19:20Chase AI · Tutorial
Claude Code + RAG-Anything = LIMITLESS
How to give your LightRAG knowledge graph the power to ingest PDFs, charts, and images without changing how you query it.
April 3rdThe argument in one line.
A free Motion Design skill gives GPT-6 Astra, Claude Code, and Codex a library of reference videos plus a Higgsfield MCP connection, turning one sentence into a finished motion graphics video.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude Code, Codex, or ChatGPT's Astra and want to generate branded motion graphics without hiring a motion designer.
- You have a product, a website, or just an idea and want a polished explainer, product reveal, or campaign video from one prompt.
- You're comfortable installing a GitHub skill and connecting an MCP server (Higgsfield) to your AI coding tool.
SKIP IF…
- You want manual control over every keyframe. This hands the actual animation to an AI video model (Seedance 2.5), not a timeline you edit by hand.
- You don't want to pay for Higgsfield credits or GPT image generation. The skill still needs those to render anything.
The full version, fast.
A creator built an open-source GitHub skill called Motion Design that plugs into GPT-6 Astra, Claude Code, or Codex and turns a text prompt, product photo, or website link into a finished motion graphics video. The skill ships with a library of reference films across ten styles, so the AI adapts a working example instead of starting from a blank prompt, then writes a detailed, scene-by-scene brief and sends it to the Higgsfield MCP, which renders the video with Seedance 2.5. Three demos, a noise-cancellation explainer, an exploded product shot, and a Canva campaign video, each came from a single casual prompt, with the system checking its own output against the brief and regenerating any scene that misses it.
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 – 00:32
01 · Intro
The creator claims GPT-6 Astra is the best model yet for motion design and previews three demos built from simple prompts.
00:32 – 03:29
02 · Motion Design Skill
Walkthrough of the open-source Motion Design GitHub skill: its reference video library, how it drafts a full production prompt from a one-line input, and how it hands rendering off to the Higgsfield MCP and Seedance 2.5.
03:29 – 08:00
03 · Demos
Three live demos: a 15-second noise-cancellation explainer from a text prompt, an exploded-product film for a keyboard built from two AI-generated reference images, and a Canva campaign video generated from a single product webpage URL.
08:00 – 08:41
04 · Final Thoughts
The creator concludes motion graphics isn't fully solved but tools like this lower the floor for non-designers, and points viewers to the skill and setup guide.
Lines worth screenshotting.
- A GitHub skill called Motion Design gives AI coding agents a library of reference motion-graphics videos so they never generate from a blank prompt.
- The skill works with GPT-6 Astra, Claude Code, or Codex because the actual rendering happens through the Higgsfield MCP, not the chat model itself.
- Higgsfield's Seedance 2.5 is the default video-generation model behind every clip, though the skill lets you swap in any model you prefer.
- The AI expands a one-sentence prompt into a full production brief covering color palette, hero object, typography, motion, and a shot-by-shot beat sheet.
- After generating a video, the system compares the result to the original brief and automatically regenerates any scene that misses the intent.
- A 15-second flat-vector explainer about noise-cancelling headphones took only two generation passes to match the brief.
- Two AI-generated product photos, a keyboard and its internals, were enough reference material to build a full exploded-product film.
- A single webpage URL is enough source material for the skill to build a full marketing campaign video for that product.
- Seedance 2.5 currently supports videos up to around 30 seconds per generation, longer than the 15-second default used in the first demo.
- The creator argues motion graphics isn't fully solved yet, but tools like this let people who aren't motion designers produce professional-looking output.
Feed AI a reference, not a blank page
AI VIDEO WORKFLOW
Giving an AI model a working example to adapt, instead of asking it to invent from nothing, is what turns a one-line prompt into a finished, on-brief video.
- Giving an AI system a library of proven reference examples to adapt, instead of a blank prompt, consistently produces stronger first-pass output than asking it to invent a solution from nothing.
- Turning a one-line request into a detailed brief, covering palette, structure, and a scene-by-scene plan, before generation starts is what separates a usable result from a rough draft.
- Building in an automatic self-check that compares the output against the original intent, and regenerating only the parts that miss it, gets you closer without re-running the whole job.
- The same reference-library approach works across model providers and tools, because the intelligence lives in the reference examples and prompt structure, not in any single AI model.
- A single source, whether that's a plain idea, a product photo, or a webpage URL, can be enough raw material if the surrounding system knows how to expand it into a full brief.
Terms worth knowing.
- Motion Design (skill)
- An open-source GitHub skill that gives an AI coding agent, Codex, Claude Code, or GPT-6 Astra, a library of reference motion-graphics videos and turns a prompt into a finished video via the Higgsfield MCP.
- Higgsfield MCP
- A Model Context Protocol server that connects an AI coding tool directly to Higgsfield's video and image generation models, so the agent can request renders without leaving the terminal or chat.
- Seedance 2.5
- Higgsfield's default AI video generation model, used to render the actual motion graphics from the detailed prompt the skill produces.
- GPT-6 Astra
- OpenAI's assistant used in this video as one of three interchangeable front ends, alongside Claude Code and Codex, for running the Motion Design skill.
Things they pointed at.
00:00toolMotion Design (GitHub skill)
Lines you could clip.
00:00
“GPT-6 Astra is the greatest model we have ever seen when it comes to motion design.”
00:40
“So we never start from scratch.”
08:00
“Is motion graphics completely solved with something like GPT-6 and the motion design skill and the Higgsfield MCP? No, no, it's not completely solved.”
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
GPT -6 Astra is the greatest model we have ever seen when it comes to motion design. And today I'm gonna show you how to unlock its full potential. When you add this motion design skill to GPT -6, you can just go to Astra and give it a simple text prompt or a reference image, or even a link to a website, and it will create motion graphics for an explainer video, a product ad, or whatever else you want.
So today I'm gonna walk you through three examples. I'm gonna show you how you can set this up for yourself. And I'm going to break down how this is all working under the hood so you can customize it however you want.
Now, what makes this all work is the motion design skill. Now, this video is all about how this works with GPT -6 Astra, but this can ultimately work with any AI model. So if you're using cloud code, this skill also works.
And what happens is this gives your AI model a motion library, a bunch of different videos to essentially base its creation off of. So . So for example, let's say I wanted to do a product video that shows sort of the inner workings of my particular product.
Well, we have something that does exactly that. So if I click on the exploded product video, we get this.
Now, chances are you aren't going to be making an exploded product video for a watch. However, if you had something else where you wanted to expose essentially the inner workings of some sort of product, well, we would use this as a template and you wouldn't have to come up with the prompt itself. That's already pretty much made for you with the skill.
It then uses this as the template for your particular creation. And that idea is extended across all of these different styles. And if you want to create your own sort of style, it can, again, use the prompts that are sort of behind each of these as a template as well.
So the idea is we never work from scratch. We're building off of things that we know actually work. Now, the other half of the equation is the Higgs field MCP, because the skill is going to have the reference video.
It's going to create the prompt for you, and then it's going to send it off to Higgs field, which is going to use Seed Dance 2 .5 to actually create the motion graphics. If we are inside of Astra, It has the ability to create images on its own, but if you're using this inside of Cloud Code or Fable 5 .1, yeah, we're going to need to use Higgs field to create the reference images as well.
Now, to do that, we're just going to go to higgsfield .ai, go to MCP and CLI, either the MCP or CLI, both of these work. You're just going to click for your platform of choice. So if we're using ChatGPT, you're just going to do the add the Higgs field plugin, and you're going to walk through the authentication process.
So it's just going to ask for your login, your password, all that stuff. And the great thing about the MCP, instead of having to do all the work inside of the Higgs field UI and the balance between that and Codex, I can just sit inside of Codex or inside of Cloud Code and do everything from there. And the skill is super flexible.
It uses Seed Dance 2 .5 as a default because that's the best AI video generation model out there. You can swap that for anyone you want. On the image side, it will default to GPT images too.
But again, you can swap that to whatever you like. Furthermore, you can do text. We can give it reference images.
You can even give it reference video. So really, you just need to and a general idea of what you're trying to do. Then you need to decide what sort of style you want it to create it in.
Again, you have these examples in front of you. And then it does everything else. so it's pretty simple you just got to show up with the idea now set this all up is really easy i give you the two prompts you can paste into side of cloud code or codex and then it's going to install the skill for you i also added a set of guide specifically for hicks field as well then all you have to do is hop into codex do forward slash motion design to call on that skill and give it your prompt and here's the three demos i created using this skill so the first one is a 2d explainer video talking about how noise cancellation works and not only does the motion graphics it also does audio now you can see down here this is the prompt i used right it just said i just said use motion design to make a 15 second flat vector explainer about how noise cancelating headphones work adapt the flat vector sample show the mechanism and then show me the full prompt before generating and from that basic prompt
This is then the prompt that was sent off to the Higgs field MCP. And you can see this is a lot more in depth. We went from a 2D explainer, use this one to this full prompt that breaks down everything from the color palette to the hero, to the type, to the motion, to a beat sheet, what we want to happen on each and every slide.
So instead of you having to show up with this, you just need the skill and a simple input. Now, the other cool thing about these videos is most of these weren't necessarily one shot. What Codex was able to do, what Astra was able to do, is it generates the video the first time.
It compares the first generation to what you're essentially prompt was. What were we actually trying to achieve here? And if it thinks one particular section needs some work, it will then regenerate using Seed Dance 2 .5.
And again, if you're worried about credits or tokens or anything like that, you can reduce that. But it essentially is a built -in system to make sure we don't just one shot and pray. It will keep adjusting it until you get what you want.
So here's a look at the first one. Noise cancellation fights sound with sound. Microphones detect the noise.
A processor creates an opposing signal. The speaker plays it. The waves combine at your ear, reducing the noise, especially steady hums.
So not bad for essentially a one -shot on our end. It did two generations and created this. it's set for 15 seconds on this first prompt it can go up to i believe 30 seconds with cdance 2 .5 so again you have a lot of wiggle room for what's actually generated here and that was just a pure text prompt for our second demo we built off that exploding product thing because remember we have sort of the style gallery like i talked to before where you have all these different options to choose from and so for this one we use the deconstructed watch template and this was the starting prompt i gave it use motion design to create a 20 second exploded product film for this keyboard using the two images and then i included two reference images that astra created itself so we have the keyboard and the keys and then the shot of the internals and from there it created this Pretty solid work.
Lastly, we created this campaign video for Canva. So the prompt here was use motion design to make a 20 second kinetic promo based on. Canva .com's visual suite.
So I gave it the link to this webpage, which is a specific Canva product, and was like, hey, can we turn this into a video? So I said, show one fictional coffee shop campaign becoming a social post, a story, a presentation, adapt the kinetic typography example, and end with one idea, every format. And this was the full prompt it made for us.
Again, you can see it breaks it down for every single scene, goes into a soundscape, all that sort of stuff. And here's what we got on essentially a one shot.
Again, I think that's a pretty solid output. There are some things we can tweak here, but I think the biggest takeaway from all three of these is that I gave it. pretty basic prompts at the beginning of every single generation, right?
This starting input is nothing serious and we're getting like legit outputs. If you had a much stronger vision of what you're trying to create and we're willing not to just like one shot this, but to iterate, iterate, iterate, you can create some pretty solid motion graphics, whether that's something that's 2D or some sort of product thing or a campaign with like this Kinect typography or really anything else within the style gallery, like you can get solid outputs with Astra.
So is motion graphics completely solved with something like GPT -6 and the motion design skill and the HIG -2 and the MCB? No, no, it's not completely solved. There's still a lot of work that can be done.
But if you compare where we're at now with where we were at even a few months ago, you can see the direction this is heading. And you can see the sort of skills and models like this are for those of us who are not motion designers. You can create some really solid outputs using Astra, Fable 5 .1, and a skill like motion design.
Make sure to check out down below to get the skill for yourself. Let me know how this works for you. I'd love to see how you're able to adapt it as well because you can customize this to your heart's content.
And besides that, I'll see you around.
The Hook
The bait, then the rug-pull.
The creator opens with a blunt claim, that GPT-6 Astra is the best AI has ever been at motion graphics, then spends the next eight minutes proving it with an open-source skill anyone can install for free.
Named ideas worth stealing.
03:29list
Three ways to start
- An idea (text prompt)
- Product images
- A website URL
The three input types the Motion Design skill accepts to kick off a video: a plain text idea, one or more product reference images, or a link to a webpage it should adapt into a campaign.
Steal forany AI video-generation workflow that needs a flexible intake format
02:29model
Choose, Adapt, Generate, Review
- Choose the closest example
- Adapt scenes, sequence, and timing to your subject
- Generate the complete film via Higgsfield MCP
- Review and request corrections
The skill's internal four-step pipeline for turning a prompt into a finished video, always starting from an existing reference film rather than a blank page.
Steal forstructuring any AI generation pipeline around a reference library instead of cold generation
How they asked for the click.
VERBAL ASK
08:30link
“Make sure to check out down below to get the skill for yourself.”
Soft CTA at the very end pointing to the free GitHub skill and description links (Higgsfield affiliate, Skool community/course, consulting site) rather than a hard sell.
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.
