Disable Dynamic Caching when using SDXL inpaint model

* produces garbage output, not sure why

Author: Acly

ai_diffusion/workflow.py

@@ -1537,6 +1537,9 @@ def prepare(
         if minfo.arch.is_qwen_like and minfo.quantization is Quantization.svdq:
             i.batch_count = 1  # Nunchaku Qwen is broken with batch size > 1 #2114
 
+    if arch is Arch.sdxl and i.inpaint and i.inpaint.use_inpaint_model:
+        i.models.dynamic_caching = False  # inpaint model incompatible with dynamic caching
+
     i.batch_count = 1 if is_live else i.batch_count
     i.nsfw_filter = settings.nsfw_filter
     return i