Ultimate guide to editing videos with AI https://x.com/i/article/2111044910203973632
50+ videos in 7 days. Zero editing experience. My complete AI video editing playbook.
Last week, I made 50+ videos with Opus 5.5, and I have ZERO video editing experience. I've never spent any time learning motion design or even installed Premiere Pro.
I'm someone who has spent time at the intersection of product and growth, and video editing has always been something I've wanted to do but never really had the time or skills for.
But what I learnt over the last week is that you don't really need to know how to edit anymore. You just need to know how to brief an editor, and you need taste.
I spent the week experimenting with different types of videos, figuring out what works, what doesn't, and how to get better outputs with every iteration.
This article is everything I've learnt, in order, along with the exact prompts I've been using that you can use as inspiration to build on.
TL;DR: I've been using Cardboard + Opus 5.5 as my video editor. I almost always give it references (form Instagram, X, Pinterest, and YouTube)
Why video, and why now?
The world is getting more performative. Everyone is on stage now. creators, founders, brands, even your co-workers.
And, in that world, videos/motion gives people a reason to stop scrolling in ways text and static images cant.
Whatever your goal, attention is the thing you’re after:
- If you're a creator, video gets you social attention.
- If you run a business, video helps people discover and remember you.
- If you work in growth, video is one of the cheapest ways to get someone to notice your product.
- If you've just come back from a trip, a good edit lets you relive the whole experience.
The biggest blocker has always been skill. Making a good video meant learning to edit or hiring someone who could, often taking days or weeks.
That barrier is disappearing fast.
Attention is still scarce, but making good videos is getting easier. And people who can make great videos quickly have a real edge.
My setup
I've tried a bunch of different models, and one thing I've realised is that every model has its own strengths and does things a little differently.
For video, Opus 5.5 has been my favourite by a wide margin. It's particularly good at motion graphics and understanding footage. You can show it a video, and it can break down what's happening, how things move, and why certain parts work so well.
(Though this space moves so fast that my recommendation might be completely different next week.)
And that second part is really important. If the model can't understand what makes a video good, it's going to have a hard time making one.
The other half of my setup is Cardboard, which gives Opus the ability to actually edit videos. It can look at footage, study references, and make changes directly inside an editor.
That's basically my setup. Opus figures out what to do, and Cardboard gives it the tools to do it.
Treat Cardboard x Opus like your actual video editor
The biggest learning from making 50+ videos is that you should work with AI exactly how you'd work with a human editor.
Dump all your raw clips, explain what you're trying to make, and give it as much context as possible. The more you tell it, the less it has to guess. And as you keep working with it, it starts understanding your taste better.
A few things that made a HUGE difference for me:
1. Give it all the context you have, especially if you're making videos for your business
If you're making a product video, one of the best things you can do is give Opus access to your actual website.
With a browser harness, Claude can open your website live, go through different pages, understand what your product does, see your actual UI, and learn your brand's visual language.
This makes a LOT of difference compared to just describing your business in a prompt.
Tell it what your business does, who you're targeting, what you're trying to communicate, and roughly how you imagine the video.
You don't need a perfect script. Even a rough storyline is enough to get started.
Here's something I'd use:
2. Reference videos make a day/night difference
I cannot stress this enough. This is probably the single biggest unlock I've had.
Give it references for everything. Not just the overall video, but specific effects, transitions, how you want the video paced, camera movements, typography, and even how certain elements should animate.
Instead of asking it to make a cool travel edit, show it 2–3 edits you love and tell it what you like about them.
And don't just say "make it like this". Ask it to watch the reference first and break down how the video works, including the cuts, timings, camera movements, and transitions.
Then tell it what to keep and what to change.
Cardboard lets you paste X, Instagram and YouTube links directly, so sharing references becomes much easier.
Example of a video from HIMYM intro as a ref
Now, you must be thinking where i find references
See the last section for my list. the habit that matters: save good motion the moment you see it. a folder of 20 links beats any prompt.
3. ALWAYS try sourcing B-roll before generating it
This is something I learnt after a few bad outputs.
Good footage makes a TONNE of difference, and I've found that real footage often looks much better than generating every video clip from scratch.
If you need a drone shot of New York, try sourcing an actual drone shot first. If you're making a travel montage, look for real footage that matches the lighting, camera movement and overall vibe of your edit.
Give Claude the ability to search the web and source B-roll. Cardboard already has this built in, so you can ask it to find footage and bring it into the edit.
But generating assets is still incredibly useful. I usually think about it this way:
- For actual video footage, I prefer sourcing first.
- For illustrations, backgrounds and smaller visual elements, generation often works better because you have more control over the look.
- For product UI and logos, I always prefer real assets (the setup can recreate UI really well given it has the context on your website and your UI)
- For specific shots that don't exist or are difficult to find, generation is great.
The important part is not forcing the AI to generate everything when better assets already exist.
4. Thing about motion graphics that feel “Premium”
References help a LOT, but understanding what makes good motion look good is equally important.
I knew pretty much nothing about motion design when I started. But after studying a bunch of really good motion graphics and going through other people's workflows, I came across a few rules that made a noticeable difference in my outputs.
1. Arrive fast, land soft. Elements should move quickly at first and gradually slow down. A good starting point is covering 12–19% of the remaining distance per frame.
1. Keep some movement in the frame. Even held shots can have a subtle push-in of around 0.25% per frame.
1. Give animations enough time. Smaller movements around 0.3s, larger ones around 0.5–0.75s.
1. Stagger elements. Offset multiple animations by 2–4 frames instead of moving everything together.
1. Use directional motion blur. Blur should follow the movement and fade as things settle.
1. Let text be read. Give text enough time to sit still before moving it again.
1. One focal point per frame. Make it obvious where the viewer should look.
1. Cut with the music. Try landing cuts on the beat or 1–2 frames before it.
1. Use fewer generic fades. Try making elements transform into the next scene instead.
I've started adding these guidelines directly to my prompts:
These aren't strict rules, but they're useful starting points.
I've also found that giving feedback with actual numbers helps a lot. Instead of saying "this transition looks off", tell it "the card should stop 6 frames earlier and the next element should enter 3 frames later".
The less the model has to guess, the better.
5. Sound design makes an INSANE difference
This is something I completely underestimated initially. You can have great motion graphics, but without good sound design, the video still feels unfinished.
A few things that made a difference for me:
1. Pick your music before editing. Find a track that matches the vibe and build your edit around it. Time your biggest visual reveals with the drops, and try to make important cuts land on the beat.
1. Sync SFX with movement. A click should land exactly when a button is pressed, and a whoosh should peak when the movement is strongest. Even being a few frames off makes a difference.
1. Don't keep repeating the same SFX. AI has a tendency to reuse the same whoosh or impact for every transition, which gets extremely annoying. Ask it to vary the sounds based on the movement, and leave some transitions silent.
1. Use good-quality sounds. I prefer sourcing real SFX over generating generic ones. Cardboard makes this much easier because it already has a huge SFX library and can generate relevant effects on the go, so I don't have to manually hunt for sounds.
Here's the prompt I use:
The biggest thing I've learnt is that sound should feel like it's part of the animation, rather than something added afterwards. And you don't need to overdo it. Sometimes removing a few unnecessary sound effects makes the video much better.
Cardboard already has all of this baked-in, it has a huge library of SFX and it can generate sounds/music on the go to match your beats to the cut making the whole sound stitching experience a lot more easy.
6. Keep directing it, just like you would with an editor
Your first output is rarely going to be your best one.
The biggest mistake is expecting a perfect video from one prompt and giving up when it doesn't happen.
I usually give feedback in plain English, like "the intro feels too slow", "use the transition from the reference at 0:08", or "the product should be much more prominent here".
You don't need to know motion design terminology. But the more specific you are, the better the edits become.
And instead of remaking the entire video every time, ask it to fix the particular section you're unhappy with.
Over time, the agent also builds more context around your preferences, the references you like, and the kinds of edits you keep rejecting.
That's really how I've been working with it. Give it the context, dump the footage, share good references, and keep giving feedback as you would to an editor sitting next to you.
7. check trends, then describe as you go
Attention formats age fast. a style that wins this month looks tired next month. so before you start, ask the editor to look at what is working right now.
then pick one you like and use it as your reference from step 2.
after that, work the way you would talk to an editor sitting next to you. watch the cut and say what you think in plain words. "faster here." "more like that one." "the first second is boring."
you do not need the right jargon. you need to say what you saw and what you want instead. the editor translates it into the edit.
Example
Where to find references?
Build a folder of links. save anything with motion you admire the moment you see it. Cardboard reads the links, so you never need to download a thing.
Feel free to DM me if you need any help
If you make something with this, tag me. I would love to see it :)