Skip to content
Back to search

pixel-avatar

SkillWorks with Claude CodeGitHub

Free Claude Code skill: turn one photo of you into a talking pixel avatar and make vertical videos

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/skillahollic--pixel-avatar/badge.svg)](/listings/skillahollic--pixel-avatar)

HTML

<a href="/listings/skillahollic--pixel-avatar"><img src="/api/listings/skillahollic--pixel-avatar/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

1. Download the zip file and unzip it. Keep the folder in a place you can find, such as Downloads. 2. Double-click `Install-Mac.command`. 3. The file has no Apple signature, so macOS blocks it the first time. macOS shows a message that it cannot check the file. Click **Done**. 4. Open **System Settings**, then **Privacy & Security**. Scroll down to the message about `Install-Mac.command`. Click **Open Anyway**. macOS asks for your Mac login. Type it into the macOS window. The script never sees it. 5. A Terminal window opens. Read the list of tools and sizes. Type `y` and press Return. 6. Answer the question about a paid Claude plan with `y` or `n`. 7. The AI tool opens by itself in the same window and starts the setup.

To avoid the macOS block, open Terminal, type `bash ` with a space, drag `Install-Mac.command` into the window, and press Return.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 127 lines
  1. # Pixel avatar skills
  2. This folder has two skills for an AI coding tool. The skills turn one photo of you into a talking pixel-art avatar with poses. The avatar speaks with a clone of your own voice, or with your own recording, and its mouth follows the words (Rhubarb Lip Sync). The skills then make vertical videos for Reels and Shorts, with graphics from a style pack, an Instagram caption, and a YouTube Shorts title, description and tags. Everything runs on your computer and is free.
  3. - `pixel-avatar-setup`: the pose sheet prompt for Google AI Studio, the voice, the tools, the style pack, and optional Hypit.
  4. - `pixel-avatar-video`: from an idea, a video you like, or your own video, to the finished MP4 and the post texts.
  5. ## What you need
  6. - A Mac, or a Windows 10 or 11 PC.
  7. - Free disk: about 3 GB at least. The installer shows the exact size for your computer before it starts.
  8. - A paid Claude plan for Claude Code. If you do not have one, the installer sets up OpenCode with a free model.
  9. - A Google account for Google AI Studio (the Nano Banana Pro image model), and a clear photo of yourself.
  10. - For the voice clone: a clip of 5 to 15 seconds of your own voice. For the other choice: a microphone or a phone to record each script.
  11. You do not need to install Python. The installer does it.
  12. ## Install on a Mac
  13. 1. Download the zip file and unzip it. Keep the folder in a place you can find, such as Downloads.
  14. 2. Double-click `Install-Mac.command`.
  15. 3. The file has no Apple signature, so macOS blocks it the first time. macOS shows a message that it cannot check the file. Click **Done**.
  16. 4. Open **System Settings**, then **Privacy & Security**. Scroll down to the message about `Install-Mac.command`. Click **Open Anyway**. macOS asks for your Mac login. Type it into the macOS window. The script never sees it.
  17. 5. A Terminal window opens. Read the list of tools and sizes. Type `y` and press Return.
  18. 6. Answer the question about a paid Claude plan with `y` or `n`.
  19. 7. The AI tool opens by itself in the same window and starts the setup.
  20. To avoid the macOS block, open Terminal, type `bash ` with a space, drag `Install-Mac.command` into the window, and press Return.
  21. ## Install on Windows
  22. 1. Download the zip file. Right-click it and select **Extract All**. Do not run the file from inside the zip view.
  23. 2. Open the extracted folder. Double-click `Install-Windows.cmd`.
  24. 3. The file has no Microsoft signature, so Windows can warn you. If you see **Windows protected your PC**, click **More info**, then **Run anyway**. If you see an **Open File - Security Warning** window, click **Run**.
  25. 4. A black window opens. Read the list of tools and sizes. Type `y` and press Enter.
  26. 5. Answer the question about a paid Claude plan with `y` or `n`.
  27. 6. The AI tool opens by itself in the same window and starts the setup.
  28. ## After the AI tool opens
  29. This part is the same on both systems.
  30. 1. **Claude Code only:** your browser opens so that you can sign in if you are not signed in. Claude Code then asks if you trust this folder. Select **Yes, I trust this folder**.
  31. 2. The AI asks about your photo, one question at a time.
  32. 3. The AI gives you one prompt and the steps for Google AI Studio: model **Nano Banana Pro**, aspect ratio **16:9**, resolution **4K**, your photo attached. The model does not always obey 16:9; another shape (for example 3:2) also works, because the script finds the 12 cells from the thin lines between them. The result is one sheet of 12 pixel-art cells of you: the base pose, four mouth shapes, closed eyes, and six poses (pointing, shrug, thumbs-up, fist pump, wave, arms crossed).
  33. 4. Download the image or copy it. Go back to the AI and type `done`. If pasting fails, drag the file into the window.
  34. 5. The AI asks: clone your voice, or send a recording for each script? It scans your computer and installs the best free local voice clone for it, or nothing for the recording choice. This can take 30 minutes or more on a slow connection.
  35. 6. The AI cuts the sheet into poses and mouth frames, and shows you a preview.
  36. 7. You pick a style pack: Checkpoint Lab (default), Vox Explainer or Kallaway from Nate Herk's Herk Edit Kit, or your own colours.
  37. 8. The AI offers Hypit (optional). It installs only if you say yes.
  38. 9. Give the AI a video idea, a video that you like (a file or a link), or your own video. It writes the script with you, makes the video, and writes the Instagram caption and the YouTube Shorts texts next to it.
  39. To come back later, double-click the installer file again. It does not repeat finished steps.
  40. ## Free models
  41. OpenCode lists GLM-5.3-Flash Free and DeepSeek V4.1 Flash Free as free with zero data retention on [its model page](https://opencode.ai/docs/console/models/). Free models are "for a limited time" and the list changes. We have not run these skills end to end on a free model.
  42. ## The voice for your computer
  43. | Your computer | Voice clone | Licence |
  44. |---|---|---|
  45. | NVIDIA card with 8 GB or more | Qwen3-TTS 1.7B Base, on the card | Apache-2.0, commercial use OK |
  46. | NVIDIA card with 4 GB, Apple Silicon Mac with 8 GB, or a Windows PC with 16 GB | Qwen3-TTS 0.6B Base (on a Mac or with no NVIDIA card: on the processor, about 20 to 60 seconds for each line) | Apache-2.0, commercial use OK |
  47. | Intel Mac, or less memory | No clone. Record each script | Your own audio |
  48. Clone only your own voice. The recording choice works on every computer. Higgs TTS 3, Breeze TTS 2 and YuE2-3B stay opt-in only, with their non-commercial terms (see the table below).
  49. ## Style packs
  50. - **Checkpoint Lab** (default): this kit's own components and 3D scenes.
  51. - **Vox Explainer** and **Kallaway**: full-screen cards from [Nate Herk's Herk Edit Kit](https://github.com/nateherkai/herk-edit-kit) (MIT). This folder includes the style library (about 4 MB, commit `10511e5`), not the brand files of the AI Automation Society. See `kit/styles/herk-edit-kit/SOURCE.md`. The cards load their fonts from Google Fonts when the video renders.
  52. - **Custom**: your own colours on the Checkpoint Lab components.
  53. ## Hypit (optional)
  54. [Hypit](https://github.com/hypit-ai/hypit) turns a video that performs well into a template you can reuse. The setup offers it and installs it in your home folder only if you say yes, with `npx skills@1.7.6 add hypit-ai/hypit -g`. This folder does not include Hypit. Its licence: free to use on your own computer, also for client work; you may not resell it, bundle it, or run it as a service for others; its name and logo must stay. The videos you make are yours.
  55. ## What it costs
  56. The software in this folder is free. Your computer, your time, your Google AI Studio limits, your AI plan, and ElevenLabs or Suno, if you pick them, are yours to check. The ElevenLabs and Suno free plans do not permit commercial use. Some of Hypit's optional AI models are paid services.
  57. ## Please read
  58. - **Make each video different.** YouTube does not monetize content that looks mass-produced from a template. This can affect a whole channel. A recurring host is fine. The idea of each video must be yours.
  59. - **No promises.** Nothing here guarantees views, followers, or income.
  60. - **Label AI content.** See [AI-DISCLOSURE.md](AI-DISCLOSURE.md).
  61. - **Use only your own photo and your own voice.** Do not use another person's likeness or voice, a celebrity, or a character from a franchise.
  62. - **Use other people's videos as a structure only.** Do not copy their words or graphics.
  63. ## What the installer and setup add
  64. Nearly all tools stay in `.vtuber/` inside this folder. Delete that folder to delete them. Claude Code or OpenCode go in your home folder on a Mac. On Windows, Claude Code goes in your home folder and OpenCode stays in this folder. HyperFrames and ONNX Runtime keep small settings files in your home folder. Your avatar, voice clip and videos are in `my-vtuber/` and `videos/`. Hypit, if you say yes, goes in your home folder. Each download is a fixed version, checked against a checksum in `vtuber.py`, and uv and npm use package versions from a fixed date.
  65. | Tool | Licence | Source |
  66. |---|---|---|
  67. | [HyperFrames](https://github.com/heygen-com/hyperframes) | Apache-2.0 | npm |
  68. | [GSAP](https://gsap.com) | GSAP's own licence. It is free to use and not open source. See [gsap.com/licensing](https://gsap.com/licensing) | npm |
  69. | [ACE-Step 1.5](https://github.com/ace-step/ACE-Step-1.5), music, on NVIDIA 6 GB+ or Apple Silicon | MIT. You can use the music commercially | GitHub commit, Hugging Face fixed revision |
  70. | [YuE2-3B](https://huggingface.co/m-a-p/YuE2-3B), music, opt-in only, NVIDIA 24 GB+ | **CC BY-NC 4.0, non-commercial only.** You download it from its Hugging Face page. This folder does not include it | Hugging Face, fixed revision |
  71. | [Rhubarb Lip Sync](https://github.com/DanielSWolf/rhubarb-lip-sync) | MIT | GitHub release |
  72. | [Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M) via [kokoro-onnx](https://github.com/thewh1teagle/kokoro-onnx) | Apache-2.0 | GitHub release |
  73. | [Qwen3-TTS](https://github.com/QwenLM/Qwen3-TTS) 1.7B and 0.6B Base, the voice clone | Apache-2.0 | Hugging Face, fixed revision |
  74. | [PyTorch](https://pytorch.org), for the voice clone | BSD-3-Clause | PyPI or pytorch.org, fixed version |
  75. | [yt-dlp](https://github.com/yt-dlp/yt-dlp), only when you give a video link | Unlicense | PyPI, fixed version |
  76. | [skills CLI](https://github.com/vercel-labs/skills), only if you say yes to Hypit | MIT | npm, fixed version |
  77. | [Higgs TTS 3](https://huggingface.co/bosonai/higgs-tts-3-4b), opt-in only, needs typed confirmation, on NVIDIA 24 GB+ and 64 GB memory | Boson's Research and Non-Commercial licence with a Creator Use Grant. You can monetize only on your own channel, with a prominent credit. The kit adds the credit. The grant does not cover a product or service for others. You download the weights from Boson's page. This folder does not include them. Runtime from [Higgs_v3-TTS-ComfyUI](https://github.com/Saganaki22/Higgs_v3-TTS-ComfyUI) (MIT) | Hugging Face, fixed revision |
  78. | [Breeze TTS 2](https://huggingface.co/BreezeBlue/Breeze-TTS-2), opt-in only, on NVIDIA 12 GB+ | Weights and audio: **non-commercial only**. You download the weights from BreezeBlue's page. This folder does not include them | Hugging Face, fixed revision |
  79. | [faster-whisper](https://github.com/SYSTRAN/faster-whisper) | MIT | PyPI, Hugging Face |
  80. | [uv](https://github.com/astral-sh/uv) | MIT or Apache-2.0 | GitHub release |
  81. | [Node.js](https://nodejs.org) | MIT | nodejs.org |
  82. | FFmpeg, built by [Martin Riedl](https://ffmpeg.martin-riedl.de) (Mac) and [gyan.dev](https://www.gyan.dev/ffmpeg/builds/) (Windows) | GPL | Those builders' releases |
  83. ### Included in this folder (`kit/`)
  84. This project's MIT licence covers the scene kit (components, motion, design tokens). The folder also includes:
  85. - **Fonts:** Archivo, Martian Mono and Reenie Beanie, under the SIL Open Font License. See `kit/licences/`.
  86. - **three.js** (MIT) for the optional 3D scenes, with a heavy cut of Archivo (SIL Open Font License). See `kit/licences/`.
  87. - **Style library** from Nate Herk's Herk Edit Kit, MIT (Copyright 2026 Nate Herk) with its use permission. See `kit/styles/herk-edit-kit/`.
  88. - **Logos:** a small set from [Lobe Icons](https://github.com/lobehub/lobe-icons) (MIT). See `kit/licences/`. The logos are their owners' trademarks. Use them only to name the company or product a video is about.
  89. The Mac build of Rhubarb is for Intel, so Apple Silicon Macs need Rosetta from Apple. Without Rosetta, the mouth follows the loudness of the voice.
  90. The kit is English only for now. This kit has no personal support.
  91. MIT licence: see [LICENSE](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.