AI components 3

These are the files Immuch360 needs to make a 360° panorama from a photo on a Windows or Linux computer, an experimental feature. Everything beyond the edges of the photo is generated by AI, and the app says so. It downloads these files only when you ask for the feature: about 5.3 GB from this release, 5.4 GB with the MoGe file it shares with the 3D scenes.

FLUX.2 klein base 4B (Q4_0 GGUF) with its VAE, Qwen3 4B as its text encoder and the 360 ERP Outpaint LoRA of nomadoor are under Apache-2.0. The LoRA was trained on CC0 and public domain images, with a part under CC BY: MODEL-LICENSES.md gives the details. MI-GAN (MIT) closes the seam and Real-ESRGAN x4plus (BSD-3-Clause) enlarges the result. stable-diffusion.cpp (MIT) runs them on the graphics card through Vulkan: my CI builds it, and on Windows sd-cli.exe and its libraries are signed. The two GGUF files are over 2 GiB, so each comes in two parts that the app joins and checks.

The AI components of the 360° panoramas of Immuch360: FLUX.2 klein base 4B, Qwen3 4B, the VAE of klein and the 360 ERP Outpaint LoRA (Apache-2.0), MI-GAN (MIT), Real-ESRGAN x4plus (BSD-3-Clause) and stable-diffusion.cpp on Vulkan (MIT), built and signed by our CI. The files of 2 GiB or more are in parts, which the app puts end to end. This is not a release of the app: the app downloads these files only when you ask it to, checks each one by its SHA-256, and works offline afterwards. Licences and sources: MODEL-LICENSES.md. Sums: SHA256SUMS.