What This Is
HivisionIDPhotos is an intelligent AI ID photo maker: a complete AI model workflow that recognizes people and cuts them out of photos taken in all kinds of settings, then generates standard-spec ID photos and 6-inch layout sheets. Its core selling point is “lightweight” — the cutout runs fully offline, with fast inference on CPU alone; no graphics card needed, and no photo ever goes to an online service.
What it can do includes: generating standard ID photos to different size specs and 6-inch layout sheets (later expanded with 5-inch photo paper, A4, 3R, and 4R layout sizes); changing the background color (common base colors built in, plus custom HEX values); face beautification; face rotation alignment; downloads default to 300 DPI, and layout sheets come with crop marks for easy print-and-trim. The roadmap also lists “smart formal attire swap,” which as of the v1.3.1 provided here is still in development — don’t treat it as an existing feature.
In essence it’s an algorithm library plus three ways to use it: a local Gradio graphical interface (app.py, operated in a web page, good for individuals); Python command-line inference (inference.py, good for batch processing); and an API service or Docker deployment (deploy_api.py, docker-compose.yml, good for standing up your own online ID photo service to share with family, friends, or a team). If you’d rather not install anything, you can try the results first on the online demo pages hosted on SwanHub, HuggingFace Spaces, and ModelScope (addresses are in the tool’s documentation).
Who It’s For
- People who want an ID photo setup where “photos never leave the house”: unlike online background-swap websites, this pipeline runs from cutout to layout entirely on your own machine, and your front-facing photo never passes through a third party’s hands;
- People with a computer who are willing to install a Python environment once: the Gradio interface is friendly enough for non-technical users — once the dependencies are in, it’s just opening a web page and clicking;
- People who want to self-host an ID photo service for a community, a dorm, or a small team: the API-plus-Docker deployment exists precisely for this scenario, and a cheap server or NAS is enough to run it;
- Developers: the core logic lives in the hivision/ package under the Apache-2.0 license — no obstacles to modifying or integrating it.
Cases where it’s not a fit: if you want zero installation and instant output on a phone, its online demo entry point is the better choice; if you have extreme custom requirements for ID photo specs (say, an exam registration system’s oddball pixel requirements), test once with the demo before deciding.
How to Use It
The easiest path: try the official online demo first to confirm you like the results, then decide whether to go local. The standard local deployment: download the source package from this site’s “Download and Verify” section and unpack it (this site provides the source package), install the dependencies in requirements.txt, then pick one of three — python app.py to run the Gradio web version; python inference.py for command-line inference; python deploy_api.py or docker compose up to start the service. The first run downloads the cutout and face-detection model files; after that, inference is offline throughout.
Making an ID photo in the interface goes: upload a casual photo → pick a size spec and background color → fine-tune (face rotation alignment can be enabled) → download the single photo or the layout sheet. When the layout size is set to 6-inch photo paper, it automatically works out how many photos fit on a sheet and draws the crop marks — print directly on a home photo printer.
A note on version choice: this site currently provides v1.3.1; new versions are added once verified.
Safety Notes
- This site provides the source package; before use, verify its integrity against the SHA256 on this page (see the “Download and Verify” section above).
- Control the exposure of any self-hosted deployment: if you open the service to the public internet with deploy_api.py or Docker, anyone can upload pictures to your server; keep it on an internal network or behind authentication, and don’t let a public service run unprotected.
- The model files are downloaded from the network on the first run — a normal dependency fetch; if you’re strict about provenance, run it once in a networked environment to confirm the model hashes, then use it offline — its inference design natively supports fully offline operation.
- Finished ID photos are often tied to identity documents; after export, keep custody of the files — how you store, transfer, and use them is on you — and this point applies no matter which tool you use.
- For photos produced by features like attire swapping or AI beautification, the issuing authority’s requirements govern formal document applications — don’t push your luck with over-processed photos.
Editor’s Review
I included it mainly for the phrase “fully offline CPU inference”: most AI ID photo solutions either upload your photo to the cloud or devour a graphics card — this one has neither flaw and runs on an ordinary laptop, which makes it the most worry-free foundation for self-hosting an ID photo service. Feature completeness is high too — from cutout, background swap, and alignment to layout and printing, the chain is complete, and the API and Docker options push the self-hosting barrier very low.
Caveats: it’s an algorithm library plus tooling, not an out-of-the-box mobile app; non-technical users should simply use the official online demo and go local only after thinking it through. The bundled documentation updates quickly, but formal releases come slowly — after v1.3.1, the latest development builds have already accumulated a good deal of unreleased goodness, so using a fixed version means accepting features slightly older than the newest development build. Also, beautification and skin whitening are “face-processing” features; use them conservatively in formal ID photo settings.
Version History
- v1.3.1 (released 2025-01-21): this site currently provides v1.3.1; new versions are added once verified.
FAQ
Why does this project only provide a source package? This project offers only a zip download — the unpacked code matches that version exactly, which is fully sufficient for everyday use and deployment.
How is it different from IDPhotoStudio above? Both are ID photo tools, but different in form: IDPhotoStudio is a packaged desktop app (macOS/Windows installers) for individuals to click through; HivisionIDPhotos is an algorithm library plus server-side solution (Gradio/API/Docker) aimed at self-hosting and secondary development, with a different cutout engine and runtime as well. Pick the former to install and use right away; pick the latter to build your own service or integrate.
Does it need a graphics card? No. The official positioning is lightweight cutout, with fast inference on pure CPU; the Docker and API services also have very modest machine requirements.
Why does this project only provide a source package? This project offers only a zip download — the unpacked code matches that version exactly, which is fully sufficient for everyday use and deployment.
Does using it upload my photos anywhere? Local runs (Gradio/command line) do all inference on your own machine and need no internet; only the first-run model download needs a network. If you use the official online demo or a self-hosted public-facing service, the photos naturally pass through the corresponding server — that’s the essential difference between the two paths, so choose per your needs.
Why does this site’s version look older than the newest? This site provides v1.3.1; new versions are added once verified. The benefit of a fixed version is clarity: what each version is, its update date, and its checksum are all plainly stated — ready to use, with no version guessing.
Sources
This page was compiled from the tool’s own documentation and files; for features and parameters, the documentation bundled with the package is authoritative.