Fix/adjust diffusion: export unload + sd3.5 alias for PR #5754

- routes/export.py load_checkpoint now unloads the diffusion
  pipeline alongside the existing inference + training unloads, so
  an export load after Images does not OOM the export subprocess.
- Remove the 'sd3.5' alias from the stable-diffusion-3 family.
  SD3.5 needs its own family + base_repo (and its own smoke test);
  pairing it with the SD3 Medium base produced a misleading load.

studio/backend/core/inference/diffusion.py

@@ -113,7 +113,13 @@ _FAMILIES: tuple[DiffusionFamily, ...] = (
         pipeline_class = "StableDiffusion3Pipeline",
         transformer_class = "SD3Transformer2DModel",
         base_repo = "stabilityai/stable-diffusion-3-medium-diffusers",
-        aliases = ("sd3-medium", "stable-diffusion-3-medium", "sd3.5"),
+        # Intentionally NOT including "sd3.5" / "stable-diffusion-3.5"
+        # here: the SD3.5 family uses a different transformer config and
+        # base repo than SD3 Medium, and silently pairing SD3.5 GGUFs
+        # with the Medium base produces a misleading load. Add a
+        # dedicated SD3.5 family with its own base_repo when we ship
+        # smoke coverage for it.
+        aliases = ("sd3-medium", "stable-diffusion-3-medium"),
     ),
     # SDXL: full diffusers path only (no GGUF). SDXL uses a UNet (not a
     # transformer) and wiring UNet2DConditionModel.from_single_file +

studio/backend/routes/export.py

@@ -81,6 +81,20 @@ async def load_checkpoint(
         except Exception as e:
             logger.warning("Could not unload inference model: %s", e)
 
+        # Also unload any active diffusion pipeline (Images page); it
+        # competes for the same GPU and would survive the inference
+        # shutdown above. Best effort; silently skip if the module is
+        # absent.
+        try:
+            from core.inference.diffusion import get_diffusion_backend
+
+            diff = get_diffusion_backend()
+            if diff.is_loaded:
+                logger.info("Unloading diffusion model to free GPU memory for export")
+                diff.unload_model()
+        except Exception as e:
+            logger.debug("diffusion unload skipped for export: %s", e)
+
         try:
             from core.training import get_training_backend