1/ Just like Skills & MCP have enriched and supercharged the world of LLMs, now we have Diffusion Tempaltes for vision models, which allow Pluggable, Extensible and Stackable components to be built atop Diffusion models. Come exploring the Diffusion ecosystem with us modelscope.github.io/diffusi… The diffusion ecosystem have been split among different techniques: ControlNet, LoRA, IP-Adapter, edit models, upscalers…… but stacking and combining them can be messy, since all were built against different assumptions. So we built the long-overdued fabrics layer: Diffusion Templates — a plug-and-play framework where controls actually compose. To kick it off, we trained 11 Templates on FLUX.2-klein-base-4B and open-sourced everything 🧩 Structural Control · Brightness · Color · Sharpness · Image Editing · Aesthetic Alignment · Local Inpainting · Age · Super-Resolution · Content Reference · + one cursed Panda Meme template 🐼