Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Clean up DiffusersLoader node.

comfyanonymous7e941f9f247f
diff --git a/comfy/diffusers_load.py b/comfy/diffusers_load.py
index 11d94c340..a52e0102b 100644
--- a/comfy/diffusers_load.py
+++ b/comfy/diffusers_load.py
@@ -1,87 +1,36 @@
import json
import os
-import yaml
-import folder_paths
-from comfy.sd import load_checkpoint
-import os.path as osp
-import re
-import torch
-from safetensors.torch import load_file, save_file
-from . import diffusers_convert
+import comfy.sd
+def first_file(path, filenames):
+ for f in filenames:
+ p = os.path.join(path, f)
+ if os.path.exists(p):
+ return p
+ return None
-def load_diffusers(model_path, fp16=True, output_vae=True, output_clip=True, embedding_directory=None):
- diffusers_unet_conf = json.load(open(osp.join(model_path, "unet/config.json")))
- diffusers_scheduler_conf = json.load(open(osp.join(model_path, "scheduler/scheduler_config.json")))
+def load_diffusers(model_path, output_vae=True, output_clip=True, embedding_directory=None):
+ diffusion_model_names = ["diffusion_pytorch_model.fp16.safetensors", "diffusion_pytorch_model.safetensors", "diffusion_pytorch_model.fp16.bin", "diffusion_pytorch_model.bin"]
+ unet_path = first_file(os.path.join(model_path, "unet"), diffusion_model_names)
+ vae_path = first_file(os.path.join(model_path, "vae"), diffusion_model_names)
- # magic
- v2 = diffusers_unet_conf["sample_size"] == 96
- if 'prediction_type' in diffusers_scheduler_conf:
- v_pred = diffusers_scheduler_conf['prediction_type'] == 'v_prediction'
+ text_encoder_model_names = ["model.fp16.safetensors", "model.safetensors", "pytorch_model.fp16.bin", "pytorch_model.bin"]
+ text_encoder1_path = first_file(os.path.join(model_path, "text_encoder"), text_encoder_model_names)
+ text_encoder2_path = first_file(os.path.join(model_path, "text_encoder_2"), text_encoder_model_names)
- if v2:
- if v_pred:
- config_path = folder_paths.get_full_path("configs", 'v2-inference-v.yaml')
- else:
- config_path = folder_paths.get_full_path("configs", 'v2-inference.yaml')
- else:
- config_path = folder_paths.get_full_path("configs", 'v1-inference.yaml')
+ text_encoder_paths = [text_encoder1_path]
+ if text_encoder2_path is not None:
+ text_encoder_paths.append(text_encoder2_path)
- with open(config_path, 'r') as stream:
- config = yaml.safe_load(stream)
+ unet = comfy.sd.load_unet(unet_path)
- model_config_params = config['model']['params']
- clip_config = model_config_params['cond_stage_config']
- scale_factor = model_config_params['scale_factor']
- vae_config = model_config_params['first_stage_config']
- vae_config['scale_factor'] = scale_factor
- model_config_params["unet_config"]["params"]["use_fp16"] = fp16
+ clip = None
+ if output_clip:
+ clip = comfy.sd.load_clip(text_encoder_paths, embedding_directory=embedding_directory)
- unet_path = osp.join(model_path, "unet", "diffusion_pytorch_model.safetensors")
- vae_path = osp.join(model_path, "vae", "diffusion_pytorch_model.safetensors")
- text_enc_path = osp.join(model_path, "text_encoder", "model.safetensors")
+ vae = None
+ if output_vae:
+ vae = comfy.sd.VAE(ckpt_path=vae_path)
- # Load models from safetensors if it exists, if it doesn't pytorch
- if osp.exists(unet_path):
- unet_state_dict = load_file(unet_path, device="cpu")
- else:
- unet_path = osp.join(model_path, "unet", "diffusion_pytorch_model.bin")
- unet_state_dict = torch.load(unet_path, map_location="cpu")
-
- if osp.exists(vae_path):
- vae_state_dict = load_file(vae_path, device="cpu")
- else:
- vae_path = osp.join(model_path, "vae", "diffusion_pytorch_model.bin")
- vae_state_dict = torch.load(vae_path, map_location="cpu")
-
- if osp.exists(text_enc_path):
- text_enc_dict = load_file(text_enc_path, device="cpu")
- else:
- text_enc_path = osp.join(model_path, "text_encoder", "pytorch_model.bin")
- text_enc_dict = torch.load(text_enc_path, map_location="cpu")
-
- # Convert the UNet model
- unet_state_dict = diffusers_convert.convert_unet_state_dict(unet_state_dict)
- unet_state_dict = {"model.diffusion_model." + k: v for k, v in unet_state_dict.items()}
-
- # Convert the VAE model
- vae_state_dict = diffusers_convert.convert_vae_state_dict(vae_state_dict)
- vae_state_dict = {"first_stage_model." + k: v for k, v in vae_state_dict.items()}
-
- # Easiest way to identify v2.0 model seems to be that the text encoder (OpenCLIP) is deeper
- is_v20_model = "text_model.encoder.layers.22.layer_norm2.bias" in text_enc_dict
-
- if is_v20_model:
- # Need to add the tag 'transformer' in advance so we can knock it out from the final layer-norm
- text_enc_dict = {"transformer." + k: v for k, v in text_enc_dict.items()}
- text_enc_dict = diffusers_convert.convert_text_enc_state_dict_v20(text_enc_dict)
- text_enc_dict = {"cond_stage_model.model." + k: v for k, v in text_enc_dict.items()}
- else:
- text_enc_dict = diffusers_convert.convert_text_enc_state_dict(text_enc_dict)
- text_enc_dict = {"cond_stage_model.transformer." + k: v for k, v in text_enc_dict.items()}
-
- # Put together new checkpoint
- sd = {**unet_state_dict, **vae_state_dict, **text_enc_dict}
-
- return load_checkpoint(embedding_directory=embedding_directory, state_dict=sd, config=config)
+ return (unet, clip, vae)
diff --git a/nodes.py b/nodes.py
index 3e4d5240b..5e755f149 100644
--- a/nodes.py
+++ b/nodes.py
@@ -475,7 +475,7 @@ class DiffusersLoader:
model_path = path
break
- return comfy.diffusers_load.load_diffusers(model_path, fp16=comfy.model_management.should_use_fp16(), output_vae=output_vae, output_clip=output_clip, embedding_directory=folder_paths.get_folder_paths("embeddings"))
+ return comfy.diffusers_load.load_diffusers(model_path, output_vae=output_vae, output_clip=output_clip, embedding_directory=folder_paths.get_folder_paths("embeddings"))
class unCLIPCheckpointLoader: