[BlackMixture]

Mix Studio now offers Qwen Image 2.1 in Create and Edit while retaining Krea 2 as the default. The integration uses native ComfyUI conditioning, the dedicated VAE, separate LoRA stacks, model-aware reuse/metadata, and animated keyboard-accessible sampling controls. Unsupported regional, mask, Expand, and camera workflows remain on their existing models.

Setup selects official INT8/W4A8 assets for lower-memory NVIDIA systems, INT8/INT8 for 16 GB+ NVIDIA systems, or BF16 for other backends. Downloads are ungated and reuse installed files. The UI and documentation show the model's noncommercial research license. Old 2511 Lightning adapters are not applied to 2.1.

Also includes the reviewed LM Studio Prompt AI support, long-running Smart request retention/deadlines, and the LTX 2.5 native-support status correction.

Validation:

• Node syntax checks and full suite: 1,325 passed, 2 skipped after rebasing on current main.
• BMBeast RTX 4090, ComfyUI v0.37.0: real token-free downloads; balanced and compact 1024-square creation; 25/40-step editing; two-reference editing with a 768×512 canvas and batch of two.
• Browser checks: mobile overflow, keyboard/pointer sampling, persistence, Krea settings restoration, Edit selection; desktop dropdown visually reviewed on BMBeast.
• A single color-edit example changed background texture at both 25 and 40 steps. This is not a claim of pixel-perfect preservation or a comprehensive quantization benchmark. Actual 8 GB hardware and BF16 inference were not tested.

The installed GitHub update path will be checked on BMBeast after merge. No stable release tag is created by this PR.