The Wizard's Brush

Create an image or video, edit it, compare variants, and keep a library you can trace back to its inputs. Generation is one part of the workspace.

From prompt to library

Generate images and video, work from source images, inpaint, outpaint, apply instruction edits, or compose assets in a layered canvas. Finishing operations such as upscaling, face restoration, resizing and background removal are explicit steps. A searchable library keeps captions, tags, collections, ratings and lineage with the result.

Sets are durable work

Variant Sets expand named axes over shared source material. Each combination is validated and tracked as an ordinary job, so failed items can be retried without rebuilding the set. Recipes preserve settings for reuse. Exports retain the chosen metadata policy and provenance.

Controls come from capabilities

Each generator declares its supported inputs, limits and defaults. The interface and JSON job schema read that contract, so a model that cannot perform an operation does not acquire a misleading control. Local and remote lanes keep separate queues and capability checks.

Use it from an agent

The HTTP API covers importing assets, discovering generators, queueing jobs, waiting for results and exporting Variant Sets. The MCP server uses that same API. Readiness reflects the configured model and live remote worker; selecting a remote lane sends its prompts and media to that worker.

Install and update

The standalone release includes a built interface and installs pinned tools inside its own directory. Linux x86-64 and Windows through WSL2 are supported, with 16 GB VRAM as the tested local baseline. Model weights and CUDA have separate terms. Use the latest release for dependency security fixes, and follow its backup and migration instructions before updating.