AI video editing for reels, promos and explainers.

My AI video editing services cover talking-head reels, promos and ads, explainers, shorts and video at volume. AI handles the slow groundwork, such as transcripts and a first rough cut of a long recording, and I make every creative decision myself. Nothing goes out until you have approved it.

I am Anuj Paudel, a video editor in Kathmandu. My clients are small businesses, nonprofits and founders, most of them in the US, so the calls are by video.

What AI video editing services mean in my hands

A talking-head reel shows how the work divides between AI and me. You send a long recording of someone speaking to camera. A tool turns the speech into text with a timestamp on every word, and another pass makes a rough first cut, so I am not scrubbing through the whole recording myself. Both jobs are slow to do by hand and quick to check, which is why I hand them over.

After that, the work is mine. I choose the sentence that opens the video, decide which pauses stay, design what appears on screen as each line is spoken, grade the colour and mix the music under the voice. These choices decide whether a stranger watches to the end, and a tool does not know what your video is for. Then you watch the cut and decide whether it goes out.

An AI video editor, or an editor who uses AI?

Now, if you search for an AI video editor, many of the results are apps where you upload footage and the software gives back a cut. That is fine for a quick clip of your own. When the video speaks for your business, it helps to have a person who watches it the way a viewer will and can say why a cut feels slow, and that part I do myself. The longer argument for this split, including where AI still fails, is in my post AI video editing: what AI does well, and where a human editor still matters.

The kinds of video I edit

You can watch the reels, ads, lessons and shorts described below in my library at appynox.com/video.

Talking-head video editing

Talking-head video editing is mostly deciding what to take out. I cut the recording down to what the viewer needs, move the strongest line to the front and add word-by-word captions that follow the voice. When the speaker mentions something that can be shown, I cut to a picture of it, and the graphics on screen are built around that one line: a name, a place, a figure.

I have edited reels like this for several brands, with titles such as “Our origin story”, “Location secured” and “Potholes to people”. They are vertical 9:16 and run longer than most reels, from around ninety seconds to just under two minutes. At that length the captions and graphics do much of the pacing.

Promo and ad video editing

In a promo I open on the product at work, so the viewer sees the point before deciding to scroll, and I close on a single action to take. For SameShot, a progress-photo camera app we make at Appynox, I cut a 52-second launch video. I also made a 29-second ad for WideMath, our free browser math tools, and a 41-second ad for GharCalc, which offers free home-loan calculators for buyers in India. I made one more SameShot promo, 32 seconds long, with AI video.

Explainer videos and animated lessons

An explainer works if a viewer can repeat the idea after one watch. I script it, design the visuals and add the voice. One example is the animated lesson “Tens and Ones with Tally”, and for SameShot I made a 47-second illustrated story. This kind of video usually runs as an agent cut, described below, because most of what you see is drawn and animated instead of filmed.

Short-form video editing for Reels, TikTok and YouTube Shorts

Whether the job is Reels and TikTok editing or YouTube Shorts editing, the starting fact is the same: the viewer is on a phone with a thumb ready to scroll. That calls for a reason to stay in the first seconds, captions that make sense with the sound off, and framing that keeps clear of the spots where each app puts its own buttons and text. I have cut shorts such as a 24-second Plant Room social short and a 20-second short for Rhino Dash, our runner game for young kids.

Event promos, films and music videos

An event promo has to say what the event is, when it happens and why anyone should go, and it has to say it fast. I have cut promos for HYPERCE x BEAMSTART, an entrepreneurs’ meetup, and for Code Camp 2024, made for LOCUS at IOE Pulchowk, along with an introduction film for an IT club.

The camera came first. I started around 2019 with cultural films in Chitwan (Bhailo nights, Thali Nach, homestays), and since then I have cut a cinematic lyric film called Big Black Car, wedding films, music videos titled Darsan and Ud Jayega, a cinematic reel called Haule Se and a Langtang trek reel. That work is the base of the classic cut, described below.

AI-generated b-roll in real footage

B-roll is the footage that plays over a speaker’s voice to show what they are describing. When the thing was never filmed, AI-generated b-roll can stand in for it.

In a walk-through film that runs 64 seconds, you see the speaker, then a shot of what they are describing, then the speaker again. Only the shots in between are generated. I graded them to match the real film, with the same colour balance and contrast, so the cut between real and generated is hard to spot, and the voice you hear all the way through is the speaker’s own recording.

Video at volume

Past a certain number of videos, editing them one by one stops making sense. For two Appynox products, SameShot and WideMath, we built a system that makes the videos in batches. The two runs came to 85 social videos for SameShot and 65 shorts for WideMath.

Each video gets captions, motion text, a sound mix and an export for every platform size. Automatic checks run on each render first, and a person reviews it after. We run each batch the way a software team runs a release, against a posting calendar. The video at volume case study has the details.

Three workflows, and which one your project needs

Every project I take runs one of three ways. The right one depends on what you are making, how much of it, and how often.

The classic cut

The classic cut is hands-on editing. I keep only the strongest takes, cut in Premiere Pro, build the motion graphics in After Effects and grade with a library of LUTs, which are saved colour looks. Music drives the cut, so the rhythm of the track decides where the edits fall.

It suits films where every shot is a deliberate choice, such as a brand film, a wedding film or a music video. AI plays a small part here, mostly in transcripts and in finding shots in long recordings. The footage usually comes from you.

The agent cut

In the agent cut, the timeline is a web page. I write the script and a storyboard, the video is assembled as an HTML composition, and a deterministic render turns it into an MP4. Deterministic means the same inputs give the same video every time, so a re-render shows only the changes I made.

My video-editor agent follows the pacing and motion rules I have written down, then checks its own rendered frames before I look at them. I review the result, change whatever I would have done differently, and send it to you.

The agent cut fits videos that are mostly built rather than filmed, such as explainers, animated lessons, product ads and captioned shorts, and any series that has to look the same every time. The Appynox promo was rendered this way, from HTML to MP4. It works on filmed footage too: my Langtang trek reel, shot on a Sony a6700, was an agent cut.

The content factory

The content factory is for volume. I stop treating each video as a separate edit and build a batch pipeline: scripts go in, finished videos come out, and a log records which checks each render passed or failed. Failed renders go into a re-render queue, and a posting calendar decides what goes out and when. I have run it in batches of about thirty short clips.

It fits a product with many features to show, a weekly series, or a content calendar that needs filling. If you also want the posts around the videos handled, that is a separate service, AI social media automation, and the AI social media system case study shows it running 10 brand accounts with a person approving every post.

Choosing between them

So, a single video that is mostly filmed suits the classic cut, and one that is mostly built suits the agent cut. Many videos every month call for the factory. Some projects combine two, such as a classic-cut film with agent-cut versions for social. On the call I say which one I think fits, and why.

What you receive

Each video arrives in the sizes your platforms need. Reels, TikTok and YouTube Shorts want vertical 9:16, and YouTube and your own website want wide 16:9. For feed posts on Instagram, Facebook and LinkedIn I can also cut 4:5 or square 1:1, which fill more of a phone screen than a wide video does.

Captions come in two forms. Burned-in captions are part of the picture, so they show on every platform and cannot be switched off. A separate caption file, usually SRT, can be uploaded where a platform accepts one, such as YouTube, and viewers can turn it on or off.

You can ask for versions too: a shorter cut, a clean copy without captions or text for your own reuse, or the same video in another size. To keep working on a video yourself, ask for the project files when you confirm the plan: the Premiere Pro or After Effects project for a classic cut, and the HTML composition for an agent cut.

How a project runs, from brief to approval

The steps are the same for one reel or a whole batch.

1. You start by telling me the purpose of the video and who will watch it, and you send whatever you have, whether that is raw footage or only an idea. Send files straight from the camera or phone where you can, since messaging apps compress them. Your logo, colours and fonts help the graphics match your brand, and so do one or two videos whose style you like.
2. Before any editing, I put the plan on one page: length, platforms, style, message and delivery date. You confirm it, and the first cut is judged against that page.
3. I edit from the plan and send you a first cut.
4. You watch it and write notes against timestamps, for example “1:05, hold the logo a little longer”, so I know exactly which moment you mean.
5. I make the changes your notes ask for. If a note changes the plan, such as a new length or one more platform, I say so, because it changes the work.
6. When you say yes, I export every size and send the files. Nothing is published before that.

Sound, music and captioned videos

I use licensed music only, so you are not risking a copyright claim on a video you have approved. The voice always sits on top: music is mixed under it and drops back when someone speaks, so a viewer on a phone speaker does not have to strain to hear a word.

For captioned videos, I build word-by-word captions from the transcript and read every one against the audio. If I am not sure how a name or a place is spelled, I ask you instead of guessing. Many people watch with the sound off, so in short-form video the captions are often the whole message.

How I disclose AI-generated people and scenes

Some of my work includes generated material: b-roll, a whole scene, sometimes a voice or a person who does not exist. It goes in only when it fills a gap and you have agreed to it, and I tell you which shots are generated. A speaker’s real voice stays on the track unless you ask for a generated one.

Where a generated person appears on screen, the video or its caption says so. The two AI creators I have made, Mila and Theo, are labelled as AI wherever they appear, and I hold client videos to the same rule. Some platforms also ask for a label when you upload realistic AI-generated content, and I will tell you when that applies to your file.

The tools I use

I edit in Premiere Pro and After Effects, which are standard programs, so another editor can open the project files if you ever want to continue the work elsewhere. A LUT library keeps the grade consistent from shot to shot, and for agent cuts I compose videos in HTML and render them to MP4.

AI tools do the transcripts and, when you want them, the generated scenes and voices. When I shoot footage myself, it is on a Sony a6700 or a drone.

What drives the cost of AI video editing services

I do not list prices, because two videos that look alike can be very different amounts of work. After a short call I send the plan in writing, with a quote. Six things move the quote most.

• A longer finished video has more to cut, caption, grade and mix.
• More raw footage takes longer to go through, even for the same short reel, and extra speakers or poor audio add to it.
• Plain captions are quick, while motion graphics designed around each line take design time, line by line.
• Each AI scene has to be generated, checked against the real footage and graded to match, and some do not survive that.
• Every extra version, whether a size, a length or a caption file, is another export and often another reframe.
• A steady monthly volume changes the method. I build a pipeline with checks instead of editing every video by hand, which is work up front that makes the later videos routine.

Two things keep a project simpler. Clean audio recorded close to the speaker saves cleanup, and choosing which platform comes first keeps the number of versions down.

Video editing for small businesses and nonprofits

Video editing for small businesses and nonprofits often starts with one long recording. That single interview can become a longer film for your website, a vertical reel and a few short clips, and fast transcripts and rough cuts make that practical.

If you are in Nepal, my AI video editing in Nepal page covers captions in Nepali, quotes in NPR and working over WhatsApp and Viber. For a production that needs more hands, Appynox, the agency I co-founded with two partners, can take it on as a team of three.

How to start

The quickest way is to book a 30-minute call. The booking page shows times in your own time zone. Bring what you have, whether that is raw footage or only an idea, plus where the video will run and a video whose style you like.

On the call I ask what the video is for and tell you which workflow fits, and afterwards you get the plan in writing, with a quote. If you would rather write first, email me a brief with a link to your footage.