AI video editor that hands back an editable timeline, not just an MP4. Edit it yourself, or let your agent drive it over MCP. Free tier, no card.
United Kingdom
Joined December 2011
- Tweets774
- Following268
- Followers70
- Likes2.2K
Pinned Tweet
you can now call your video editor.
live call on vidmoat: tell otto "15 sec teaser, with captions" and watch the edit happen while you talk.
hang up and it's all on a real timeline you can still tweak.
try it free: vidmoat.com
forty minutes down to under one minute is a great first pass. the part that matters next is how fast you can fix the one cut that's wrong. on an editable timeline it's a nudge, otherwise it's another prompt
Holy f*ck, Opus 5.5 just replaced After Effects and my entire editing job
I'm a video editor and motion designer with years of experience myself, and this one scared me
it takes a 40 min raw video and cuts it into a 58 sec edit. hook, cuts, captions, keyframes, sound. by itself
so I turned the whole thing into an interactive After Effects desk you can play with:
• 8 real stations: ingest, find the hook, cut, 9:16 reframe, captions, keyframes, sound, render
• watch the AE timeline build itself: layers, playhead, graph editor, auto-ducking
• break it on purpose: kill the silence cut, center-crop the face out, linear keyframes, drown the voice
want Claude to actually edit your videos? send it this skill:
> github.com/heygen-com/hyperf…
try it for free, link in the first reply, prompt inside.
flub a line while recording? stop, pause two full seconds, then say it again. on the waveform that gap is a big flat spot right before the retake, so finding the cut takes a second instead of scrubbing through everything
captions read best in short chunks that break where you'd naturally pause. two lines max, split at the end of a thought, not in the middle of one. if you have to rush to read it, it's cut wrong
end the video the second the point lands. no outro sentence, no "so yeah". the cut right after your best line almost always feels better than the one after the extra ten seconds
batch your edits. cut dead air on all three videos first, then do captions on all three, then export. switching tasks every few minutes is where the hours go, not the editing itself
got talking head footage sitting in your camera roll? i'm building vidmoat.com so you can chat "cut the dead air, add captions, make it vertical" and still open the timeline to fix what it got wrong. free tier to try (biased, i'm building it)
this is exactly why the rough cut matters. an agent or a script can do the assembly pass. the decision pass is yours, so the assembly needs to land somewhere you can still move things around
your first cut is a transcript with pictures.
everything in order, every sentence kept, every pause respected. that's assembly, not editing.
the second pass is where you actually decide.
what goes entirely, what moves earlier, what gets a beat of silence because it earned it. try cutting a quarter of the video and see what you miss.
usually it's nothing. nobody asks about the footage you deleted. they just feel the pace.
funny, but the real version of this is true. cut on the open question, not just the pause. a pause with nothing before it is dead air, a pause after a question is the part that holds people
the QA line is the honest part. if the agent works in an editable timeline, a false flag means fixing one clip. if it hands back a flat render, it means redoing the whole pass
Make full documentaries made with AI
AI wrote, researched, and sourced all the content
Built with @claudeai and @GeminiApp:
Opus 5.5:
- Research
- Editor in Davinci Resolve
- QA (lots of issues and false flags still)
- Script: Argon
- Voice: Gemini 3.8 Flash TTS (custom)
- Video QA: Gemini 1.5 Flash
- Music: Flow Music
Full video is about 90 minutes and took about 50% of a weekly Claude subscription ($200/m)
That's with a lot of automated checks that ended up being false flags
nice. the next thing you hit is when the first pass is almost right and you want to nudge one cut or one caption without rerunning everything. that's when an editable timeline pays off
the sameness comes from one shot renders. when the human still gets to make the last few choices on the timeline, like pacing, a trim, a moment of silence, that's what makes it look like theirs
ai video editing will create an explosion of similar-looking short form content...just like how vibe-coded landing pages all looked alike in the beginning.
don't get me wrong though - i'm all for using ai to edit videos, and i'm leaning into it more now. it saves a BUNCH of time for us creators and video editors, and enables us to create more.
but when i see trends like this, i often think: how can i be different?
tip: get inspiration from others, but make your art...your art. make it your own.
Vidmoat retweeted
New AI video tools from makers:
Vidmoat by @vidmoat: edit in chat
agentscouthq.com/tools/vidmo…
Vellm: photo-to-video
agentscouthq.com/tools/vellm…
Reference to Video: product ads
agentscouthq.com/tools/refer…
Motion Transfer: animate photos
agentscouthq.com/tools/motio…
#AIVideo #VideoEditing
ALT AgentScout card: New video tools on AgentScout. Tool pages for Vidmoat, Vellm, Reference to Video and Motion Transfer.
The editor job isn't disappearing, it's moving up a layer. The agent does the dead-air pass and the first assembly, you spend your hours on pacing and the cut nobody else would make. The new skill is reviewing fast: scrub the cut points, not the whole timeline.
I’m a pro editor and over the last three months I have transitioned to a project manager of AI video editing agents. It is scary how good these agents are.
Couple of my thoughts:
- at some point (for now) a pro editor needs to step in and make sure the fine details are excellent
- my taste is the differentiator
- editing has always been perceived as a major “time” investment, and I think my clients will continue to treat it as so for a long time
- since editing is very technical, I think that will scare a lot of people off from AI editing beyond simple edits (which overseas editors are already taking)
- until transferring large amounts of footage over the internet is solved, localized editors have a strong advantage
Many of my colleagues are terrified about AI, and I think it will take awhile for adoption. Thankfully I’ve been in the sandbox for awhile, so this transition has been pretty seamless.
the fastest way to tighten a talking head: cut every breath between sentences, then put back only the ones that sound like you're thinking. most of them aren't
save a new project version before every big change. "final_v3_actual" is a joke until a crash eats two hours and the only backup is from this morning
reviewing an ai edit? don't rewatch the whole thing. jump to each cut point, watch one second either side. that's where it breaks, and it takes minutes instead of the full runtime
messy timelines usually come from never cleaning up mid project. every hour or so: delete unused clips, label tracks, collapse what's done. future you will thank you
chat editing removes friction only if you can still see what it did. say the edit, then open the timeline and nudge the cut it got wrong. without that it's just a slot machine
The "Claude replaced my editor" videos are unwatchable for the same reason: nobody watched the cut. An agent is great at the boring first pass, dead air, filler, rough captions. Taste is still a human scrubbing it before render.
AI video editing.
YouTube is full of people claiming "Claude is replacing video editors".
But when you watch the video, you notice a problem:
EVERY single video is literally unwatchable.
Sure, it can look nice at first glance. But unless you understand video editing and spent stupid amount of time training the AI, the video is garbage.
The pacing is insane. Like Mr. Beast on coke (zero). You are constantly bombarded with stimulus and have no chance of soaking information in.
Don't get me wrong. I love using AI. BUT AI LACKS TASTE.
Taste is developed through experience and preferences (see the problem?).
That's why (at least for now) anyone who doesn't understand editing, will get a garbage output.
