Field guide

Local AI Image Generators: Privacy and Setup

For image generation on your computer, start by choosing a local model and a workflow that actually runs it locally. ComfyUI offers desktop and manual installation paths, but a workflow can also call an external image API. Running the interface on your computer alone does not establish where an image or prompt is processed. This guide helps you check that boundary before you use private source images.

By Unpacking the World · Published Oct 11, 2026 · Updated Oct 12, 2026

Choose an installation path before choosing hardware

The ComfyUI project documentation describes a desktop application for Windows and macOS and a manual installation path with broader hardware support. Its Windows portable build is another option, although the project directs regular users toward the desktop app. It also describes where model files belong for the portable installation. Last verified: 2026-10-12.

That is a map of installation choices, not a tested performance comparison. Before buying hardware, choose the specific model and workflow you want to run, then check their current memory and platform requirements against your machine. A generic claim that all image models need a particular amount of VRAM would be misleading: model files, settings, and acceleration paths differ. We have not run timing or quality tests on Windows, macOS, NVIDIA, AMD, or Apple hardware.

Check every node in the workflow

The project README also says ComfyUI can run locally. That does not mean every available workflow stays local. Its GPT Image 1 API node guide documents an API backed image operation that needs service access and credits. Last verified: 2026-10-12. Treat any node described as an API or partner service as a separate data and cost decision, even when the workflow canvas is local.

For a private image, write down what enters the workflow: prompt text, source image, mask, and output. Check the documentation for every nonstandard node and model. If the workflow requires an account, credits, or a remote API, do not describe it as offline processing. A local node graph and a remote API call can coexist in the same app.

A practical selection checklist

1. Decide whether your task is text to image, editing an existing image, or batch production. Those are different workflows and may need different models or nodes. 2. Check the project's current installation guidance for your operating system and hardware. Confirm the model's own file and memory requirements before making a purchase. 3. Download the application and model from the project's official sources. Record the exact version and model you selected; a recommendation without those details cannot be reproduced. 4. Review the workflow graph for external API nodes. For sensitive source images, use only a processing path whose data handling you understand and that your own policy permits. 5. Try a non-sensitive sample first. Judge setup effort, output suitability, and local resource use on your machine. We have not done this test for you.

The next step is to pick one documented local workflow and verify it with your own sample. If you need private text processing instead, see the desktop language model guide. The selection checks themselves require no purchase.