Stable Diffusion WebUI Codex v0.3.0-beta adds FLUX.2, LTX2 video, WAN 2.2

The latest release of the popular Stable Diffusion WebUI extension, Codex, packs over 40 new features and improvements. Chief among them is full support for FLUX.2 Klein, including dedicated tabs, model metadata handling, prompt-token counting, and even img2img continuation. For video creators, Codex now natively handles LTX2 generation with both text-to-video and image-to-video UIs, plus explicit two-stage execution profiles and GGUF asset validation. WAN 2.2 gets its own dedicated model lanes for the 14B and 5B variants, along with exact capability lookup, shared video history, and a zoom overlay.

On the image front, the update adds SDXL Fooocus Inpaint and BrushNet inpaint, plus SUPIR integrated directly into the normal img2img/inpaint workflow. IP-Adapter support with reference-image conditioning is now included, and a new safetensors merge tool gives model fusion capabilities. Many UI improvements target inpaint workflows: per-step blend window control, live blur/padding previews, invert-mask controls, and better crop/visual feedback. Sampler and scheduler selection now reflects explicit choices, and model selection is more reliable. For Windows users, an optional build with PyTorch 2.9.1 and Flash Attention 2 is available for SM80–SM90 GPUs, though the default build omits FA2 due to platform constraints.