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.

Support VAEs in diffusers format.

comfyanonymousa53288884680
diff --git a/comfy/sd.py b/comfy/sd.py
index c6be900ad..4df149fe1 100644
--- a/comfy/sd.py
+++ b/comfy/sd.py
@@ -14,6 +14,7 @@ from .t2i_adapter import adapter
from . import utils
from . import clip_vision
from . import gligen
+from . import diffusers_convert
def load_model_weights(model, sd, verbose=False, load_state_dict_to=[]):
m, u = model.load_state_dict(sd, strict=False)
@@ -504,10 +505,16 @@ class VAE:
if config is None:
#default SD1.x/SD2.x VAE parameters
ddconfig = {'double_z': True, 'z_channels': 4, 'resolution': 256, 'in_channels': 3, 'out_ch': 3, 'ch': 128, 'ch_mult': [1, 2, 4, 4], 'num_res_blocks': 2, 'attn_resolutions': [], 'dropout': 0.0}
- self.first_stage_model = AutoencoderKL(ddconfig, {'target': 'torch.nn.Identity'}, 4, monitor="val/rec_loss", ckpt_path=ckpt_path)
+ self.first_stage_model = AutoencoderKL(ddconfig, {'target': 'torch.nn.Identity'}, 4, monitor="val/rec_loss")
else:
- self.first_stage_model = AutoencoderKL(**(config['params']), ckpt_path=ckpt_path)
+ self.first_stage_model = AutoencoderKL(**(config['params']))
self.first_stage_model = self.first_stage_model.eval()
+ if ckpt_path is not None:
+ sd = utils.load_torch_file(ckpt_path)
+ if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format
+ sd = diffusers_convert.convert_vae_state_dict(sd)
+ self.first_stage_model.load_state_dict(sd, strict=False)
+
self.scale_factor = scale_factor
if device is None:
device = model_management.get_torch_device()