jingyi-wu-richael/rachel-digital-human-production — explained in plain English
Analysis updated 2026-05-18
Generate a short preview talking-head video from a script, portrait, and voice sample before committing to a full render
Automate a repeatable pipeline that combines voice cloning with image-to-video generation
Track the status of voice, asset, and video generation jobs across a production run
Add a Codex skill for authorized digital human video production to an existing content workflow
| jingyi-wu-richael/rachel-digital-human-production | google-cloud-ai/alphaevolve-on-googlecloud | humanmllm/swim | |
|---|---|---|---|
| Stars | 75 | 75 | 75 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | hard |
| Complexity | 3/5 | 4/5 | 5/5 |
| Audience | developer | developer | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires your own paid MiniMax and HeyGen API keys and rights to the voice, portrait, and script used.
rachel-digital-human-production is a Codex skill, meaning a packaged set of instructions and helper scripts for the Codex AI coding assistant, that walks through producing a talking head video of a digital human. It combines two paid third party services: MiniMax, used for cloning a voice and generating narration from a script, and HeyGen, used to turn a still portrait photo into a talking video driven by that narration. The skill is built as a repeatable workflow rather than a one-off script. Before making any paid API calls, it checks that the script, portrait image, and voice sample files are valid. It then generates a short 15 second preview video first, and only moves on to producing a full 1080p video after the user explicitly approves the preview. Throughout the process it tracks identifiers for the voice, asset, and video jobs, along with their status, inside a file called work/job-state.json, so a person can see where a production run stands. To use it, you copy the skill's folder into your local Codex skills directory and then invoke it by name inside a Codex task, for example asking it to make a 15 second preview first. The project expects a specific folder layout, with input files like the portrait, voice source, and script kept separate from working files and final output videos. You also need to set your own MiniMax and HeyGen API keys as environment variables, the README is explicit that the skill itself does not include any API access, paid credits, or legal permission to use anyone's voice or likeness, and that real API keys and generated videos should never be committed to the project. A couple of small Python helper scripts are included for initializing the job state file and for running the preflight checks on the script, portrait, and voice files before the paid steps begin. The README states this is meant only for authorized production work, where the person running it already has the rights to the voice, portrait, and script involved. It is released under the MIT license.
A Codex AI skill that automates producing an authorized digital-human talking-head video using MiniMax voice cloning and HeyGen video generation.
Mainly Python. The stack also includes Python, Codex, MiniMax API.
MIT license: free to use, modify, and distribute, including commercially, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.