The argument in one line.
An agentic video editor lets someone who can't operate After Effects describe a visual effect in plain language and get a real, tunable version built directly on their own footage instead of hand-keyframing it.
Read if. Skip if.
- You can describe what you want a shot to look like but don't know which layers, keyframes, or shaders would actually produce it.
- You make videos solo and want motion graphics or custom effects without learning a compositing tool like After Effects.
- You're deciding whether an AI video editor is worth trying and want to see an agent-built effect before signing up.
- You already work comfortably in After Effects or DaVinci Resolve and don't need a plain-language shortcut.
- You want a technical breakdown of how the underlying shader or prompt-to-effect pipeline actually works.
The full version, fast.
Luuk Alleman calls himself a builder, not a motion designer: he knows what a good effect looks like but not how to build one in After Effects or DaVinci. He tests InVideo's new agentic editor by creating an AI agent and describing a custom effect in plain language: pull the brightest highlights into soft light trails, leave the colors untouched, keep darker areas clean, apply it to every scene. The agent builds the shader directly on his footage and hands back five sliders to fine-tune manually. When he asks for motion graphics, the first pass is text-only; restating what he actually wants in plainer language gets a better second pass. The lesson: prompting replaces keyframing, but you still have to iterate.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
01 · An effect I could never build by hand
Cold open: the gap between knowing what an effect should look like and having no idea how to build it manually sets up the whole video.
02 · What InVideo Editor is
InVideo sponsors the video; InVideo Editor is introduced as an agentic video editor where AI agents take plain-language editing instructions on a browser-based timeline.
03 · My plain footage
Starting point is raw, unedited footage from a previous video, already restructured once by an agent, used as the test bed for the effect.
04 · Creating the agent and writing the prompt
An agent is created and given a detailed plain-language brief: pull the brightest highlights into soft light trails, don't change colors, don't generate anything, and expose named controls.
05 · Asking for controls
The agent confirms its defaults in plain language and returns five adjustable parameters: highlight threshold, softness, trail length, direction, and mix.
06 · The light-trail effect
The built effect is inspected clip by clip; a lamp visibly glows and trails in one clip but the effect is absent from the next, showing per-clip inconsistency.
07 · Putting it on every scene
A single instruction ('keep the darker areas clean, only the brightest light should trail, put on all scenes') propagates the tuned effect across the entire edit.
08 · Tuning it with the sliders
The five exposed controls are manually adjusted, softness, trail length, direction, and mix, to push the AI-built effect toward the intended look.
09 · Asking for motion graphics
A voice prompt asks the agent for text and motion graphics keyed to what's being said, with a specific brand color code, restrained to one graphic per point.
10 · First pass: only text
The agent's first attempt renders as plain text overlays only, missing the visual storytelling graphics that were actually intended.
11 · Fixing it in plain language
A more specific follow-up request, explaining what's happening on screen and explicitly saying 'not only text,' triggers a reworked, better second pass.
12 · My honest take
The finished effect is judged against footage and saved effects the creator never used himself, with an open question about how much editing still needs a traditional tool.
13 · Try it free
CTA: InVideo Editor is free to use, agents cost credits, and viewers are pointed to an affiliate link for up to 50% off annual plans.
Lines worth screenshotting.
- An AI video editor can build a custom shader effect on your own footage from a single plain-language prompt instead of requiring manual keyframing.
- Asking an agent to build an effect AND expose manual controls (threshold, softness, trail length, direction, mix) beats asking for a finished, non-editable result.
- The first AI pass on a vague request like 'some motion graphics' defaulted to plain text overlays instead of the visual storytelling that was actually wanted.
- Restating a request in more specific plain language, rather than manually fixing the output, is what corrected an underwhelming first pass.
- Explicitly telling the agent what NOT to change (colors, footage) kept it from over-applying the effect beyond what was asked.
- A single follow-up instruction ('put it on all scenes') propagated one tuned effect across an entire multi-clip edit instead of reapplying it clip by clip.
- Even after an AI built the effect, four of five exposed sliders still got manually tuned before the result was accepted.
- The agent's build step ran visibly on the actual timeline, letting the editor watch the change happen instead of trusting a black-box render.
- Judging AI-built effects against footage and folders of effects you'd already given up on is a much lower bar than judging against professional motion-design work.
How to get After Effects-style results by describing them, not keyframing them.
An agentic video editor can turn a plain-language description of an effect into a real, tunable result on your own footage, but only if you prompt for named controls and iterate on vague first passes.
- InVideo Editor is an agentic video editor: you hand its AI agents editing tasks in plain language and watch them execute directly on a browser-based timeline you can still edit.
- The agent doesn't just render a black-box result, it also builds manual controls for whatever it creates, so a plain-language request still ends in an editable parameter set.
- Starting from unedited, plain footage is a fair baseline test for an AI editor, since it isolates what the agent adds rather than what a human editor already fixed.
- An agent can also reorder and cut a raw recording on its own, handling structural editing before any visual effect work even begins.
- Setting up an agent starts with naming it and giving it a short introduction, the same way you'd brief a freelance editor before handing off work.
- A workable effect prompt names the visual outcome, pulling the brightest highlights into soft light trails, states the direction, and adds an explicit constraint: don't touch the colors, don't generate new footage.
- Asking for named, adjustable parameters up front turns a one-shot AI output into a tool you can keep tuning after the fact.
- Requesting explicit sliders instead of a finished look is what separates a reusable AI-built effect from a static render you can't adjust.
- The agent explained its own defaults in plain language back to the user before handing over the controls, which is what made the result trustworthy enough to accept.
- A single applied effect can behave differently per clip: the same shader visibly glowed on a lamp in one clip while doing nothing to a similar light source in the next.
- Watching the agent apply an effect to one clip at a time, rather than trusting an instant global change, is how you catch inconsistencies before they spread.
- Once an effect is dialed in on one clip, a single follow-up instruction can propagate the same settings across an entire multi-clip edit instead of reapplying it manually.
- Explicit negative constraints, keep the darker areas clean, only the brightest light should trail, kept the effect from bleeding into parts of the frame that should stay untouched.
- Even an AI-built effect benefits from manual fine-tuning afterward: softness, trail length, direction, and mix were all still hand-adjusted after the agent's first pass.
- Treating the agent's output as a strong first draft, not a final answer, is what let a generic effect end up looking intentional on its footage.
- Vague creative requests get vague results; the more specific the described intent, react to what's being said, don't cover the face, the closer the first pass lands.
- A follow-up voice or text prompt works the same way a second round of notes to a human editor would: restate what's missing rather than starting over.
- The agent's first interpretation of 'motion graphics' defaulted to the narrowest reading, plain text overlays, rather than the richer visual storytelling that was actually meant.
- A disappointing first AI pass is diagnostic information, not a dead end: it tells you exactly which part of the instruction needs to be made more specific.
- The fix for an underwhelming AI pass was a clearer plain-language re-request, not manual cleanup after the fact.
- Naming what the output should NOT be was as useful a piece of the second prompt as naming what it should be.
- Judging an AI-built result against your own low bar, effects you already owned but never used, is a more realistic test than judging against professional motion-design work.
- The open question left unanswered, how much of a finished edit still needs a traditional tool like After Effects, is the honest limit of a single quick demo, not a full review.
Terms worth knowing.
- Agentic video editor
- A video editor where an AI agent executes editing tasks described in plain language directly on the timeline, instead of requiring the user to manually set keyframes or layers.
- Light-trail shader
- A visual effect that pulls the brightest highlights in a shot into soft trailing streaks in one direction, while leaving the rest of the image unchanged.
- Highlight threshold
- A control that sets how bright a pixel has to be before an effect, like a light trail, gets applied to it.
Things they pointed at.
Lines you could clip.
“Do you recognize this? The feeling that you know what you want to create for your video, but actually doing that manually...becomes really hard.”
“I'm a builder, not a motion designer. I have a folder full of effects I've saved over the years and just never used.”
“I made the effect I imagined and I never opened After Effects because I don't know how that actually works.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
Luuk Alleman opens with the exact frustration every non-editor has felt: you know what you want your video to look like, but building it by hand in DaVinci or Premiere is out of reach. His fix this time is prompting an AI agent to build the effect for him.
Named ideas worth stealing.
Effect prompt structure
- Name the visual outcome
- State the direction/behavior
- Set an explicit constraint (what not to touch)
- Request named, adjustable controls
The working prompt for the light-trail effect named the look wanted, constrained what shouldn't change, and explicitly asked for tunable parameters rather than a finished render.
How they asked for the click.
“Try it on your own footage, use my InVideo link in the description. You get up to 50% off on eligible annual plans.”
Direct affiliate link with a discount code placed as the final chapter right after the honest-take section, paired with a like/subscribe ask.
Visual structure at a glance.
