Big Idea

The argument in one line.

AI video tools can now recreate nearly any visual effect a YouTuber wants using only public apps, but each tool trades precision for spectacle differently — generative video models look flashier while code-driven tools like Remotion get text and facts exactly right.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You make talking-head YouTube videos and want punchier intros, transitions, or motion graphics without hiring an editor or learning After Effects.
You're comparing AI video generators (Runway, Gemini, Seedance, Kling, Veo) and want to know which one is actually good at which specific effect.
You're curious how creators generate b-roll, logo reveals, or lower thirds using AI coding tools like Remotion, Codex, or Claude Code.

SKIP IF…

You're looking for a single best AI video tool recommendation — this is a tour of a dozen tools, each suited to a different effect.
You want fully AI-generated videos rather than AI effects layered onto real human footage.

TL;DR

The full version, fast.

Matt Wolfe breaks down every AI visual-effect trick in his own YouTube videos, using only public tools. His core technique: feed two still frames — before and after a change — into an AI generator's keyframe mode (Runway with Seedance, Kling, or Veo) to produce a seamless surreal transition, like bursting through a wall or teleporting cities; matching wardrobe and prompting the model not to make the subject speak sharpen the illusion. For b-roll, motion graphics, and text-accurate logo reveals, code-driven tools (Remotion via Codex or Claude Code, NotebookLM's video overviews) beat general AI video models on precision, while Runway's character script-to-video feature dramatizes written exchanges — like a leaked Sam Altman text thread — from one photo and recorded narration. He caps his own AI use at roughly 5% of any video to keep it honestly, visibly AI.

Free for members

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 →

Chapters

Where the time goes.

00:00 – 00:40

01 · Why AI VFX (And Not AI Slop)

States his thesis: he doesn't make full AI-generated videos, but uses AI for the fun, attention-grabbing effects viewers ask about most.

00:40 – 04:27

02 · The Runway Intro Formula

Exports two still frames from DaVinci Resolve (empty room, seated) and feeds them into Runway's Keyframe mode with Seedance 2.0 to generate a wall-burst transition, then blends it in with the Smooth Cut transition.

04:27 – 06:16

03 · Intro Variations & The Animorph Hack

Shows a claw-machine pickup variant and an animal-to-human morph experiment that AI models still handle poorly; generates an intermediate wolf image in ChatGPT to use as a starting keyframe instead.

06:16 – 09:11

04 · Location & Travel Transitions

Uses the same before/after keyframe trick to transition between two real shooting locations (home office to a Big Bear cabin), plus a wormhole/mic-toss transition made for a conference talk.

09:11 – 11:39

05 · Background VFX With Gemini

Uses Google's Gemini/Omni model to composite fictional elements — a Yeti walking by, Godzilla stomping through a shot, an explosion, smoke, weather change — into real, otherwise unedited footage.

11:39 – 14:25

06 · AI B-Roll & Text Highlight Animations

Generates screen-recording-style b-roll of a webpage scrolling and highlighting using text prompts in Claude Cowork (Opus 4.8) and Fable, comparing the two outputs.

14:25 – 15:28

07 · Stock Footage On Demand

Generates generic 'corporate stock footage' style clips (handshakes, money tosses) with the same video models, noting how well they nail this familiar style.

15:28 – 20:13

08 · Logo Reveals & Lower Thirds

Uses the Remotion best practices skill inside Codex/Claude Code to build a particle-reform logo, a paint-splatter logo, and text-accurate lower thirds, contrasting them against AI video model attempts that mis-spell the on-screen text.

20:13 – 25:11

09 · Motion Graphics & Animated Infographics

Builds a mock text-message exchange, thought bubbles, SVG and stock-chart explainer animations in Remotion, pulls explainer motion graphics from NotebookLM's video-overview feature, and stress-tests three tools on a specific San Diego-to-New York flight path animation.

25:11 – 27:08

10 · Animated Talking Heads

Uses Runway's character script-to-video feature with a single reference photo and his own recorded narration to dramatize the leaked Sam Altman / Mira Murati OpenAI text exchange.

27:08 – 29:37

11 · Final Thoughts

Explains his 95% human / 5% AI philosophy and why he deliberately keeps AI effects a little conspicuous rather than seamless, then closes with a CTA tied to his two weekly upload types.

Atomic Insights

Lines worth screenshotting.

Two exported stills — the frame before and after a change — fed into an AI generator's keyframe mode as start/end images can generate a seamless surreal transition, like a wall bursting open.
AI video models are still consistently bad at morphing an animal into a human, even the newest generation of models.
Wearing the exact same shirt in two different real locations sells an AI-generated scene transition far more than the animation itself does.
Explicitly prompting 'the man does not speak' prevents AI transition models from making a person mouth gibberish at the end of a generated clip.
Google's Omni/Gemini model is currently the best option for compositing a new fictional element — a Yeti, Godzilla, an explosion — into real footage without changing anything else in the shot.
Claude Cowork and Fable can generate usable b-roll of a screen-recorded article scroll, complete with zoom and highlighter animation, purely from a text prompt, no screen recording software needed.
On an identical prompt, Fable produced smoother highlight-animation b-roll than Claude Opus 4.8.
AI video generators reliably nail generic 'stock footage' style clips like handshakes and money tosses because that footage type is overrepresented in training data.
Remotion, driven through a Codex or Claude Code skill, gets on-screen text and logo animations correct every time, where general-purpose video models like Seedance frequently misspell names and titles.
The tradeoff between Remotion and AI video generators for lower thirds and logo reveals is precision versus flash: Remotion nails text accuracy but looks less cinematic; video models look better but need four or five prompt attempts to get text right.
NotebookLM's video-overview feature generates genuinely usable motion-graphic b-roll when fed source documents on a topic, not just slide decks.
None of the AI video generators tested, including Seedance, Google Omni, and Remotion, could accurately animate a specific point-to-point flight path; Remotion's code-driven arrow was the closest, everything else drifted geographically.
Runway's character script-to-video feature turns a single reference photo plus recorded narration audio into a talking video, useful for dramatizing text-message leaks without ever filming the person.
Capping AI usage at roughly 5% of a video's content is a deliberate strategy so an audience never has to wonder whether what they're watching is real.

Takeaway

Precision tools beat flashy ones for anything that has to be exactly right.

TOOL SELECTION

The best AI video effect for a job depends on whether the shot just needs to look real, in which case use a generative video model, or has to be exactly correct, in which case use a code-driven tool like Remotion.

01Why AI VFX (And Not AI Slop)

A creator can use AI for a small fraction of a video's content and still keep the whole video feeling human-made; the ratio, not the tool, is what determines whether something reads as slop.
Every AI effect discussed runs on tools that are publicly available right now, not a private production pipeline.

02The Runway Intro Formula

Exporting a 'before' still and an 'after' still from your editor and feeding both into an AI generator's keyframe mode produces a far more controlled transformation than a single freeform text prompt.
Seedance 2.0 currently outperforms Kling and Veo 3.1 for this specific before/after transformation style, according to hands-on comparison.
DaVinci Resolve's Smooth Cut transition can disguise the seam where a generated AI clip meets real footage, even when the AI clip's colors don't perfectly match.

03Intro Variations & The Animorph Hack

As of mid-2026, every major AI video model still struggles to convincingly morph an animal into a human; it's a reliable weak spot to know before planning an effect around it.
Generating an intermediate image in a still-image tool like ChatGPT to use as one of your two keyframes lets you sidestep a video model's weaknesses, turning a hard animal-to-human video morph into an easier human-to-human video generation plus a separate image edit.

04Location & Travel Transitions

A location transition sells itself on wardrobe continuity; wearing the identical shirt in both real locations does more work than the AI transition animation itself.
Explicitly prompting 'the man does not speak' at the end of a transition prompt prevents the model from generating gibberish lip-flap once the transformation completes.

05Background VFX With Gemini

Google's Gemini/Omni model is the current best option for adding a fictional element, such as a creature, an explosion, or weather, into real footage while leaving everything else in the shot untouched.
Telling the model explicitly not to react to the added element keeps the original footage's performance intact instead of forcing a re-shoot.

06AI B-Roll & Text Highlight Animations

Claude Cowork and Fable can each turn a plain-text prompt into a screen-recording-style b-roll clip, scrolling, zooming, and highlighting a specific paragraph on a real webpage, with no screen-recording software involved.
On an identical prompt, Fable produced a smoother, more natural highlight animation than Claude Opus 4.8, which had minor image artifacts.

07Stock Footage On Demand

Generic corporate stock-footage scenes are one of the easiest things for any current AI video model to nail, likely because that footage style is heavily represented in training data.

08Logo Reveals & Lower Thirds

Code-driven tools like Remotion, used through a Codex or Claude Code skill, get on-screen text exactly right every time, where general AI video models frequently misspell names and titles on lower thirds.
The tradeoff is precision versus spectacle: a Remotion-built lower third or logo reveal is fully accurate but visually plainer than what an AI video model produces when it happens to get the text right.
Expect to prompt an AI video model four or five times before it renders on-screen text correctly; budget for iteration, not a single generation.

09Motion Graphics & Animated Infographics

NotebookLM's video-overview feature can generate usable explainer-style motion graphics directly from source documents, not just slide decks or podcasts.
A narrow, specific task like animating a single flight path between two named cities exposes real gaps in current AI video models; none of the three tools tested got it fully accurate, and the code-driven Remotion version was the closest despite looking the least cinematic.
Mock text-message exchanges and simple explainer diagrams are reliable wins for Remotion-driven motion graphics.

10Animated Talking Heads

Runway's character script-to-video feature can animate a single reference photo speaking uploaded narration audio, making it possible to dramatize a written exchange like a leaked text thread without ever filming anyone.
Recording your own voice reading both sides of a dialogue and running each half through character script-to-video separately, then editing them together, produces a convincing back-and-forth conversation from a single tool.

11Final Thoughts

Capping AI usage at roughly 5% of a video's total content is a deliberate strategy to keep a channel's output feeling trustworthy as AI-generated content becomes harder to distinguish from real footage.
Purposely leaving AI effects a little obvious rather than seamlessly hiding them is itself an editorial choice, not a limitation; it reassures the audience about what they're watching.

Glossary

Terms worth knowing.

Keyframe mode: An AI video generator feature where you supply a start image and an end image and the model generates a video interpolating between the two.
Smooth Cut: A DaVinci Resolve video transition that blends two clips together to disguise what would otherwise be a hard, visible cut.
Delta Keyer: A tool inside DaVinci Resolve's Fusion page used to key out, or remove, a solid background color such as a green screen.
Character script-to-video: A Runway feature that animates a single reference photo of a person speaking uploaded audio or typed dialogue, producing a talking-head clip without filming anyone.
Video overview (NotebookLM): A Google NotebookLM feature that generates a short animated video summarizing the source documents loaded into a notebook.

Resources

Things they pointed at.

00:54toolRunway

00:54toolDaVinci Resolve

01:45toolSeedance 2.0

01:45toolKling

01:45toolGoogle Veo 3.1

05:52toolChatGPT

08:23toolLeonardo

09:27toolGoogle Gemini (Omni)

11:37toolClaude Cowork

13:34toolFable

15:30toolCodex

15:43toolRemotion (best practices skill)

21:47toolNotebookLM

25:26toolRunway character script-to-video

Quotables

Lines you could clip.

27:17

“I still prefer to make 95% human videos and use AI about 5% of the time.”

clean philosophy soundbite, no setup needed→ newsletter pull-quote↗ Tweet quote

28:19

“So when I use AI, I almost, like, purposely show off that it is AI.”

counterintuitive stance stated in one breath→ IG reel cold open↗ Tweet quote

The Script

Word for word.

Read-along

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.

metaphor

00:00One of the easiest ways to make videos stand out is to use visual effects. If you watch my Friday news videos, they always start with some sort of like random visual effect, and it's probably the thing that I get the absolute most questions questions and comments about. Not to mention, they're just super fun to make.

00:17Now before I get into this, I don't like to make full on AI generated videos. I still like humans to create stuff. So here's a bunch of cool effects that you can make with AI tools, and all of these tools are actually publicly available right now.

00:33Some of the effects you can do in different tools and we'll get different results with different tools, But let's just get into it and start playing around with some of them. Let's start with some of those intros that you see in my videos, you know, stuff like this. And stuff like this.

00:54I've done a ton of intros that follow that same formula, and most of them are done inside of runway. You can just see I've done a lot of them now. But I actually start in DaVinci Resolve.

01:05So if I open up DaVinci Resolve here, I actually start by recording my video, like, off of my seat. So here's an example that I'll pull in from my most recent news video. And you'll notice in the very beginning of the video, I'm not on camera.

01:19I actually hit record by sort of like reaching in, pressing the button while I'm not on screen so that I have a frame of what my room looks like without me sitting in it yet. If I scrub forward a little bit, you'll see I sit down, position myself, and then start talking.

01:35So I need two still frames from this video to get the effect. So I'll go back to the beginning before I'm on camera, select my timeline, come up to file export in DaVinci Resolve, and then take current frame as still.

01:47I'll usually save this as like frame one, and then I come to right before I start talking. So let me zoom in even more on my timeline to right before I start talking here. So the transition feels more seamless right into me talking and I will export another frame here.

02:01So file export, current frame is still, and then frame two. Once I have those two frames, I then jump into runway, click on custom on the left side, and then select key frame. And then it gives you the option to upload the first video frame and the last video frame.

02:17So for our first video frame, we're gonna upload frame one. And for our last video frame, we'll upload frame two. So you could see me without the chair, me in the chair.

02:25I'll then give it a simple prompt like the man burst through the back wall and then sits down in the chair. I like to use Seed Dance two point o, but there are other models you can use here. The cling model's actually pretty good and Google's v o 3.1 model's also pretty good for this kind of thing.

02:40But Cdance honestly, like, really kinda tops them all right now. So I'll use Cdance two point o, and then you could decide whether or not you wanted to output with audio on or off. I always output with audio on because if I don't like the audio, I could just mute it inside of DaVinci Resolve.

02:57Also, can use Seed Dance and most of these models in, like, almost any video generator. So if you're using runway, you can do it. I believe Korea has all of these.

03:06I believe Higgs Field has all of these. I believe Leonardo has most of these. I just mostly use runway because at some point like a year and a half ago, runway gave me a ton of credits to play around with this tool, and I still have a lot of those credits.

03:21So that's the main reason I use runway, but feel free to use whatever tool has these models in it. We'll click generate and when it's done, I get a video that looks like this.

03:35Now, typically, I would reroll it and have it do it again so it doesn't do this little like hard cut from zoomed in on me to zoomed out. But just for demo purposes, I'll show you the next step here. I'll go ahead and download this.

03:47Jump back into DaVinci Resolve. I'll cut out everything before I start talking here. Drag this in and it should be pretty close.

03:55The colors change a little bit, but it's not too bad. And then in DaVinci, they have a effect. So if I open up my effects box and then go to video transitions, they have one called smooth cut, which kinda cleans up the transition a little bit so it's less noticeable.

04:10So I'll zoom in here. I'll take a couple frames off of this side, take a couple frames off of this side here, and then I'll use this smooth cut transition and you'll notice that it kinda tries to blend it a little bit better so the transition is less noticeable.

04:25And that's how I do most of those intros. There's a few variations that I've tried. Like, for example, I did this one where a claw machine comes and picks me up here.

04:34It lifts me up and then sets me in my chair, and then I start talking.

04:39The only difference between the other one is I didn't start off frame. I hit record, walked over, sat in the corner of my room, got up, moved to my chair, and then just spliced out everything in between. So frame one is just me sitting in my corner, and then the final frame is me sitting at my chair about to talk.

04:56And here's the exact prompt I used. If you want, feel free to pause and read the exact prompt. Now, another variation was this sort of like anamorph one that I was trying to do, and this is something I still find all of these AI models are really bad at is starting from an animal and having it like morph into a human.

05:14You could actually see as I scrub through these, none of them are that great at morphing from animal to human. I think that might have been the one I ended up using right here.

05:27I like this one here where my head's there first, and then my body grows out of it. I did eventually find one that I think worked decently well.

05:36And for this one, what I did was I actually started the way I would normally start, me not in a chair at all, walk in, sit down, and then I still exported the first frame without me in it and also the frame right before I start speaking. I then went over to chat GPT, uploaded the image of me not in my chair yet and just said make a wolf sitting in the chair, make sure to keep the same 16 by nine aspect ratio, and it gave me this image.

06:01And then I used this image as my starting frame. So you could see Wolf as the starting frame, me as the finishing frame, and then here was my prompt right here if you wanna pause and look at the exact prompt. But that's pretty much how I'm making all of those intros.

06:16Here's another thing I love to do if I know I'm gonna be shooting in a different location is I like to do transitions to the new location. So for example, a news video I did a couple weeks ago started like this.

06:28In fact, for this one, let's change up the scenery a little bit.

06:36So there wasn't a ton of AI news. So for this one, I knew I was gonna be up in Big Bear at a cabin recording. So before I left for Big Bear, I actually recorded the first, you know, five seconds of this in my office saying, here's the AI news.

06:50Let's have a change of scenery. So it was just a little bit of foresight knowing that I was gonna be recording in another location. You can see here's my frames, the very last shot from me at my house, and then the very first frame of me talking out in Big Bear.

07:05And I just said the camera rotates above the head of the man on the first frame and then rotates around him to reveal the second image. It does it all in one smooth motion, and the man does not speak. And I put and the man does not speak at the end because oftentimes these models will do the transition and then just make me speak like gibberish at the end.

07:24So I need to often prompt it, like, make sure the person doesn't speak. And this was the output of that. It took it and it just transferred between the two scenes.

07:32You can see it's actually a different microphone. I'm holding this microphone here, and in this one, I'm holding this style microphone, and it happened so quick.

07:39You barely even noticed that the microphone changed. But I also knew to wear the same shirt. So I was wearing the same shirt before I left in my studio.

07:47And when I was recording out in the cabin, I made sure to wear that exact same shirt when I was recording just to kinda sell the transition. When I found out I was speaking at NAB, they asked me to make some video samples of some of the cool, like, animations I've done and transitions. And I made one, and it never ended up getting shown.

08:05So here's actually that video that I was going to show at NAB that I'm actually showing for the first time anywhere. I'm currently in San Diego, and I need to be out in Vegas pretty quickly here. So I've got a trick.

08:17All I gotta do is this.

08:21And I'm here. That was easy. So it's always fun to make little transitions like that when you're going from one place to the other.

08:28That one was done in exactly the same way I just showed you with this one. The only real difference with this one is that I used the cling video model instead of seed dance, and I did it inside of Leonardo instead of runway because I had credits over there when I made that one. But the process exactly the same.

08:46And here's another variation of that same effect. Toss the mic in the air. It sort of slow motion spins and then falls into my hand in the new location.

08:53And here's an example of me getting sucked through a wormhole and spit out the other side.

09:02Here's another one.

09:07So, yeah. Transitions are a lot of fun with AI. It's also fun to make things happen in your video that didn't really happen in real life.

09:17Like, the majority of the video is something that really happened, but something else going around you didn't happen. For example, I did this intro in one of my videos where I'm talking and a Yeti walks by in the background and I don't acknowledge it or even talk about it at all.

09:32It just kinda happens. For these kinds of effects, the best tool right now is probably Google's Omni. So to do this, I would record the first, you know, ten seconds of me talking in my video here, and then take that ten second clip of me talking in the video, upload that video here into Google's Gemini, and put a prompt like a Yeti walks by in the background behind the man.

09:55And then you get a clip like this where I'm talking in the foreground and the Yeti just walks by in the background, and I splice it into the front of my video just like I do any other video. This was something else I was playing around with in a video from, like, last week sometime where I had this clip from Google IO of this band playing.

10:14And I thought it would be fun to have Godzilla come in and stomp around in the scene. So I took this video clip, tossed it into Gemini, and said make Godzilla enter the scene and start stomping around. And this was the output from that one.

10:35You can also add little effects to the back of your video. Like, example, I took one of the intros to one of my videos here, and I said, add an explosion behind me in the video. Don't change anything else.

10:44Make it so I don't even react or notice that happened. Now, I have to add that extra bit to the prompt because otherwise, like, I'd turn around and go and act like I noticed it happened, and I want in this particular case to just be like nothing's going on. So for this one, here's an explosion in the background.

11:01Here's the AI news you probably missed this week. So for this one, I took that same exact video clip and said make my head smoking. And so this one, do my normal intro and there's just smoke coming off of my head.

11:15And then here's another one that started as a clip from Google IO where everybody's just sort of walking in. It's clearly a bright sunny day. Everything looks nice.

11:23The weather's amazing. And I said take that same video, but make it cloudy and rainy day. But now you can see it's the same location, but it's a cloudy rainy day and everybody's wearing jackets and holding umbrellas and it totally changed the environment, but the location's exactly the same.

11:39Now let's talk about generating b roll for your videos because this is something that anybody who makes videos knows they need from time to time, and sometimes you can't find the exact footage you need on stock video sites. For example, a channel like mine where I share news and articles and curate things, I like to be able to highlight things on a web page and make it look nice.

11:58For that, I actually use a surprising tool that gets the job done, and that's Claude Cowork. So check this out. This is my Claude account here, my Claude Cowork app.

12:08And for this, I'm using Opus 4.8. Now you can use Fable for this, and Fable will probably do a little bit better of a job than Opus, but Opus honestly does the trick. Now, if you scroll all the way up, you could see what my prompt here is.

12:21Create a video for me. The video starts with a screenshot of this page. I linked it up to this news article about Claude Cowork is coming to mobile and web, and I want it to zoom down to a paragraph down here that says your work follows you.

12:36You started task at your desk, check out from your phone, pick up the finished output anywhere. I want an animation that starts on this article, scrolls down to that text, and then highlights the text for me. So I said, create the video for me.

12:48The video starts with a screenshot of this page. It scrolls down to the text that says, your work follows you, etcetera, etcetera. It zooms in on the page slightly and then uses a highlighter animation to highlight the text.

13:00And then what it does is it actually goes and it screenshots that website, and you can see it wrote a bunch of code for me and did a bunch of tasks and used a bunch of tools. And once it was all done, it literally spit out an m p four video for me. So if I click download and open here, we can see this is the video that it created.

13:18There's a little bit of a wonkiness in the image here, but it's not horrible. But you can see I press play and it scrolls down the video for me, zooms in on that section, and then highlights it. And that's the whole video.

13:31But that's usable b roll if I'm trying to explain some news and have a nice, like, little highlight animation. Now when I used Fable to do the same prompt, it actually did even better.

13:41So here's some examples of that. This time, I used the redeploying Fable one, and you can see the highlight looks a little more natural. And when it animates to different highlight areas, it just feels a little bit smoother, but it's still the same process.

13:56It's still used code behind the screen and used screenshots. You can also do things like this where it zooms in and points an arrow at a specific piece of text just like this, or you can do it where it zooms to an area and then highlights that area like this. But you could do those sort of, like, text b roll animations for your videos literally using Claude code.

14:18It would probably work in codex as well, but I tested it in Claude code or I guess Claude co work for this example. You can also find that cheesy looking corporate b roll style video that you'd find on a stock video websites if you don't wanna use a stock video website. You can see my prompt here, stock footage of a man in a business suit shaking hands with a woman in a business suit while standing in a conference room.

14:39And that's exactly what you get. And here's another example of that exact same prompt. I mean, it looks pretty spot on to what you would get out of one of those b roll websites.

14:50Here's one where I have it with a stock b roll footage. A man sits on a pile of money and scoops it from the ground and excitedly tosses it in the air. Like, come on.

15:00That's exactly what you'd find on a b roll website. Here's another version of that same prompt. If you need some decent looking b roll, but don't wanna use a stock video site, you wanna dial it in for exactly what you want, well, you can use runway and something like seed dance for that or, you know, Leonardo and cling or v o three.

15:19Pretty much any of them these days will make this style stock footage probably because there's a ton of it trained into the training data already. Another thing you can use AI tools for are text animations and logo reveals.

15:33So the way I like to do these is I actually like to do these inside of Codex, but this also works just as well in Claude code and you install a skill called remotion best practices. Now, if you haven't seen my video about skills, definitely check that video out. This is one of the skills we kind of deep dive on.

15:48But you install this skill inside of Codex or inside of Claude Code. To do that, you open up Codex, go to plugins, go to skills up here, search up Remotion, install Remotion best practices, and you can use this skill.

16:03So I uploaded my future tools logo here into Codex, and I tagged Remotion best practices. So I just type slash and then type Remotion, and then I'll select Remotion best practices here, and then I would enter my prompt next to that.

16:17There's an explosion of particles, and then the particles reform to become the future tools logo, which, you know, is the logo that I uploaded with it. And here's that animation that it created for me. All the particles explode, and then they form directly into that future tools logo for me.

16:32So that's something that you can use like right at the intro of a video if you want. Here's another version of the particle logo reveal using my color version of my logo, and here's what this one looks like. You could see it's now colorful, but pretty much doing the same thing.

16:45Here's another one I did where I used the emotion skill. Start from a blank screen, paint splatters all over the screen in the colors of my avatar. The paint drips and morphs into the avatar image, and I uploaded my avatar.

16:57And here's the video that that generated. You can see all of the colors splat all over the screen, and then they sort of rearrange and form into my avatar. You can also use AI to create lower thirds.

17:06Remotion does this really really well. You can also use the video generators to do that. Albeit, it doesn't always work quite as well.

17:13Here's some examples of using Cdance where it spells my name wrong and then says CEO of Busy Bench, where if you look at my prompt here, it's supposed to create a lower thirds saying my name correctly and CEO of Busy Bench. But we can see it did create a sort of green screen background here that if we were to download this, pull it into DaVinci Resolve here, I can actually open this in the fusion editor here.

17:40I like to use the delta keyer. So if I add the delta keyer in here, come select my background color as green by dragging this over right here. Now, I watch my video, you could see this animate onto the screen like that.

17:54And then I can adjust the position like this, position it over here like this, maybe make it a little smaller. And now, I've got a little animation. It's kind of a bummer that it didn't do the text right because it actually looks pretty good.

18:06Here's another example that's even worse, CEO of Busty Bench, but you get the idea. You can prompt it to create a green screen lower third graphic, tell it what you want to say on the lower third, and it might create it.

18:19Most likely, you're gonna have to prompt it like four or five times before it gets it right. Unless you're just okay being the CEO of Busty Bench. Don't look that up.

18:28Now, if I switch the model over to v o 3.1, it actually does a much better job with the text. So, you know, different models are better for different things here.

18:38If I pull this one into DaVinci Resolve, there you go. You got a pretty nice looking animated lower third that is just better than what you're gonna get directly inside of something like DaVinci. Now, the other option to get this kind of thing is to go back to remotion.

18:52And remotion, it's gonna dial in the text perfectly every time because it's using essentially code to make these videos, but it's not gonna look quite as epic with the animations. So you can see I used Remotion, gave it the exact same prompt inside of Remotion, and well, here's the version that Remotion made.

19:10You could see not quite as flashy, but it nailed the text and everything I asked for and it's got a perfect green screen background. So adding our delta keyer on that one, this one looks pretty perfect as well.

19:22If you wanna animate text that comes in behind somebody, these days it's not that hard to do it in something like DaVinci Resolve, but you can ask Gemini to do it. Here's my prompt for that, and it made a video where the future tools text animates in behind me instead of in front of me. Maybe I wanna do a fun little like thought bubble.

19:39I can use a prompt like this one right here, and you get a video where I'm sitting here, and now I'm thinking everyone should really subscribe after this one. So, yeah, AI is great for your logo reveals, animating images onto the screen, putting text behind you, putting thought bubbles, adding lower thirds.

19:56You can do all of that with AI. It's just sometimes it might take you up more than one prompt to get it looking the way you want. But if you really don't know your way around a video editor like DaVinci Resolve or Adobe Premiere, this is a pretty easy way to get what you're looking for without needing to go and hire a separate editor.

20:13Another thing that AI is getting better and better at is doing like these animated motion graphics. Like, the best way I can sort of describe them is the type of things you might see in like a BuzzFeed or Cleo Abrams or Johnny Harris style video.

20:26Now, they're not as good as if you were to hire somebody and do them in After Effects yourself, but you can get some fairly decent style motion graphics. For example, here's one that I did with Remotion inside of Codex where I basically asked it to create like a mock text message exchange, and here's what it gave me.

20:45We can see it looks like an iPhone. It says, have you subscribed to future tools yet? No.

20:49Why? Because it's literally how people are keeping up with AI news these days. Oh, okay.

20:53That's how they're doing it. Just sub. Thanks.

20:55Like, this exact text exchange would probably work decent as an ad to get people to go subscribe to the newsletter. If you're telling a story about a text exchange that you had on a video, well, you can use this kind of animation for that.

21:08If you wanna see the exact prompt I used, feel free to pause and and take a peek, but, uh, it works well for that kind of thing. Another example is if you're trying to explain a complex topic and you need a sort of visual to go along with that explanation. Like, when I gave it this prompt here to explain what SVG images are and how SVG actually uses code, it generated this video here showing that the code is what turns into the image.

21:31Or this one here, we're explaining how my Gary Busey leaderboard works. I mean, it's a pretty decent little animation, especially when you know the context of the rest of my video.

21:42Now, another good way to get some of these motion graphic type things is to use Notebook LM. You can actually feed it a bunch of documents on the topic that you're trying to explain. And inside of Notebook LM, you can do things like create audio podcasts and slide decks and things like that.

21:57But there's also this video overview feature here where you can have it create cinematic overviews or explainers or shorts. And in this case, with this sort of birds aren't real one that I've played with in the past, you could see that it made this animation for me that looks pretty good. Now, I can scroll through this and there's parts of this animation that would make for a great video on this topic.

22:18Like, here's a funny chart that it made about drones going up as the amount of birds are going down. Here's one that explains that the birds get their energy by sitting on telephone wires. And here's like a really cool animation of a guy holding a bird.

22:32Like, if I was making a legit video about this topic, the animations that NotebookLM creates around the sources that you feed it are actually really good animations that are very usable for b roll. Jumping back to Codex here, another really useful animation style, especially if you're doing an explainer video or like a mini doc or something like that.

22:54You might wanna show some graphs. Here's one that shows NVIDIA's stock chart growth that, again, another really solid animation that can be used if you're making a video about NVIDIA's growth.

23:06Now one thing that's been sort of hit or miss, and I haven't really got it to work that well, is I wanna get this visual of, like, a plane going from one city to another that might work really well for like a travel video, except all of the video generators are kind of pretty bad at it. So here's what C Dance does. You can see that here's my plane starting in the state of Sklincke, and it flies over e e over the state of Nara and Dirk and Bradu and lands somewhere near Navork when I told it to go from San Diego to New York.

23:39And, yeah, you you you could just see. It's it's not great. Here's my second attempt, Maybe a little bit better.

23:44Like, San Diego is not here. It's way down here. But you could see this video.

23:50It sort of flies over the state, sort of weaves up, and then overshoots New York a little bit. I mean, better than the last one, but still not super accurate. I tried to get the same thing to work over in Gemini using the Omni model, and we got a similar issue.

24:07It did start in San Diego. It flies across, and it does seem to land kinda where New York is. My main complaint about this one is, like, halfway through the video, you could see it sort of, like, fades out and sort of goes backwards.

24:22Right? So right here, we're over, like, Nebraska area.

24:25Right here, we're over Illinois. But then all of a sudden, we're back over Arizona again.

24:31So it sort of has this weird blip in the middle of it. But the starting and ending are a lot more accurate than the other options I got. But the best one does seem to be using remotion in codex.

24:42But again, it's just not quite as cool looking. Here's what that one generated. You could see it's just an arrow and it flies from San Diego to New York.

24:51It kinda takes this weird upper arc instead of just kinda shooting it across in a straight line, but it's definitely the most sort of accurate city to city of them all. So depending on what you're going for, try some options. If you're gonna use one of the AI models and not, you know, a code model like Remotion, you might have to prompt it a few times before you get what you're looking for.

25:12Okay. There's one last video type I wanna talk about because it was another little trick that I used in one of my videos, and a lot of people ask me, how'd you do that? So in this video that I released back on May 8, I sort of broke down this text exchange between Sam Altman and Mira Moradi, and here's what that looked like.

25:27Sam asked, can you indicate directionally good or bad? Satya and others are anxious. Directionally, very bad.

25:34Okay. Can you wrap up soon? Lots of pressure from Microsoft for an update, Sam, this is very bad.

25:39And, again, I got a lot of questions about how did you actually do that. So what I did was I read the text messages out loud. I just used DaVinci Resolve's audio recorder and just recorded myself reading all of the text exchange.

25:53Inside of Runway, they have this feature called character script to video. And if I select this feature, I can upload an image. So I uploaded this image of Sam Altman that I found.

26:03I think it was actually an AI generated image of Sam because it doesn't really look totally like him. And then under character script, you have the option to type a script or upload audio.

26:13So I uploaded the audio of me speaking out this exchange. I did the same thing for Mira Moradi. The output looked like this.

26:20It was an entire video of one person saying both the lines. Sam asked, can you indicate directionally good or bad? Satya and others are anxious.

26:29Mira said, directionally very bad. So both videos looked just like that with the entire dialogue of both. I pulled them into DaVinci, my editing software, and I just edit them having the conversation back and forth.

26:42I find little stuff like that keeps people engaged in the videos longer. If it was just a video of me just rambling and talking about the news for the entire video, people tend to check out and tune out.

26:53So when I do stuff like this, it tends to get people to stick around and watch videos longer because they're always waiting to see like what little fun Easter eggs or toys or tools am I gonna throw in to make the video more entertaining, more interesting to watch.

27:08And those are some of the ways that I use AI to create visual effects inside of my videos to keep them exciting and intriguing and fun and entertaining. And also, let's be honest, I'm doing it to entertain myself. Like, I have a lot of fun figuring out what tools to use and how to get these different things to make the animations that I'm looking to get them to make, and it's a really enjoyable process.

27:30Now, I know this video probably wasn't for everybody. There's a lot of people that come to my channel and tune in for the AI news that I put out every Friday, and well, this wasn't that. But there's also a lot of people that probably wanna learn how to do some really fun little things with AI video tools to improve their videos.

27:48That's who I made this video for. I mean, you could use these AI video tools to create slop that no one wants to see, or you could use them to add little touches of spice to your own videos and really grab that attention. Me personally, I still prefer to make 95% human videos and use AI about 5% of the time.

28:08And again, it's more just to inject a little bit of fun and silliness into the video and not to like be the video itself. If I'm being honest, I'm not really looking forward to a world where everything we see online has very little human involvement.

28:22Like, I'm already getting on a lot of these platforms and noticing that I'm having a harder and harder time telling what's actually AI and what's not anymore, and I don't really like that. So when I use AI, I almost, like, purposely show off that it is AI.

28:38So nobody's wondering if this was done with AI or not, But that's just sort of the style that I like doing. But that's what I got for you today. Hopefully, you found this helpful.

28:46Hopefully, you learned something. If you like videos like this and you actually wanna learn practical, actually useful ways to leverage these AI tools that are coming out, I try to make at least one video a week where I'm showing you how to use these tools and how to implement them in your business or your life.

29:00And I'm also making at least one video a week where I break down all of the AI news that I think you need to know. I drink from the fire hose all week. I get sort of overwhelmed by all the AI news so that you don't have to.

29:11And then every Friday, I break down just the most important AI news that I think the most people would wanna know. If those are the types of videos that you think you'd enjoy more of, consider liking this video and subscribing to this channel. I'll make sure that kind of stuff shows up in your feed and that you'll always feel looped in.

29:26Thanks so much for hanging out, nerding out with me on this one. It's always fun when I get to make videos like this. So hopefully, you enjoyed it.

29:32If you wanna see more of them, let me know in the comments. Thanks again. Hopefully, I'll see you in the next one.

29:36Bye bye.

The Hook

The bait, then the rug-pull.

Matt Wolfe has spent years building a signature bag of visual tricks for his YouTube intros — wall-bursting entrances, teleporting location transitions, a Yeti photobombing his shot — and here he opens the entire toolbox: which app, which model, and the exact prompt for every effect.

Frameworks

Named ideas worth stealing.

00:54model

Two-Frame Keyframe Transition

Export the still frame right before the change
Export the still frame right after the change
Upload both as start/end keyframes in Runway (or Kling, Veo, Higgsfield)
Prompt the exact physical transformation, ending with 'the man does not speak'
Blend the generated clip into the timeline with a smooth-cut transition

Matt's signature intro trick: instead of prompting a video generator from a blank scene, you anchor it with two real photos of the before/after state so the model only has to invent the transformation between them.

Steal forany talking-head cold open, a location transition, or forcing a specific action an AI model wouldn't otherwise generate on its own

CTA Breakdown