How to Generate Notion Graphics With Claude

Adding Notion graphics to a video normally means opening an editor, designing shapes, and timing every animation by hand. A new workflow built around a custom Claude skill automates that pipeline: you supply the raw video and transcript, and the model plans and renders the b-roll for you.

What the Claude Notion Graphics Skill Actually Does

The Claude Notion graphics skill is a custom instruction package, published by YouTuber Bart Slodyczka in September 2026, that turns Claude into a b-roll generator for existing videos. You upload a raw clip, its transcript, and a timestamp file; the skill tells Claude how to read them, plan where graphics belong, and render animated clips matched to the rhythm of your words.

The skill is scoped deliberately narrow. It exists only to insert Notion-graphic b-roll into footage you already have. It does not edit your video, cut your footage, or handle your intro. In the demo session, the creator had already trimmed his talking-head clip and reframed himself to one side of the screen before uploading it. That framing matters: because the skill can see that the speaker occupies one third of the frame, it positions the animated graphics in the empty space instead of covering him.

Everything the skill does is encoded in its instruction files. The skill.md file is the master set of directions, covering everything from reading the video and transcript to generating the graphics so they stay synchronized with the script. Supporting folders hold scripts, rendering instructions, and example assets the creator accumulated while building the workflow. A skill like this is also reusable knowledge: over time you can extend it with your own learned techniques, saved colors, and brand assets.

How the End-to-End Workflow Runs, Step by Step

The whole process runs inside Claude Code, Anthropic agentic coding tool, in a desktop project folder. The creator's argument for a dedicated project folder is practical: you will want one place to store the skill, reusable resources, and documented lessons as you refine your own Notion-graphics workflow.

The sequence he demonstrates is short enough to reproduce from memory:

1. Create a fresh project folder on your desktop and open it in the Claude Code desktop app.
2. Point Claude the skill's repository URL and ask it to read and import the b-roll skill into the workspace, setting up any supporting agent or memory files it needs.
3. Upload three things: the raw video, the transcript, and the timestamp file for the segment.
4. Ask Claude to run the b-roll skill, specifying density (medium means roughly three clips, intense means four to five) and a color palette.
5. Ask Claude to embed all generated clips into a local HTML page so you can preview them in a browser instead of downloading files one by one.
6. Review the clips, keep what you like, and delete the rest before final assembly in your video editor.

Two configuration details come from his session. He ran the model at its default medium effort setting and has not tested low or high effort with this skill. And on his first session in a new folder, he explicitly waited for Claude to finish importing and configuring the skill before uploading any media, so the model read the instructions cold rather than improvising.

The Continuous-Shape Technique Behind the Animations

The visual signature of this skill is a single technique: each animated object changes continuously from one shape into the next, so a shape morphs rather than cut-transitions through a sequence. The creator credits this continuous fusion between shapes for the engaging, dynamic quality of the finished clips, and the skill file carries explicit instructions for how Claude should apply it.

The skill also encodes editorial judgment, not just rendering rules. In the demo, the animations appear at the exact moments the speaker says the matching words, because the skill instructs Claude to sync graphics to the transcript. The creator points to one segment where the skill distinguished between two effort levels in the source material being described, producing a graphic only where the content actually justified one. Timing, placement, and the decision of whether a graphic belongs at all are all part of the instruction package.

Placement adapts to your framing as well. The skill gives Claude the ability to overlay a full-screen graphic when it fits, and the demo shows it placing graphics in the black space beside a speaker framed to one third of the screen. If you prefer to control density yourself, choosing the intense setting up front and deleting clips afterward is his recommendation: having more clips than you need beats asking for more later.

Inputs You Must Supply: Video, Transcript, and Timestamp

The skill currently requires three inputs from you, and the transcript is the one that adds manual work. Claude reads the video and its timestamp, but the skill has no built-in way to produce a transcript automatically, so you upload one alongside the footage. In the demo the creator mentions two possible fixes: sending audio to the ElevenLabs transcription API, or wiring in a local transcription model on your own machine. Neither was integrated at the time of the video.

Your raw footage should also be editorially prepared. The creator recorded his talking-head segment, cut it in his editor, and built the transition that shrank himself to a corner of the frame before uploading. The skill handles graphics, overlaying, and placement, but first-pass editing still belongs to you.

One preview setting is worth copying. The skill's default behavior hands you downloadable clip files; he instead asked Claude to embed every generated clip in an HTML page inside the workspace, which let him OpenAI browser tab and review all five animations side by side without leaving the project folder.

Results, Density Options, and What a Session Produces

On a 33-second clip at intense density, the skill produced five clips and the creator estimated the full asset generation took about fifteen minutes. He reviewed them once, with no re-prompting, and decided to keep them as generated rather than revising the plan. His on-camera reaction was that the sync between graphics and specific spoken words held up, including a sequence where a b-roll element slides onto the screen as a multi-track moment.

Density and palette are the two main dials. Medium density yields roughly three clips across a 33-second segment; intense yields four to five. He defaulted to the standard palette for the demo but suggests that first-time users upload their own brand colors and style assets at this stage to personalize output. One clip from his run did not survive review: the opening graphic conflicted with a cut he had planned later in the edit, and he flagged it for removal rather than regenerating.

All of these figures come from the creator's own session and are first-hand experience, not an independent benchmark. Results will vary with your footage, your framing, and how much editing context the skill has to infer.

Current Limits and Where the Creator Is Taking the Workflow

Three limits are worth knowing before you try this. First, the skill covers only b-roll insertion, so trimming, transitions, and reframing remain your job. Second, automatic transcription is not built in; you supply the transcript yourself until the creator finds a reliable local-model process for it. Third, it has only been exercised at the default medium effort setting, so there is no documented evidence yet about how low or high effort changes the output.

The creator says he is building two additional skills in the same family, one for product launches and one for explainer videos, including faceless explainers. His stated long-term goal is more ambitious: a fully automated pipeline where he records a 45-second talking-head script, uploads it, and Claude handles the cuts, the engagement edits, the Notion graphics, and everything else an editor would normally do. That automation does not exist yet in what he demonstrated; the b-roll skill is the first piece of it.

It is also fair to note what this article cannot verify. The skill is distributed through the creator's channel rather than through a documented public specification we audited, so treat the described behavior as his demonstration and testing, and expect the workflow to evolve as he ships the follow-up skills.

FAQ

• Do I need to edit my video before using the skill? Yes, for first-pass cuts and framing. The skill inserts Notion-graphic b-roll into footage you provide, and the demo workflow had the speaker already trimmed and reframed before upload. If you want graphics placed in empty screen areas, position yourself in the frame accordingly.

• Does the skill generate the transcript for my video? No. As shown in the September 2026 video, the transcript is a required upload, and the creator named an external transcription API and a local transcription model as possible future integrations. Neither was wired into the skill at the time.

• How many graphics does one session produce? The density setting controls it: medium produced about three clips over a 33-second segment, while intense produced four to five in the demonstrated run. You can generate at intense density and delete clips you do not want.

• What model and settings were used in the demonstration? The creator ran Claude its default medium effort setting and reported that he had not tested the low or high effort levels with this skill, so medium is the only documented configuration.

• Can the graphics cover the whole screen? Yes. The skill gives Claude the option of a full-screen overlay when the content calls for it, and it also positions graphics in empty regions of the frame when the speaker is off to one side.

Turn Your Own Videos Into Written Articles

This workflow works because a video's knowledge, the exact steps, timing decisions, and limits demonstrated on camera, can be captured in a structured form another person can follow. The same is true of your own channel: every tutorial, review, or walkthrough you have published already contains an article waiting to be written.

Skala Blog does exactly that. Paste a YouTube URL, let it transcribe the video, and generate a written article from the transcript, so the knowledge inside your footage reaches readers who search instead of watch.

Source video