Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

1.7 KB
Raw Blame

Qwen-Image 2.1

Choose Local → Qwen-Image 2.1 in Image Gen. The model uses Qwen/Qwen-Image-2.1 and Diffusers' QwenImage21Pipeline, with 40 steps and guidance disabled by default. Existing Qwen models remain available. The registry adds the new choice to existing installs without changing their selected model or customized settings.

Install or repair the shared FLUX.2 image runtime through the app, or run:

INSTALL_FLUX2=1 FLUX2_FORCE_REINSTALL=1 bash scripts/setup-image-video.sh

Both installers use current Diffusers from git and Transformers >=5.17. An older runtime missing QwenImage21Pipeline reports the repair command before loading weights. Setup does not download model weights; those load when a render is requested.

Text-to-image and editing through Init Image and reference-image uploads are supported, with up to ten image inputs in total. An Init Image uses one slot, leaving nine reference-image slots. Editing preserves the reference's aspect ratio and alpha channel; this unified pipeline does not use denoising strength or per-reference strength controls. PNG output retains generated transparency. For transparent output, explicitly describe an RGBA image with an alpha channel and a transparent background in the prompt. PortOS does not expose Qwen's separate-mask interface.

The model ships under the Qwen Research License, unlike the earlier Apache-licensed Qwen image models. See the model and license and upstream usage examples. The integration's automated checks use fixture pipelines; they do not establish render quality or hardware memory requirements for the full model.