what is appnz-character-cog-template fr?

This repository is a starter template for building a specific kind of hosted AI endpoint on a platform called app.nz, aimed at things like animating a portrait, syncing lips to audio, generating voice, or running 3D character models. It is not a working demo of any actual model itself. Instead, it is a minimal, tested scaffold you fill in with a real model later, and its automated tests run without downloading any model or using any GPU time. The template runs a small FastAPI server on port 5000 with a fixed contract: a health check endpoint reporting setup or ready status, a schema endpoint for app.nz's generated interface, and a predictions endpoint taking a portrait plus audio or a driving video, returning a standard success or error result. Before you add a real model, it just returns placeholder deterministic responses. To turn it into something real, the README walks through a checklist: pin the exact version and software stack of the upstream model you are wrapping, load its weights once and get it fully warmed up before marking the service ready, return either a public link to the generated output or the output embedded directly as data, and set clear limits on things like resolution, video length, and batch size so a single request cannot use unbounded resources. It also asks you to actually measure how long the container takes to start, load, and run inference, and to record the peak memory it uses, so the app.nz listing states a truthful hardware requirement rather than guessing. There is also a provenance checklist meant to be filled in before publishing any model built from this template, covering things like where the upstream code and model weights came from, what license they carry, whether you are actually allowed to run inference on them commercially, and any caveats about the data the model was trained on. The README explicitly flags that character animation systems can be misused for impersonation, and asks anyone deploying one to make sure they have rights to the likeness or voice involved and to label synthetic output clearly. Pushing to the main branch runs the tests, builds a container image, and publishes it automatically. The template itself is released under the MIT license, though any real model code or weights you plug in keep whatever license they came with.