How to edit AI video into a finished short film

The AI video editing workflow, start to finish

This is the final lesson of the free course Make a short film with AI: you have rendered shots and a soundtrack, and now you cut them into a film.

Editing AI footage is ordinary editing with a few extra failure modes. The order of work matters, because every step depends on the one before it. Color-grading a shot you later cut is wasted effort; mixing sound before the picture is locked means mixing twice.

1. 01

   Organize

   Name every file by shot number and take (S07_T2) and keep a sheet with the prompt, model and cost of each. You will need it for retakes and for disclosure questions later.

2. 02

   Assemble

   Drop the best take of each shot onto the timeline in script order, untrimmed. Watch it once.

3. 03

   Rough cut

   Trim heads and tails, remove shots that do not earn their place, and reorder if the story asks for it.

4. 04

   Fine cut

   Pacing, cutting on action, continuity fixes, and a list of retakes for the shots that still fail.

5. 05

   Picture lock, then sound and color

   Once the timing stops changing, lay in the soundtrack from the sound design lesson, match color, then add titles and captions.

6. 06

   Export and label

   One master, then a version per platform, each with its AI-content disclosure.

Where should this happen? PromptLab's Composition node stitches your Flow's video outputs end to end into one 1920×1080 clip and mixes audio tracks with per-track gain. That is a quick way to see the whole film assembled and to deliver a simple 16:9 cut. Frame-accurate trimming, color, titles and vertical versions belong in an editor such as DaVinci Resolve, Premiere Pro or CapCut, and most of this lesson assumes you have one open.

Trim heads and tails: where AI clips go wrong

A generated clip is not a shot. It is a few seconds of footage with, usually, a good stretch in the middle. The edges are where problems cluster. The first frames can stall or settle before motion really starts, and in an image-to-video clip the first frame is simply your still. The last second is where the camera drifts past the subject, an action loops back on itself or a detail starts to melt.

Start frame

→

pick the window, not the whole clip

Image promptFlux 2 Pro · 8 creditsCopy

Tortoiseshell aviator sunglasses floating above rippled desert sand at dusk, the lenses reflecting a pink and orange sky, long shadow, luxury eyewear campaign, hyper-detailed

Motion promptPixVerse V6 · 5 s clip · 57 creditsCopy

The sunglasses rotate slowly in mid-air, fine sand drifting below them, the sky colours shifting from pink to orange, product commercial

Here is what that looks like in the clips from this course, frame by frame:

The practical rule: generate longer than the shot you need. If the cut needs three seconds, render five and choose the best three. It costs a little more per clip and saves whole retakes. When the image-to-video first frame is your approved still, you can also use it as a match frame: cut from the still on screen to the moving clip and the transition is invisible.

Cut on action and pace the shots

The oldest editing rule is also the most useful one for AI footage: cut on action. Start a movement at the end of one shot and finish it in the next, and the eye follows the motion across the cut instead of noticing that the lighting, the lens or the model changed. A hand reaching, a head turning, a cup being set down: each gives you a cut point.

• Shorter is safer. There is no correct shot length, but the longer a generated shot holds, the more time the viewer has to spot a flaw. Hold long only on clips that survive repeated viewing.
• Vary the rhythm. A run of equal-length shots feels mechanical. Mix quick inserts with a longer establishing or emotional beat.
• Let sound lead. Start the next scene's audio a moment before its picture (a J-cut) or let a sound carry over the cut (an L-cut). Both smooth the join between separate generations.
• Cut on the music. If you chose a tempo that matches your shot lengths, cuts on the downbeat will feel deliberate.

To practice, here is an edit decision list for a ten-second vertical reel built from four PixVerse V6 clips (228 credits in total), with in and out points chosen by stepping through each clip:

Run a continuity check on AI footage

Continuity errors are the giveaway of an AI film, because each shot was generated without knowing what the previous one looked like. Watch the fine cut once just for continuity, ideally with someone who has not seen it before.

Our own tea-party clip shows how easily this happens. The prompt asked for a hare in a brocade waistcoat; the clip delivered a red brocade dress with a white apron. Fine for a single shot, but cut it next to a shot where the hare wears a waistcoat and the audience will notice. The fix is upstream: build start frames from the same character references, as covered in the consistent characters lesson, and re-render the shot rather than hoping the cut hides it.

Match color across clips from different models

Every model has a default look. Put three of them side by side and the differences are obvious: warm gold, cool pastel, low-key amber. In one film they need to read as one world.

1. Balance each clip first. Neutralize white balance and set exposure shot by shot, so the starting points are comparable.
2. Match to a hero shot. Pick the shot that best represents the film and match the others to it: skin tones, then shadows, then highlights.
3. Apply one look over everything. A single grade on an adjustment layer or the timeline ties the shots together. A light, consistent film grain helps too, because it gives every clip the same texture.
4. Grade gently. Heavy contrast and sharpening exaggerate compression and generation artifacts. If a clip needs a drastic grade to match, re-rendering it with a better start frame is often faster.

The cheapest color matching happens before you render: the same style words, light and palette in every prompt, and start frames built from the same style references.

Titles, captions, upscaling and frame rate

Titles and captions. Add them in the editor, not in the generation prompt; video models still garble text. Burn captions into social cuts and upload a caption file for YouTube. Keep all text away from the edges of vertical frames, where the app interface sits.

Frame sizes. Expect mixed footage. The ten clips in this course all arrived at 24 fps, but in five different frame sizes, from 1280×720 to 720×1280. The three image-to-video clips came back in the shape of their start frames (roughly 4:5 and 4:3), not 16:9. Set your timeline to the delivery format first, then decide per clip whether to crop, scale or reframe. That is why the planning lesson fixes the aspect ratio before any start frame is made.

Frame rate. Keep the timeline at the rate your clips were generated at, usually 24 fps for cinematic footage, and export at the same rate. Converting to 30 or 60 fps with frame interpolation can add warping around fast motion, so check every converted shot carefully.

Upscaling. Several clips in this course came back at 720p. If you need 1080p or 4K, upscale the final, trimmed selects only, with your editor's super-resolution feature or a dedicated video upscaler. Upscaling sharpens detail and artifacts alike, so review the result at full size.

Export settings for YouTube, TikTok and Reels

Export one high-quality master first, then a version per platform. YouTube publishes its recommended upload settings; for TikTok and Instagram, the values below are the commonly used ones. Check each platform's help pages before a launch, because limits change.

For audio, deliver the loudness you mixed to in the previous lesson (around -14 LUFS integrated). Watch each export on a phone before posting.

How to label AI video on YouTube, TikTok and Meta

All three major platforms ask creators to disclose realistic AI-generated content. The rules as of October 2026:

• YouTube requires you to disclose realistic altered or synthetic content in YouTube Studio: a real person appearing to say or do something they did not, altered footage of real events or places, or realistic scenes that never happened. Since a May 2026 update, the label sits under the player on long-form videos and as an overlay on Shorts, and YouTube can apply it automatically when its systems detect significant photorealistic AI. Clearly unrealistic or animated content gets its disclosure in the expanded description instead.
• TikTok asks creators to label realistic AI-generated content with its AI-generated label, and can label uploads automatically when they carry C2PA Content Credentials.
• Facebook and Instagram require disclosure of photorealistic video or realistic audio that was digitally created or altered, shown as an "AI info" label, and Meta can add the label itself when it detects industry-standard signals.

A surreal tea party may not strictly need a label; a realistic AI ad usually does. When in doubt, label it: it costs nothing, and platforms can penalize realistic content that should have been disclosed. Ads have their own rules on each platform, so check the advertising policies too.

Final checklist before you publish

1. 01

   Watch it once without stopping

   Note every moment your attention slipped. Those are pacing problems.

2. 02

   Step through every cut

   Look for a stalled first frame, a melting last frame, a continuity break or a flash frame.

3. 03

   Listen with your eyes closed

   Every word understandable, no music fighting the voice, no ambience holes at the cuts.

4. 04

   Check text at 100%

   Titles, captions and any on-screen signs spelled correctly and inside the safe area.

5. 05

   Export, then watch the export on a phone

   Right frame, right length, captions readable, loudness consistent with other videos.

6. 06

   Disclose and credit

   Platform AI label where it applies, and the tools listed in the description.

That is the whole course: from a one-line idea to a planned, rendered, scored and edited short film. Go back to the course page to revisit any lesson, or browse the guides on prompting, product shots and social video for your next project.

Questions people ask