x

custum_3d_diffusion/custum_pipeline/unifield_pipeline_img2mvimg.py

@@ -214,8 +214,9 @@ class StableDiffusionImage2MVCustomPipeline(
         # 3. Encode input image
         emb_image = image
         
-        image_embeddings = self._encode_image(emb_image, device, num_images_per_prompt, do_classifier_free_guidance).to(device="cuda", dtype=torch.bfloat16)
+        image_embeddings = self._encode_image(emb_image, device, num_images_per_prompt, do_classifier_free_guidance).to(device=self.unet.device, dtype=elf.unet.dtype)
         print("DEBUG: image_embeddings", image_embeddings.dtype, image_embeddings.device)
+        print("DEBUG: version v111")
         cond_latents = self.encode_latents(image, image_embeddings.device, image_embeddings.dtype, height_cond, width_cond)
         cond_latents = torch.cat([torch.zeros_like(cond_latents), cond_latents]) if do_classifier_free_guidance else cond_latents
         image_pixels = self.feature_extractor(images=emb_image, return_tensors="pt").pixel_values