init

app.py

@@ -22,7 +22,9 @@ from packaging import version
 
 from accelerate.utils import set_seed
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
-
+from models.unet_spatio_temporal_condition_controlnet import UNetSpatioTemporalConditionControlNetModel
+from pipeline.pipeline import FlowControlNetPipeline
+from models.svdxt_featureflow_forward_controlnet_s2d_fixcmp_norefine import FlowControlNet, CMP_demo
 from diffusers import AutoencoderKLTemporalDecoder, EulerDiscreteScheduler
 from diffusers.utils import check_min_version
 from diffusers.utils.import_utils import is_xformers_available
@@ -153,10 +155,6 @@ class Drag:
     @spaces.GPU(duration=100)
     def init_models(self, pretrained_model_name_or_path, resume_from_checkpoint, weight_dtype, device='cuda', enable_xformers_memory_efficient_attention=False, allow_tf32=False):
 
-        from models.unet_spatio_temporal_condition_controlnet import UNetSpatioTemporalConditionControlNetModel
-        from pipeline.pipeline import FlowControlNetPipeline
-        from models.svdxt_featureflow_forward_controlnet_s2d_fixcmp_norefine import FlowControlNet, CMP_demo
-
         print('start loading models...')
         # Load scheduler, tokenizer and models.
         image_encoder = CLIPVisionModelWithProjection.from_pretrained(