Qwen Image Lora training fix from #11090 (#11094)

Author: comfyanonymous

comfy_extras/nodes_train.py

@@ -623,7 +623,7 @@ def loss_callback(loss):
                 noise = comfy_extras.nodes_custom_sampler.Noise_RandomNoise(seed)
                 if multi_res:
                     # use first latent as dummy latent if multi_res
-                    latents = latents[0].repeat(num_images, 1, 1, 1)
+                    latents = latents[0].repeat((num_images,) + ((1,) * (latents[0].ndim - 1)))
                 guider.sample(
                     noise.generate_noise({"samples": latents}),
                     latents,