Local-first Β· open core Β· runs everywhere

Every open image model.
Running on your machine.

Planar is a local studio for image generation β€” from a one-click result to iterative, instruction-driven refinement. Diffusion and beyond, on the hardware you already own.

macOSApple Silicon
macOSIntel
⊞Windowsx64
🐧LinuxAppImage
β˜… Star on GitHub
macOS Β· Apple Silicon & Intel Windows Β· x64 Linux Β· AppImage Free
Untitled Project juggernautXL Β· stable-diffusion.cpp
juggernautXL Β· loaded β–Ά Generate
Prompt
a leviathan rising from black water, mtg card art, 1990s ink▏
Photoreal Illustration 3D Anime + preset
2565127681024
Runs the open ecosystem you already use
Diffusers stable-diffusion.cpp SDXL FLUX ControlNet LoRA Qwen-Image-Edit
One prompt Β· then keep going

Generate a result. Then talk it into shape.

Most images aren't right on the first roll. Planar's refine loop lets you describe a change in plain language β€” "make the sword glow blue" β€” and applies it with an instruction-edit model, without re-rolling the whole image.

Generate Refine
Untitled Project generate mode
juggernautXL Β· loaded β–Ά Run
Prompt
a leviathan rising from black water, mtg card art, 1990s ink▏
Photoreal Illustration 3D Anime + preset
2565127681024
Built for the whole open stack

From a first generation to a research rig.

The same app handles a one-line prompt and a deep model-management workflow β€” and it does it on every platform you work from.

Truly multiplatform

One codebase across Apple Silicon, Intel Mac, Windows x64 and Linux. Switch machines without relearning the app or rebuilding your setup.

macOS Β· Windows Β· Linux

Sandboxed plugin pipeline

Extend Planar with WASM plugins β€” no network or filesystem access by default. Load one, activate it, and batch-run it across every image in your gallery.

WASM Β· no WASI Β· plugin API

Experimental research models

Because it speaks Hugging Face Diffusers natively, Planar runs the newest research checkpoints the day they drop β€” not months later.

HF Diffusers native

Beyond diffusion

Background removal ships built in. Upscalers, segmentation and restoration models run as plugins on the same ONNX runtime.

ONNX runtime Β· bg removal bundled

100% local & private

Generation runs on your hardware. No uploads, no prompt telemetry, no queue β€” your work never leaves the machine.

No cloud Β· no account

Agent & MCP ready

Drive Planar from an agent session over MCP β€” queue jobs, poll status, stream results β€” while the same queue powers the UI you click through by hand.

MCP server Β· async job queue
Models

Bring any checkpoint. Planar figures out the rest.

Point it at a folder or a Hugging Face repo. Planar detects the architecture, picks sane VAE and sampler defaults, and loads β€” swapping backends without leaving your project.

  • Diffusers & stable-diffusion.cpp backends, hot-swappable
  • Auto VAE, sampler and resolution detection
  • LoRA stacks and ControlNet built in
Model Β· backend
stable-diffusion.cppβ–Ύ
juggernautXL_v2●
Sampler
Euler A ↩
DPM++ 2M ●
LCM ↩
Inference engines

Two backends. One switch.

Run the same project on whichever engine fits the job β€” a lean quantized runtime or the full research stack. Pick per project, swap anytime, no reinstall.

C++
stable-diffusion.cpp
lean Β· quantized Β· low-VRAM
  • GGUF quantization β€” run big models on small GPUs
  • CPU, Metal, CUDA & Vulkan acceleration
  • Fast cold starts and a tiny memory footprint
Best for laptops & quick iteration
⇄
swap
anytime
HF
Hugging Face Diffusers
maximal Β· research-ready
  • Every architecture β€” SD, SDXL, FLUX and new research
  • Day-one support for checkpoints as they drop
  • Full pipeline: LoRA, ControlNet, schedulers
Best for breadth & the bleeding edge
Beyond diffusion Β· WASM plugins

Bring your own image models.

Planar runs any ONNX image model as a plugin, sandboxed with no network or filesystem access. One ships in the box; drop in the rest.

BGBackground removerbundled
clean alpha matte
One-click subject isolation with a crisp alpha channel β€” ready for compositing nodes. Ships with the app, u2net-based.
DTPS1 / Bayer ditheringexample plugin
ordered dithering
A reference WASM plugin that applies retro ordered dithering β€” the template for building your own image-processing nodes.
+Add ONNX model
+
Point Planar at any .onnx graph, wrap it with the plugin SDK, and it shows up as a loadable, batch-runnable plugin.
Open core

Free to create. Pro when you scale.

The full studio is free, forever. Pro adds team and power-user features down the line β€” you’ll never lose access to what’s open today.

Planar
Free / forever
The complete local studio.
  • Generate & iterative refine modes
  • Both open backends & every model they support
  • Plugins & custom nodes
  • Every platform
Download
Pro β€” coming later
Soon
For teams and heavy pipelines.
  • Shared model & plugin library
  • Batch & queue orchestration
  • Priority model support
  • Everything in Free
Join the waitlist

Generate locally. Refine until it's right.

Download Planar and make your first image in under a minute β€” no account, no cloud.

macOSApple Silicon
macOSIntel
⊞Windowsx64
🐧LinuxAppImage
β˜… Star on GitHub