Hosted database viewer

Run a new analysis on your computer

This website displays saved results. New analyses run in the local app with a separate analysis backend.

01 Check your computer

What you need	Computer	Additional setup
View saved results	Windows or macOS; Linux also supported	Modern browser. The installer supplies Node.js when needed. No Python, Hugging Face account, or AI GPU required.
Run analyses	Windows or Linux with an NVIDIA GPU recommended	Python environment, compatible NVIDIA driver, Ollama, and access to the SAM 3 model on Hugging Face. The backend installer checks compatibility.
CPU-only analysis	Windows or Linux, using the explicit CPU profile	Much slower and dependent on model memory requirements. GPU-only stages must remain disabled.
macOS	Supported for the viewer	The complete backend pipeline is not supported on macOS. Use a compatible Windows/Linux machine for analysis.

AI model downloads are large and require free disk space, RAM, and GPU memory. The backend installer shows its plan before installation; available memory determines which model sizes are practical.

02 Install the viewer

Open the separate image-forensics-frontend repository supplied with this project. Choose Code → Download ZIP, then extract the ZIP into a normal folder before opening the installer.

Windows: double-click Install.cmd. macOS: open Install.command. If macOS blocks a downloaded file, follow the included README instructions for that file.
Follow the installer's checks and progress until it reports completion. It downloads the viewer's dependencies and builds the app.
Open Start.cmd on Windows or Start.command on macOS. Keep its terminal open while you use the app.
Open http://127.0.0.1:3100 in your browser. Select View database to explore the included results.

03 Add the analysis backend

Open the separate image-forensics-backend repository supplied with this project. Choose Code → Download ZIP, then extract the ZIP into a normal folder before opening the installer.

On Windows, open Install.cmd. On Linux, run bash install.sh from the extracted backend folder. Choose the standard NVIDIA profile, or explicitly choose CPU if you do not have a compatible GPU.
Install Ollama ↗ when prompted. Ollama runs the vision language model locally; downloading the app and downloading a model are separate steps.
Create a Hugging Face account ↗ and request access to facebook/sam3 ↗. Accept its terms with that account. Approval is controlled by the model owner.
Create a token with read access in Hugging Face settings ↗ and enter it only in the backend's local setup prompt. Never put the token in this website or commit it to Git.
Let the backend installer check model access and install the selected dependencies. Keep optional native forgery and synthetic-analysis extras disabled unless you have completed their separate setup.

About hardware, drivers, and model downloads

A GPU's memory is different from your computer's RAM. An NVIDIA name alone does not guarantee enough memory for the selected vision and segmentation models. Close other GPU-heavy applications and use the backend's compatibility check before downloading.

The PyTorch GPU package and your NVIDIA driver must be compatible. Install or update the driver using NVIDIA's official instructions; the installer does not replace system drivers. CPU mode cannot make GPU-only add-ons compatible.

Hugging Face access errors usually mean the account has not been granted access, the model terms have not been accepted, or the local token belongs to a different account. Ollama model files and Hugging Face caches stay on the machine that runs the backend.

04 Start an analysis

Start the backend using Start.cmd on Windows or bash start.sh on Linux. It starts Ollama when needed.
Start the frontend on the same computer and open the local New analysis page.
Choose an image, select the stages you need, and select Start analysis. Progress and previews appear while it runs.
Completed results download into the frontend's own data folder. You can view them after stopping the backend.

Starting your backend does not connect it to this hosted website. Use the local frontend for analysis. To update the hosted database, publish a new static build with your selected saved datasets.

The local app still cannot find the backend

Keep the backend terminal open and check its last error. The default API address is http://127.0.0.1:8766. The frontend and backend must run on the same computer with their default ports, or the local frontend's ANALYSIS_API_URL must match the backend's port.

Return to New analysis and select Try again. Saved results remain available even when the backend is stopped.