switch to tf backend

Switch to openvinvo

Change Defaults

Add CFG rescale control (default 0.5)

Remove dynamic thresholding

Fix dynamic thresholding to faithful Imagen (drop upper cap)

Add sampler controls: steps, guidance, spacing, dynamic thresholding

Load HF checkpoints via correct Keras-3 weights loader

Fix endless sprite re-fetching during generation

Swap to from-scratch DC-AE + TinyCLIP + UNet stack; frontend updates

Pin ML deps to verified versions; use tensorflow-cpu

Deploy Senior LDM FastAPI Docker Space