By Lucas Desfossez · Rawrr project operator
To run AI locally, install a compatible runtime and model on your own computer, check that the model fits your hardware, and run inference there instead of sending the job to a hosted API. Start with one small workflow before building a production setup.
1. Decide what you want to make
A local chat assistant and a local image generator need different models and runtimes. Designers may need image generation, editing, video or 3D workflows rather than a language model alone. Choose one repeatable task: for example, exploring a concept with a reference image you own.
2. Check your GPU, memory and storage
Record your operating system, GPU model, dedicated GPU memory (VRAM), system RAM and free disk space. Compare those with the exact model and runtime documentation. Image size, video length, precision and model size all affect memory use. There is no single minimum specification for all local AI. Start with a smaller supported model and resolution if memory is tight.
3. Use a maintained runtime and trusted model
Follow the runtime’s official installation instructions and check supported hardware. Read the model licence before commercial use. Download weights from their publisher or a trusted source; avoid running unknown installation scripts or extensions. For a concrete image-workflow starting point, see the official ComfyUI documentation. This is an independent tool, not a Rawrr installation guide.
4. Run a non-confidential test
Use a public or synthetic input. Record the model version, settings, resolution, generation time and peak memory use. Check the result for visual defects before increasing complexity. Keep a known-good workflow so that later model or driver changes can be compared.
5. Verify the data boundary
Local installation alone does not prove that every node runs locally. Inspect remote API nodes, telemetry, extensions, sync folders and network requests. After necessary downloads, test the intended workflow with network access disabled where practical. Keep device security and backups in scope.
Where Rawrr fits
Rawrr is a local-first creative platform in development for image, video and 3D workflows, with LoRAs part of its creative direction. It is not available as a public download yet. The planned closed beta will help establish supported hardware and workflows. Local inference avoids a third-party per-generation API bill, but hardware, power and licences still matter.