Models & backends

Planar supports two local generation paths. You had adjust the backend in the settings panel and use either gguf or safetensors checkpoints.

stable-diffusion.cpp

The stable-diffusion.cpp backend is the primary native path and uses Metal acceleration on macOS. It expects local model files and supports checkpoints plus supporting assets such as LoRAs, VAEs, CLIP, and ControlNet models.

Hugging Face Diffusers

The Diffusers backend runs through Planar’s Python bridge and can work with Hugging Face model IDs directly. It is useful for pipelines and model families that are not available in the native backend.

Model manager

Use Settings → Models to scan configured model roots and inspect the local library. Planar can also resolve downloads through CivitAI and Hugging Face. Private or gated Hugging Face downloads require an API key; CivitAI downloads require a CivitAI API key.

Edit models

Instruction-based Refine models are selected independently from the generation checkpoint. The selected model is persisted in settings. Check its license before using it in a commercial workflow; some edit models are non-commercial.

Samplers

The native backend exposes Euler A, Euler, Heun, DPM2, DPM++ 2S a, DPM++ 2M, DPM++ 2M v2, and LCM.