You can build an AI influencer with Claude for free by cloning Kevin’s Claude Codes Influencers repo, running mascot.python on one uploaded image to build a mascot, generating the facial expression asset set, training a voice in the built-in voice studio, and dropping the character into your editor. The source lives at github.com/kevinbadi/claude-codes-influencers. This post walks the five steps in order.
one image -> mascot.python -> asset set -> voice studio
|
v
mascot dropped into your editor
|
v
posts that react to your edits
This is Kevin’s own project. It is not a Creator OS feature. Creator OS comes in at the end, when you want the finished mascot to publish, reply and report across your accounts. The repo and the walkthrough are what you learn here. Watch the demo on the KevBuildsApps YouTube channel if you want to see the steps run before you touch a terminal.
What Is an AI Influencer With Claude and Why Build Your Own
An AI influencer with Claude means you run an agent driven pipeline you own instead of renting a hosted avatar tool. HeyGen and Higgsfield are hosted avatar and video platforms: you pick or upload a character, type a script, and their servers render the talking head video. That route is built around their account and their terms.
The Claude Codes Influencers repo takes the other route. You clone a public repository, run a Python file to build the character, and keep every file the process produces. No seat, no per minute render charge, no character that only exists inside someone else’s dashboard. What you trade for that is setup time and a local machine that can run Python and your editor. If you want the short version of that trade, the five steps below are the whole thing.
How the Claude Codes Influencers Repo Works in 5 Steps
The pipeline is five moves: open the repo, build the mascot from one image, generate the asset set and expressions, train a voice, then edit. The video description lists them exactly that way, and each one is a separate action you can stop and restart.
- Open the Claude Codes Influencers repo at github.com/kevinbadi/claude-codes-influencers and read the instructions there.
- Run
mascot.python, the file that builds any character you want from a single uploaded image. A photo of a person or a cartoon drawing both work as input. - Let it generate the asset set, which includes the facial expressions.
- Open the built-in voice studio and train a voice model for the mascot.
- Drop the mascot into your editing workflow and start cutting the edits it reacts to.
Step five is the one people skip and it matters most. A mascot that only exists as a still portrait is a logo. A mascot that sits in your editor and reacts to cuts, zooms and b roll is a character. Build the asset set with that in mind and you end up with a library you can reuse across dozens of videos instead of a one off render.
If you have not run a Claude Code skill before, the shape of this workflow will feel familiar. the one line launch video skill walkthrough shows the same pattern: describe what you want, let the skill produce assets, then review the output yourself.
How to Build a Mascot From One Image With mascot.python
You only need one image to start. mascot.python takes that upload and builds the character from it, whether the source is a real person or a cartoon design.
A few practical notes that the repo’s step list implies but are worth stating plainly:
- Pick the image carefully. Choose a clear, well lit, front facing shot. The character it builds inherits the shapes it can read from that file.
- Person or cartoon both work. The video description says “any character you want from a single uploaded image, person or cartoon”, so the input is not limited to photography.
- Keep the original file. You will want to rebuild the mascot later with a different image once you know what you like.
- Run it once before you design anything around it. See what the mascot step actually outputs on your machine before you plan a whole channel.
That last point saves time. People sketch a content plan, then discover their first generated mascot does not look the way they imagined, then restart the plan. Build one character, look at it, decide, then plan.
How the Asset Set and Facial Expressions Get Generated
You don’t draw the expressions yourself. The repo generates the asset set, and facial expressions are part of it, which is what turns a static portrait into something you can animate.
Think of the asset set as the character’s wardrobe of faces. When you edit a video, you want the mascot to look surprised at the right beat, unimpressed at the right punchline, or neutral while a caption explains something. A generated expression set gives you those options without opening a drawing app.
Two habits keep that library usable:
- Label what you generate. Name expression files by emotion and version so you can find the one you want during an edit instead of scrubbing a folder.
- Regenerate rather than repair. If a face comes out wrong, rerun the generation step with a better input image rather than trying to fix it by hand.
If you later want those expressions animated rather than swapped, the motion side is its own craft. the motion style atlas covers the range of movement styles you can aim for, which helps you decide how expressive your mascot should be before you cut anything.
What Is the Built-In Voice Studio and How Does It Compare to Hosted Voice Tools
The voice studio is the part of the repo where you train a voice model for your mascot, and the video describes it as a free version of 11 Labs. That description is about the workflow, not a claim of identical output.
Here is the honest framing. ElevenLabs is a hosted voice cloning and text to speech service that you use through a web account. The voice studio inside this repo is a local component of Kevin’s project that lets you train a voice model for the character so the mascot can talk. You run it yourself and you keep the model file.
| Approach | What you do | What you keep |
|---|---|---|
| Hosted voice service | Upload samples, generate speech through a web account | Audio clips, subject to that account’s terms |
| The repo’s voice studio | Train a voice model inside the project | The trained model on your own machine |
Two cautions that apply either way. Only clone a voice you have the right to use, meaning your own or one you have written permission for. And test the trained voice on a short script before you build a whole video around it, because a voice that sounds fine on one sentence can sound strange on a long read.
How to Drop Your Mascot Into Your Editing Workflow
The last step is the one that decides whether the project pays off: get the mascot, its expressions and its voice into the editor you already use and start cutting edits for it to react to.
Two of Kevin’s other open source tools slot in here naturally.
- OPEN VIDEO EDIT is a set of open source Claude Code skills that turn a raw talking head clip into an animated short, with captions, logos, faces, GIFs, b roll cards and counters. It renders locally with Python and ffmpeg. The skills include
split-animated-talking-head,longform-animated-talking-head,fireship-style-edit,fireship-style-short,split-animated-shortandreverse-engineer. - The mascot asset set from this repo feeds those edits with the character faces and reactions the cuts land on.
A worked example: record a two minute talking head, run the animated talking head skill to get the cut with captions and b roll cards, then place your mascot’s expression assets on the beats where a reaction sells the point. The output is a short with a recurring character, not a stranger reading a script. The launch video at youtube.com/watch?v=-QBJH_PK3pY shows the edited result and is worth watching before you spend an afternoon on your first attempt.
If you would rather automate the whole short form assembly, the launch video skill post covers that path, and the local video model guide by GPU VRAM helps if you want to render on your own card.
Is This AI Influencer Pipeline Actually Free
The repo is free and open source, and every step Kevin shows runs locally. Free here means no recurring subscription to a hosted avatar or voice tool, not zero cost in time.
What you actually spend:
- Time. Cloning a repo, installing Python dependencies and iterating on a mascot image takes an afternoon, not five minutes.
- Your machine. Python and your editor have to run on hardware you own. If your laptop struggles, the GPU VRAM breakdown tells you what local video work needs.
- An AI app. The workflow is built around Claude Code. Creator OS includes a hosted MCP server, a CLI and the Creator OS agent skills on every plan, so an agent can run your accounts without extra tooling. Details are in the Claude Code plugin docs.
What it does not cost is a recurring fee to a hosted avatar tool. If you make one video a month and never plan to reuse the character, a hosted service is the faster choice. If you want a mascot you own, rebuild and reuse across a channel, this pipeline is the one to learn.
AI Influencer With Claude FAQ
How do I make an AI influencer with Claude for free?
Clone the Claude Codes Influencers repo, run mascot.python with one uploaded image to build the character, generate the asset set with its facial expressions, then train a voice in the built-in voice studio. Drop the result into your editor and start building edits the mascot reacts to. Everything runs locally, so there is no recurring avatar fee.
Can Claude Code generate facial expressions for a mascot?
Yes. In this repo the expression set is part of the generated asset set, so you produce a range of faces for the character without drawing them by hand. You then pick the expression that fits each beat in the edit. Regenerate rather than repair if one comes out wrong.
Is the voice studio a free alternative to 11 Labs?
The video describes the built-in voice studio as a free version of 11 Labs. It is a local component of Kevin’s repo that trains a voice model for your mascot so the character can talk, and you keep the model file. ElevenLabs is a separate hosted service, so treat the two as different products that solve a similar problem rather than as identical output.
Do I need Python to build an AI avatar from one image?
Yes, the mascot step is a Python file, mascot.python, so you need Python installed and a machine that can run it. You don’t need to write Python yourself; you run the file and follow the repo instructions. If you have never set up a Python project, budget extra time for that first step.
HeyGen vs building an AI influencer with Claude: which should I pick?
HeyGen is a hosted avatar and video platform: fast to start, and your character lives in their account under their terms. Building with this repo gives you files you own and no recurring fee, at the cost of setup and local compute. Pick the hosted route if you need one video quickly; pick the repo if you want a reusable mascot for a whole channel.
Get started
Build the mascot first, then give it somewhere to post. Creator OS connects 11 platforms plus Skool in one place, with a REST API, a CLI and a hosted MCP server so a Claude agent can publish, reply to comments and DMs, and read analytics. The agent skills install with npx @creatoros/cli@latest init, and the tool list is at the MCP docs. For the bigger picture on running socials with an agent, read how to build a social media agent with Claude Code.
For a worked example, Social Agents at github.com/kevinbadi/social-agents is an open source agent harness that runs your socials on Creator OS. Start it with npm start creatoros social-agents, answer the brand interview, and it posts, replies and reports.
Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. New subscribers start for $0 with a free trial and can cancel anytime. Create your account at https://www.creatoros.ca/sign-up, then follow the walkthroughs on the KevBuildsApps YouTube channel to see each step run.
Get started for $0
Free trial for new subscribers. Cancel anytime.