be worth it. It is not buying capability the free tools lack.

What generation is and is not good for

Good for: short establishing shots, abstract or stylised sequences, background plates, product turnarounds, moving a still photograph slightly, and filling a gap where stock footage would otherwise be bought.

Bad for: anything with speech, anything needing a specific person to look the same across shots, precise action, hands doing detailed work, text on screen, and any sequence longer than a few seconds without a cut.

The mechanism explains the pattern. These models generate a short clip at a time, and consistency across clips is the hard problem. A one-minute finished piece is assembled from many short generations, most of them discarded.

Cost, which is the part people underestimate

Video generation is priced per second of output and the failure rate is high. A usable five-second shot commonly takes five to fifteen attempts. That makes the real cost of a shot several times the headline price, and a short film a serious budget.

Before committing to a project, generate one shot you actually need and count the attempts. The number is your multiplier for everything else, and it varies enormously by how specific your requirement is. Vague and atmospheric is cheap; a particular thing happening in a particular way is expensive.

Open video models exist and run locally on a strong GPU, which converts the cost into time and electricity. That is a real option for anyone doing this regularly with hardware to spare.

The workflow that works

Generated footage is best used as an ingredient. Generate the shots you cannot film, edit them together with material you have, and do the assembly in a real editor. Almost every good result you have seen was made this way rather than produced end to end by a single tool.

The same applies to the assistive features. Automatic captions are excellent and wrong often enough that you read them; automatic reframing is good and occasionally cuts somebody’s head off; filler-word removal is fast and sometimes takes a breath the sentence needed. Each of these saves most of the work and none of them removes the review.

The assistive features worth having whatever else you use

Three of them justify installing a modern editor on their own, and all three are free in Resolve.

Automatic subtitles, which take a job that used to cost hours and make it a review pass. Subtitles also matter more than most people assume: a large share of social video is watched without sound.

Voice isolation, which separates speech from background noise well enough to rescue a recording made in a busy room. This is the feature most likely to save footage you thought was unusable.

Reframing for vertical and square formats, which tracks the subject and follows it. It gets it wrong often enough that you check every cut, and it still turns an afternoon into twenty minutes.

Disclosure

Requirements for labelling synthetic video are emerging in several jurisdictions and on most large platforms, and they differ. Platform rules are the ones you will meet first: most major video and social platforms now require creators to disclose realistic synthetic content, and enforcement is uneven but real. Check the rules of the place you are publishing, and be aware they change more often than the law does.