Planar

Appearance

Overview ​

Planar brings local image generation and editing into one focused desktop workspace. The main screen combines prompt controls, a canvas, output history, model selection, and optional agent activity.

What Planar does ​

  • Generate images from a text prompt with txt2img.
  • Vary an uploaded or generated image with img2img.
  • Refine a design using instruction-based editing.
  • Manage local checkpoints and supporting model families.
  • Extend the pipeline with ONNX-backed ML plugins.
  • Connect agents through the embedded MCP server.

Extensible ML plugins ​

Planar includes a plugin system that uses ONNX Runtime as an inference engine for commonly used ML models, such as background removal and Real-ESRGAN upscaling. This provides a foundation for extending Planar with additional ML model capabilities over time without changing the core generation workflow.

Why Planar ​

There are many diffusion tools available, and each takes a different approach. DiffusionBee is approachable, but it uses its own model format. Draw Things is also compelling, but it is tied to the macOS ecosystem and its own model workflow. ComfyUI is powerful and flexible, but it requires manually setting up a Python environment and is more oriented toward research and highly customized workflows.

Planar takes inspiration from the focused UI of tools such as Fooocus, while aiming to provide a straightforward local workspace for standard image generation and editing. The goal is to work with established model ecosystems and keep the generation simple, we will look to workflow-specific tools for more complex tasks in future.

We believe local compute is good enough for standard images. Local video and 3D generation are possible areas of exploration, but they remain a stretch for typical local hardware.

stable-diffusion.cpp is Planar’s default generation engine. If you switch to Diffusers, Planar creates a managed Python virtual environment and installs the required Diffusers dependencies before using that backend. This keeps the default native workflow lightweight while still making the Python-based pipeline available when you need it.

A useful mental model ​

The workspace has three layers:

  1. Generation turns a prompt and model configuration into an image.
  2. Editing keeps a selected image active on the canvas and creates new versions.
  3. Project state stores generations and their embedded image assets. App-wide generation settings are stored separately.

Start with Getting started, or jump directly to Workflows.