Wunjo CE keeps its v2.0.8 code behind a 4,096 star goal

What Wunjo CE actually packages for local media work

Wunjo CE is a desktop toolkit that runs generative media models on the machine sitting in front of you rather than uploading frames and audio to a hosted service. The project description names face swap, lip sync, control-guided removal of objects, text and backgrounds, restyling, an audio separator, voice cloning and video generation, and the topic tags add `controlnet`, `segment-anything`, `txt2video` and `img2video`. All of that sits behind one interface and one model directory, which is the practical difference from the single-purpose face-swap and video-diffusion utilities it resembles. Its primary language is `JavaScript`, so the shell that wires the pipeline together is frontend code while the models arrive as Python dependencies. That split matters for anyone extending it: the readable part is the glue, and the weights are not in the tree at all.

Four requirements files and no install command to copy

The install story turns thin at the repository root. Four requirements files sit side by side, and the README does not map them onto specific machines:

requirements_cpu.txt
requirements_gpu.txt
requirements_macos.txt
requirements_zluda.txt

Named for ZLUDA, the fourth file arrives with no explanation of which cards it targets or whether it is the intended route on Windows. No pip invocation, no clone step, no virtual environment command and no environment variable appears anywhere in the documentation. Two badges carry the rest: Installer links back into the repository, and Docker points at the `docker/` directory. Expect either a packaged build or a container, then a manual read of those four files to discover what the build pulls in. Sitting in the same badge row are Patreon and Boosty links, which fund the paid product rather than the build you are about to unpack. Python 3.10 is the only version constraint stated, and no RAM or VRAM floor is given for any of it.

The 4,096 star goal sitting inside the open-source promise

The README opens with 'Goal: 4,096 Stars on GitHub!' and states that once that milestone is reached, Wunjo CE's open-source code of v2.0.8 will be updated with the latest version changes. Until it is reached the tree holds one fixed snapshot, and the improvement described in the same paragraph is withheld behind a popularity target. The surrounding lines lean on it harder: 'Star the project on GitHub to help make Wunjo Pro open-source!' plus a year of full functionality for anyone who posts a video review and emails the link to [email protected]. Those incentives point away from the repository and towards the paid product, inside a project described as local and free. A later README paragraph argues the opposite case, holding that keeping the two products apart is a matter of clarity for users and of compliance with open-source licensing. A permanent wall and a promised merger sit in the same file, and nothing says which one governs when they meet.

Smart crop announced for CE and kept for Pro

Under a heading that reads 'Smart Crop and Cut for Video!' the v2.0.8 update describes video editing with text and smart cropping for any format, where a neural network selects the most interesting frame so a vertical clip can be turned horizontal and back again. Then, in the same paragraph: smart crop is only available in the Pro version. A headline capability of the newest Community Edition note is therefore described in detail and placed outside the edition that publishes it. Anyone stopping at the bold heading takes away a feature the download cannot produce, and the Pro sentence sits far enough down that paragraph to be skipped.

An API claimed for CE with no endpoint or key

One line of the same note says 'Also available API now in CE version.' Alongside the `public-api` topic tag, the capability is asserted twice and specified zero times. There is no endpoint, no request format, no authentication key, no port and no example call anywhere in the visible documentation. For a builder that gap is the whole story: the claim can be confirmed and nothing can be built on it. Automating the toolkit therefore means driving its own files rather than calling a service, a heavier maintenance commitment than the two letters carry.

Docker, models, portable and tools at the root

Five top-level directories matter to a first run:

docker/
models/
portable/
tools/
example/test/

A Docker badge points at `docker/`, `portable/` sits beside it in the tree, and `example/test/` holds sample inputs to try before anything else. What none of the text explains is what lives in `models/` or `tools/`, or which weights Community Edition expects at first launch. A Hugging Face badge points at a model page under the same author name, which acts as a pointer rather than a manifest. The repository is MIT-licensed and the last push was on 2026-08-25, and that date is the only freshness signal a reader gets.

Deepfake tags and governance files nobody links

Topic tags including `deepfake`, `face-swap` and `voice-clone` name exactly what a user can produce and, by extension, which consent obligation arrives with the face-swap and cloning paths. Sitting in the tree are `SECURITY.md`, `CODE_OF_CONDUCT.md`, `CONTRIBUTING.md` and `PULL_REQUEST_TEMPLATE.md`, and the README links none of them; the wiki is the only documentation pointer in the header. A security file nobody is sent to makes a thin disclosure path. Documentation sits outside the repository too, since a wiki link, an issue tracker, a discussion board and a YouTube tutorial playlist are all reachable from the header while no tutorial is shipped inside the tree. The language list claims English, Spanish, Chinese, Korean and Russian, yet the single other README present is `README_ru.md`, labelled RU v2.0 rather than v2.0.8.

The real alternative is the paid product at wunjo.online

A genuine alternative exists, and it is not another open-source face-swap repository but the commercial product at wunjo.online, with the README stating the difference rather than hinting at it: that product has a different codebase, does not use or derive from this repository, and may carry commercial features not available here. Smart crop is the concrete case, named as Pro-only inside the Community Edition note itself. So the trade is CE against a feature list nobody can confirm in advance, and Pro against a codebase with no shared history and no condition attached to its progress. Roadmap decisions run through boosty.to/wunjo, where CE users vote on which features migrate, even though the header also links an issue tracker and a discussion board for the same purpose. For a pipeline that needs a documented HTTP contract or a pinned, reproducible dependency set, neither side of that pair is ready, and this tree is the wrong tool until the API line grows an endpoint.

Editorial conclusion

Wunjo CE suits a creator who wants face swap, lip sync, background removal and voice cloning on their own hardware, treating the tree as a starting point rather than a finished product. It does not suit anyone who needs smart crop, the Pro feature set, or a written API contract before writing a client, since the README places all three outside CE. Verify three things first: how your machine maps to the four requirements files at the repository root, whether the v2.0.8 code you download already carries the updates the star goal defers, and whether the face-swap and voice-clone paths fit the consent terms of whoever appears in the footage.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. wladradchenko/wunjo.wladradchenko.ru on GitHub