Offload H3 before latent upscale

dumas_h3_longvideos.py

@@ -6662,6 +6662,9 @@ class H3LongVideos:
                 parts = _nested_tensor_parts(out_samples)
                 if not getattr(out_samples, "is_nested", False) or len(parts) < 2:
                     raise RuntimeError("latent upscale expects a nested AV latent")
+                if latent_upscale_mode == "model" and str(latent_upscale_param.get("device", "cuda")) == "cuda" and hasattr(model, "clone_base_uuid"):
+                    mm.unload_model_and_clones(model, unload_additional_models=False)
+                    mm.soft_empty_cache()
                 upscaled_video, up_h, up_w = _upscale_latent_video(parts[0], latent_upscale_param)
                 full_audio = parts[1]
                 # Drop the first-pass sampling state before we start the refinement
@@ -6674,9 +6677,6 @@ class H3LongVideos:
                 target_h = int(up_h) * 16
                 if target_w <= 0 or target_h <= 0:
                     raise RuntimeError("latent upscale target size must be positive")
-                if latent_upscale_mode == "model" and str(latent_upscale_param.get("device", "cuda")) == "cuda" and hasattr(model, "clone_base_uuid"):
-                    mm.unload_model_and_clones(model, unload_additional_models=False)
-                    mm.soft_empty_cache()
                 upscale_cond, upscale_latent = _build_shot_conditioning(
                     clip, vae, prompt, target_w, target_h, ln, fps, handoff,
                     ref_images=refs, ref_image_size=ref_image_size,