Update custom_pipeline.py

custom_pipeline.py

@@ -152,10 +152,10 @@ class FLUXPipelineWithIntermediateOutputs(FluxPipeline):
 
              # Yield intermediate result
             latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
-            yield self._decode_latents_to_image(latents, height, width, output_type)
             torch.cuda.empty_cache()
 
         # Final image
+        yield self._decode_latents_to_image(latents, height, width, output_type)
         self.maybe_free_model_hooks()
         torch.cuda.empty_cache()