mirrors/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-09-10 11:35:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Initial support for qwen image model. (#9179)

Browse Source
This commit is contained in:
comfyanonymous
2025-08-04 19:53:25 -07:00
committed by GitHub
parent 03895dea7c
commit c012400240
8 changed files with 557 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
comfy/sd.py
View File

@@ -47,6 +47,7 @@ import comfy.text_encoders.wan
import comfy.text_encoders.hidream
import comfy.text_encoders.ace
import comfy.text_encoders.omnigen2
import comfy.text_encoders.qwen_image
import comfy.model_patcher
import comfy.lora
@@ -771,6 +772,7 @@ class CLIPType(Enum):
CHROMA = 15
ACE = 16
OMNIGEN2 = 17
QWEN_IMAGE = 18
def load_clip(ckpt_paths, embedding_directory=None, clip_type=CLIPType.STABLE_DIFFUSION, model_options={}):
@@ -791,6 +793,7 @@ class TEModel(Enum):
T5_XXL_OLD = 8
GEMMA_2_2B = 9
QWEN25_3B = 10
QWEN25_7B = 11
def detect_te_model(sd):
if "text_model.encoder.layers.30.mlp.fc1.weight" in sd:
@@ -812,7 +815,11 @@ def detect_te_model(sd):
if 'model.layers.0.post_feedforward_layernorm.weight' in sd:
return TEModel.GEMMA_2_2B
if 'model.layers.0.self_attn.k_proj.bias' in sd:
return TEModel.QWEN25_3B
weight = sd['model.layers.0.self_attn.k_proj.bias']
if weight.shape[0] == 256:
return TEModel.QWEN25_3B
if weight.shape[0] == 512:
return TEModel.QWEN25_7B
if "model.layers.0.post_attention_layernorm.weight" in sd:
return TEModel.LLAMA3_8
return None
@@ -917,6 +924,9 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
elif te_model == TEModel.QWEN25_3B:
clip_target.clip = comfy.text_encoders.omnigen2.te(**llama_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.omnigen2.Omnigen2Tokenizer
elif te_model == TEModel.QWEN25_7B:
clip_target.clip = comfy.text_encoders.qwen_image.te(**llama_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.qwen_image.QwenImageTokenizer
else:
# clip_l
if clip_type == CLIPType.SD3: