12:15Bart Slodyczka · Tutorial
Claude Code + LM Studio: FREE Unlimited AI Agents (Don't Pay $200/month)
A 12-minute setup guide for running Claude Code multi-agent Dynamic Workflows on free local models with no Anthropic account required.
May 31stThe argument in one line.
A packaged Claude Code skill turns Opus 5.5 into a motion-graphics editor that plans, times to the spoken words, and renders B-roll animations from nothing more than a raw talking-head clip, its transcript, and a density setting.
Read if. Skip if.
READ IF YOU ARE…
- You make short-form or talking-head video and want B-roll motion graphics without hand-animating or hiring a motion designer.
- You already run Claude Code Desktop projects and want a concrete, working example of how a reusable 'skill' folder is structured and installed.
- You're curious what Opus 5.5 at default Medium effort can actually produce on a real creative-production task, not a coding benchmark.
SKIP IF…
- You need the tool to also generate your transcript for you; this version requires you to supply one yourself.
- You want a polished, finished-product walkthrough rather than a rough live first run the creator himself says he'd normally go back and re-edit.
The full version, fast.
A creator built a Claude Code skill, published free on GitHub, that turns Claude Opus 5.5 into a motion-graphics editor. Fed a raw talking-head clip, its transcript, and a chosen density (light, medium, or heavy), Opus 5.5 plans a clip-by-clip B-roll sequence, mapping specific lines to specific treatments like full cutaways, on-face holds, and transparent panels, for approval before it renders anything. The skill's signature technique keeps one shape continuously morphing into the next instead of cutting between separate graphics, which is what makes the output read as one fluid animation. Five finished clips took about 15 minutes to render, timed to the words automatically, with zero re-prompting.
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 →Where the time goes.
00:00 – 00:26
01 · Cold open: before/after reveal
The creator opens with the payoff already proven: an original talking-head clip is played next to the same clip with Opus-5.5-generated motion graphics cut in, before he explains any of the process.
00:26 – 02:24
02 · Why this matters: the effort-level test
He recaps a companion video where he rebuilt Age of Empires 2 with Opus 5.5 at every effort level and tracked cost, time, and token usage for each, noting this B-roll skill wasn't ready in time to use on that upload.
02:24 – 02:53
03 · The pitch
He states what the skill actually does, inserting motion-graphics B-roll into an existing video, and previews two more skills in progress, one for product launches and one for faceless explainer videos.
02:53 – 03:49
04 · Setting up a dedicated Claude Code project
He opens a fresh Claude Code Desktop project folder called 'yt-demo', recommends a dedicated project per recurring motion-graphics work, and leaves Opus 5.5 on its default Medium effort.
03:49 – 05:26
05 · Touring the skill's GitHub repo
He walks through the public repo 'Barty-Bart/motion-graphics' and its 'motion-broll' skill folder (engine, examples, reference, scripts, templates, SKILL.md), naming the core technique: one shape continuously morphing into the next instead of cutting between separate graphics.
05:26 – 05:49
06 · Installing the skill into the project
He pastes the repo URL into the Claude project and asks it to read the skill, import it, and generate its own CLAUDE.md/AGENTS.md notes so the setup reproduces automatically in future sessions.
05:49 – 08:23
07 · The raw footage and the bigger goal
He plays the unedited talking-head clip he manually cut and shifted into a side-of-screen frame before handing it to Claude, and lays out his longer-term goal: a workflow where Claude does the cuts, framing, and motion graphics for an entire short on its own.
08:23 – 09:35
08 · Feeding Claude the clip, transcript, and density
He uploads the video and a separately supplied transcript, since the skill doesn't transcribe on its own yet, asks for the output to be embedded in a local HTML preview page, and chooses Heavy density (4-5 clips) with the default color palette.
09:35 – 10:01
09 · Reviewing and approving the plan
Claude returns a plan mapping five specific transcript lines to five specific treatments, full cutaways, an on-face hold, a transparent panel, and he approves it without requesting changes.
10:01 – 12:22
10 · The result
About 15 minutes later the five clips are rendered into an HTML gallery; he reacts to each one live, calling the results 'insane,' and ships the output as-is rather than re-editing.
Lines worth screenshotting.
- A Claude Code skill is just a folder convention, a SKILL.md file plus engine, scripts, examples, and reference subfolders, that Claude reads once and reuses across sessions.
- The B-roll's core visual technique keeps one shape continuously morphing into the next instead of cutting between separate graphics, which reads as one fluid animation instead of a slideshow.
- The skill doesn't transcribe video on its own; a transcript still has to be supplied separately before it can time animations to the words.
- Choosing a density setting, light, medium, or heavy, controls how many B-roll clips get generated without the creator describing a single animation.
- Before rendering anything, Claude produces a reviewable plan mapping specific transcript lines to specific clip treatments, so approval happens on a plan, not on finished renders.
- Five finished, word-timed B-roll clips rendered in about 15 minutes with zero re-prompting after the plan was approved.
- Testing one model across every effort level on the same prompt, as the creator did in a companion video building an Age of Empires 2 replica, is how you find out what each effort tier actually costs in time and tokens.
- The manual work still required going in, cutting the raw footage and shifting the frame from full-screen to a side third, happens before Claude ever sees the clip.
- The stated long-term goal is a workflow where Claude takes a raw talking-head recording and produces the entire edited short, cuts, framing, and motion graphics included, not just the B-roll layer.
- Dedicating one Claude Code project folder to a single recurring task lets custom skills and reusable resources accumulate there instead of scattering across one-off sessions.
A Claude Code Skill Can Now Plan and Time Its Own B-Roll
WHAT TO LEARN
Handing Claude Opus 5.5 a talking-head clip, a transcript, and a packaged 'skill' folder is enough for it to plan, time to the words, and render B-roll animations without being told what shape to draw.
02Why this matters: the effort-level test
- Testing one model across every effort level on the same prompt is how you find out what you're actually paying for: the creator tracked cost, time, and token count separately for Opus 5.5 at low through max effort.
- A feature or skill built today can outdate content published yesterday; the creator's own effort-level test video from the day before couldn't use this skill because it wasn't ready yet.
03The pitch
- A single skill is being scoped narrowly, B-roll only, rather than as one do-everything tool, with separate skills planned for product launches and explainer videos.
04Setting up a dedicated Claude Code project
- Dedicating one Claude Code project folder to a recurring task, here motion graphics, lets reusable resources and custom skills accumulate there instead of getting lost across one-off sessions.
05Touring the skill's GitHub repo
- A Claude Code skill is just a folder convention: a SKILL.md file plus supporting engine, scripts, examples, and reference subfolders that Claude reads once and reuses.
- The signature visual trick, one shape continuously morphing instead of cutting between separate graphics, is what makes generated motion graphics read as a single fluid animation rather than a slideshow.
06Installing the skill into the project
- Pointing Claude at a public GitHub URL and asking it to import the skill plus set up its own CLAUDE.md/AGENTS.md notes is enough to make the setup persist and self-document for future sessions.
07The raw footage and the bigger goal
- The manual work still required going in was doing the cuts and the full-screen-to-side-third framing shift by hand before Claude ever sees the clip.
- The stated end goal is a Claude workflow that takes a raw talking-head recording and produces the entire edited short, cuts, framing, and motion graphics included.
08Feeding Claude the clip, transcript, and density
- The skill doesn't transcribe video yet, so a transcript has to be supplied separately; the creator used one he already had and floated a local-model transcription path as a future fix.
- Choosing a density setting up front, light, medium, or heavy, is how the creator controls output volume without having to describe individual animations.
09Reviewing and approving the plan
- Before rendering anything, Claude produces a reviewable plan mapping specific transcript lines to specific clip types, so the creator approves a plan, not five finished renders he might reject.
10The result
- Five finished B-roll clips took about 15 minutes to generate with zero re-prompting, which the creator treats as good enough to ship even though he says he'd normally go back and re-edit.
Terms worth knowing.
- Claude Code skill
- A packaged folder of instructions and resources, a SKILL.md file plus supporting scripts, examples, and reference material, that Claude reads once and can reuse across sessions in a project.
- SKILL.md
- The core instructions file inside a Claude Code skill folder that tells Claude how to execute the task end to end, from reading input to rendering final output.
- Effort level (Opus 5.5)
- A setting in Claude's model picker, Low, Medium, High, Extra, or Max, that trades more reasoning time and token usage for higher-effort output on the same prompt.
- Density (motion graphics)
- A setting that controls how many B-roll clips get generated per minute of video, ranging from light (1-3 clips per minute) to heavy (4-5 clips covering most lines).
- B-roll
- Supplementary footage or graphics cut into a video alongside the main shot, used here to mean the generated motion-graphic overlays and cutaways.
Things they pointed at.
07:56toolElevenLabs API (mentioned as an alternative way to get a transcript)
Lines you could clip.
00:00
“So I just created a skill that lets Opus 5.5 create motion graphics like these.”
04:34
“Our core technique is to have one object continuously shift into different shapes.”
10:58
“That's insane. That's actually, I am, this is my first time seeing this clip and it is, that's so sick.”
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.
So I just created a skill that lets Opus 5 .5 create motion graphics like these. So on screen right now I've got two different clips. I have an original clip which is what I uploaded into my Cloud Opus 5 .5 session and on the right hand side I have a clip which has been motion graphics animated by Opus 5 .5 using this new skill, this new process that I'm going to show you in this video.
Now I'm going to leave the sound on and I'm just going to let you guys watch the difference for yourselves. So Opus 5 .5 just came out and in this video I built a replica of Age of Empires 2 against every reasoning level. To set this up I created a blank folder on my desktop, I opened up Cloud Code, gave it this prompt and set Opus 5 .5 to low.
I let it start building and then I replicated this process once for each reasoning level. Because of this I was able to measure exactly how much each build cost to make, how long it took and how many tokens we consumed. And I'll be sharing all these details with you throughout the video.
Now, as you can imagine, the results are going to vary quite a lot between the low effort and the max effort. So at the end of the video, I also did one more bonus build. I ended up taking notes of all the things that I liked, didn't like, and things that I thought were missing from all the builds.
And I created one more master prompt that I fed back into an Opus 5 .5 low session with the goal of seeing if I give a bunch of information up front, can I get outstanding results from the low that will rival the results of like a high or a max. So make sure you watch to the end of the video to see the results of that test as well.
Okay, let's get into the first build. So everything you just saw was completed by the skill, including how these animations would pop up exactly when I'm saying certain words. The actual style of the animations themselves, all completed in the skill.
All that information is contained in there. This part I actually really liked when it was saying there's like a difference between low and high. How when it goes across to the high, it actually builds out the age of vampires.
But when it's on the low, you see that it's like not really built out. This kind of stuff is literally built into the skill. This was all thanks to Opus 5 .5, just intuitively understanding that there should be a graphic and it should be a castle because it's most related to Age of Empires.
Now, if you do watch that video that I posted yesterday, unfortunately, I didn't have this skill ready by then. So the motion graphics there are not as good as this. But I also like this part.
where it like drags the element and it puts it into the position. And yeah, everything else here is just super cool. Like all these animations, I did not have to describe.
I just gave the video into my Cloud Opus 5 .5 and it produced this final result. I'm really excited to share this with you guys. Now I can that was absolutely ripper.
So in this video, I'm going to show you my exact process and give you access to this skill so that you can go ahead and build motion graphics like these for yourselves. Now as you can see, this skill is just to insert motion graphics as B roll into videos. But I'm also working on two additional skills, one skill that's going to be for product launches and another skill that's going to be for explainer videos like faceless explainer videos.
So if you want to keep getting access to all these skills as soon as they come out, make sure you subscribe to my channel. Okay, so let's get right into the process. Okay, so let's first start by setting up our workspace.
So I've got a brand new project folder open inside my Cloud Code Desktop app, and this is just called YouTube Demo. So if you want to do more work with motion graphics in general, I would suggest getting a brand new folder, then going into your co -work mode, and basically using this as a way to work on each kind of motion graphic or B -roll or whatever it is that you want to create, because over time, you might actually create your own skills.
You might start storing reusable resources that you... create over time and then maybe you want to document certain learning so definitely working inside a project is the best way to go and then of course i'm going to be using opus 5 .5 now i want to be using the default medium effort setting i i haven't tried this on low and i haven't tried this on high i've just been using medium and i'm getting amazing results what you saw in the intro of this video that was all done with opus 5 .5 medium so now that we have our workspace ready The next thing we want to do is go across to my GitHub repo, which is called motion graphics.
And so far, I've installed one skill into here, which is just called the motion b -roll skill. But like I just mentioned, I'm going to be popping in some more skills across this next week, because I'm already refining a process for doing product launches and then faces explainer videos. So let's go inside this motion b -roll skill.
Now this skill is very built out and includes a lot of different resources. At a high level, the skill .md file is the core set of instructions for our Claude session. So it knows how to do everything from reading our video, reading our transcript, creating the motion graphics, making sure that the motion graphics actually match the beat of our words so that everything stays in rhythm.
It also includes instructions for the core technique that we use here. Our core technique is to have one object continuously shift into different shapes. And that's how we get this really engaging and dynamic movement.
The rest of these folders here include information about scripts on how to create motion graphics, how to read our videos, how to render the videos, and a couple of examples that I've created along the way. Over time, if I create any new and interesting resources, I'll just keep dumping them back into here. So the easiest way to work with this skill is just go into this motion graphics home folder.
We're just going to copy the URL over here. I'm going to go back into my Claude session and paste in the URL. Hey, can you please go across to this GitHub repo?
And I want to use the B -roll motion graphics skill. I'm going to be uploading a video as well as my timestamp and transcript for that video. It's about 30 seconds long.
And I'm going to want to use the B -roll motion graphics skill on the video that I upload. Now, because this is my first session inside this project folder, I'm just going to give Claude some time to actually read that skill, understand what has to be done before I upload the video. Can you please read that skill and then import it into our workspace?
I'm going to be using this project folder to do all of my motion graphics needs. So if there's any other agent MD or Claude MD files that you need to set up, go for it as well. So let's send this across to Claude.
And now while we wait for Claude to get that initial bit set up, I've actually got the intro for this video and we're going to be adding the B -roll together. So whatever you just saw for that first, I think it's like 35 seconds of the clip, whatever you saw as the B -roll is literally what we're going to be building out right now.
So I think it's going to be pretty cool. So just to watch this without any of the B -roll. I'm just going to put it down to low volume so I can talk over it.
So what I actually did here was I recorded myself and then I chopped up inside my video editing tool. And then I kind of made this transition so I could at least push myself to the side of the screen because I knew that I wanted some kind of visual over here. Now, the skill that we have will also give Claude the ability to completely overlay.
a motion graphic so if it thinks that at this stage for example it wants to completely cover the entire screen with like a nice engaging motion graphic it'll do that as well and when it sees my like basically instead of me being full screen going into that little third of the screen it knows exactly where it has to land the motion graphics as well So as you saw, that was just the raw video.
And we're going to be plunking that in as well as the transcript for Claude to work with. So it looks like Claude has set everything up. But before I get into this, I also want to say that technically with the video that I'm working on there, I had to do the initial cuts and I had to do the initial like shifting from being full screen to being in that side third.
My goal in general is to be able to upload at least for like YouTube shorts. I don't do enough of them and I don't do anything on Instagram shorts either. So I want to get into that space.
And my goal is to be able to do like a talking head video where I'm just looking at the camera and I'm reading a script or whatever for like 45 seconds. And I can dump in that talking head and I can build a bunch of motion graphics and video editing skills around that. So then my Claude is able to take that video and chop it up and make it engaging and then to make all the motion graphics and basically do everything that an editor would normally do.
So that's my long term plan. And I'm actually working on that in the background. As soon as I have anything tangible and interesting and actually production ready, I will bring it out to YouTube and I'll make some videos about it.
So it looks like Claude has set everything up for us. I'm just going to be putting in my video and the transcript. Now, the reason that I'm putting in the transcript is because the skill doesn't yet include some way to get a transcript.
You could use the 11 Labs API and send your video across to there and get a transcript back. But I know that a lot of us might actually have a local transcription model running on our computer. So I want to try and figure out a process to just use our local model to do the transcription for us.
But for now, I haven't figured that part out yet. So we just have to upload the transcript as well. So just putting in the video and putting in the transcript.
Okay, thank you very much. So I just uploaded my video and the transcript. Can you please run the B -roll motion graphics skill against this media?
And then for all the clips that you produce, can you please embed them into a HTML page so that I can just see them within this local workspace? So I'm opting to get the motion graphics clips put onto a HTML page. So I can just open up this browser and then view everything that's being worked on over here.
But I'm pretty sure the default setting of this skill is just to give you those files ready for you to download. So you might just want to copy what I'm doing here and ask to get them embedded into a HTML page. Okay, so how many motion graphics do we want across the 33 seconds?
So medium is going to be three clips and then we have heavy four to five clips. I'm actually just going to start with the heavy four to five. I'd rather have the ability to just tone it down if I want to versus not having enough clips.
And I'm just going to be using the default palette. But at this stage, it might be interesting for you, especially in the first session, to just upload your style, to upload your brand colors and any other resources that would just personalize this for you. Let's go with default for now.
I'm just going to minimize this so we can have a better experience looking at the plan that Claude created for us. So we have five clips, heavy density with the default palette. We have the first line.
So a skill that lets Opus 5 .5 create motion graphics like these. We have the full cutaway. I reckon that was absolutely ripper.
We're just going to be on the face, show you my exact process, give you the access and build motion graphics, transparent panel in the blanks in the black space. So that's when my that's when my face kind of goes off to the side. We have another full cutaway, another full face.
So I'm just going to automatically approve this plan. And it looks like we've got all the resources created. So let me just see if I can find how long that took.
Where was the... Okay, so I can't find the timer, but I think it was about 15 minutes to generate all of these assets. So let's open up the HTML.
Let's play our first one. Okay, this first frame is not glorious for me.
Yeah, okay. So that's interesting. I'm not sure if I'm going to keep that very first part because I then cut away to like...
i don't know whatever it was like 30 seconds of actually showing us the side by side completed motion graphic from a different video absolutely so in this video i'm going to show you my exact process and give you access to the skill so that you can go ahead and build motion graphics like these for yourselves that is cool that actually looks fantastic and that this is literally like i didn't do any reprompting at all you can see this skill is just to insert motion graphics as b -roll into videos but i'm also working on two additional skills One skill that's going to be for product launches and another skill that's going to be for explainer videos like faceless explainer videos.
So if you want to keep getting access to all these skills as soon as they come out, make sure you subscribe to my channel. That's insane. That's actually, I am, this is my first time seeing this clip and it is, that's so sick.
That is actually very cool. And then we have all the individual clips we created here. I'm glad that I went with five because I can kind of remove them if I don't want to keep them.
uh then we have the sideways one yeah all this stuff this is so neat this is i love when you can kind of like bring the additional thing in like just being able to bring that element into this canvas is it just tells so much like you're doing b -roll you're bringing the b -roll and you have like a multi -track sequence this is incredible this is actually insane you know i'm probably gonna just have to end the video here because i thought i would go back and do some re -editing and like re -prompting to make things look a little bit nicer but i'm literally just gonna go with exactly what i got here so yeah guys if you've watched this far and you enjoyed the process and you're excited to try this skill out for yourself please like the video please drop a comment below and also please subscribe to my channel all those things will signal to youtube that this is a great video and will push it out to more people so that more people can get amazing b -roll as well all right thank you very much guys i'll see you in the next one
The Hook
The bait, then the rug-pull.
He opens by running the side-by-side proof: the same clip, once raw and once run through a Claude Opus 5.5 skill that adds word-timed motion graphics, and lets the difference speak before he explains a word of the process.
Named ideas worth stealing.
04:34concept
The single-shape morph technique
Instead of cutting between separate graphic assets, the skill keeps one object on screen that continuously reshapes itself into the next form, which is what makes a run of B-roll read as one continuous animated sequence instead of a slideshow of unrelated graphics.
Steal forany B-roll or motion-graphics workflow (Remotion, After Effects) that wants visual continuity instead of hard graphic swaps
09:08list
Motion-broll density levels
- Light: 1-3 clips per minute
- Medium: 3 clips per minute (recommended)
- Heavy: 4-5 clips covering most lines, including the subscribe call-to-action
The skill asks for one density choice up front and uses it to decide how many B-roll clips to plan, rather than requiring the creator to specify each animation individually.
Steal forany content-generation tool where a single coarse setting can replace per-item manual direction
04:10model
Skill folder anatomy
- engine
- examples
- reference
- scripts
- templates
- SKILL.md
A Claude Code skill packages its instructions (SKILL.md) alongside the code, examples, and reference material it needs, so importing one GitHub URL into a new project folder is enough to reproduce the whole capability.
Steal forpackaging any repeatable Claude Code workflow into a shareable, reusable skill
How they asked for the click.
VERBAL ASK
11:52subscribe
“please like the video, please drop a comment below, and also please subscribe to my channel”
Stacked end-card ask (like, comment, subscribe), explicitly justified as a signal to YouTube's algorithm to push the video to more people; standard structure with no unique twist.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
