feat: add Qwen-Image-2.1 prompt enhancer
Browse files- README.md +2 -0
- chain_injectors/qwen_image_2_1_prompt_enhancer_injector.py +259 -0
- core/model_manager.py +13 -6
- core/pipelines/pipeline_input_processor.py +10 -0
- core/pipelines/sd_image_pipeline.py +14 -0
- core/pipelines/workflow_recipes/_partials/conditioning/qwen-image-2.1.yaml +4 -0
- mcp_tools/common.py +23 -0
- ui/events/change_handlers.py +4 -2
- ui/events/main.py +9 -3
- ui/events/run_handlers.py +7 -0
- ui/imagegen_ui.py +2 -1
- ui/shared/ui_components.py +33 -0
- yaml/chain_features.yaml +22 -0
- yaml/file_list.yaml +9 -4
- yaml/model_architecture_features.yaml +1 -0
- yaml/model_list.yaml +2 -2
- yaml/task_features.yaml +5 -0
README.md
CHANGED
|
@@ -172,6 +172,8 @@ models:
|
|
| 172 |
- Qwen/Qwen-Image
|
| 173 |
- Qwen/Qwen-Image-2512
|
| 174 |
- Qwen/Qwen-Image-2.1
|
|
|
|
|
|
|
| 175 |
- Qwen/Qwen-Image-Edit
|
| 176 |
- Qwen/Qwen-Image-Edit-2509
|
| 177 |
- Qwen/Qwen-Image-Edit-2511
|
|
|
|
| 172 |
- Qwen/Qwen-Image
|
| 173 |
- Qwen/Qwen-Image-2512
|
| 174 |
- Qwen/Qwen-Image-2.1
|
| 175 |
+
- Qwen/Qwen-Image-2.1-PE-I2I
|
| 176 |
+
- Qwen/Qwen-Image-2.1-PE-T2I
|
| 177 |
- Qwen/Qwen-Image-Edit
|
| 178 |
- Qwen/Qwen-Image-Edit-2509
|
| 179 |
- Qwen/Qwen-Image-Edit-2511
|
chain_injectors/qwen_image_2_1_prompt_enhancer_injector.py
ADDED
|
@@ -0,0 +1,259 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Qwen-Image-2.1 Prompt Enhancer Chain Injector.
|
| 2 |
+
|
| 3 |
+
This injector automatically integrates the Qwen-Image-2.1 prompt enhancement model into
|
| 4 |
+
the generation workflow:
|
| 5 |
+
- Detects whether the task is pure text-to-image (T2I) or involves images (I2I, inpainting,
|
| 6 |
+
outpainting, hires.fix, or multi-reference images).
|
| 7 |
+
- Automatically selects the appropriate prompt enhancement checkpoint:
|
| 8 |
+
- T2I: qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors
|
| 9 |
+
- I2I / Reference: qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors
|
| 10 |
+
- Handles single or multi-reference inputs: multiple reference images are stitched into
|
| 11 |
+
a structured grid layout via ComfyUI native ImageStitch nodes and scaled via ImageScaleToTotalPixels.
|
| 12 |
+
- Supports reasoning mode (thinking=True) with automatic reasoning tag stripping via RegexReplace
|
| 13 |
+
(^.*?</think>\\s*) to prevent thought chain text from contaminating downstream image conditioning.
|
| 14 |
+
"""
|
| 15 |
+
|
| 16 |
+
try:
|
| 17 |
+
from utils.app_utils import ensure_file_downloaded
|
| 18 |
+
except ImportError:
|
| 19 |
+
ensure_file_downloaded = None
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
def create_node(assembler, class_type, title):
|
| 23 |
+
"""Create a workflow node dictionary with fallback template generation."""
|
| 24 |
+
try:
|
| 25 |
+
node = assembler._get_node_template(class_type)
|
| 26 |
+
except Exception:
|
| 27 |
+
node = {
|
| 28 |
+
"inputs": {},
|
| 29 |
+
"class_type": class_type,
|
| 30 |
+
"_meta": {"title": title}
|
| 31 |
+
}
|
| 32 |
+
if "_meta" not in node:
|
| 33 |
+
node["_meta"] = {}
|
| 34 |
+
node['_meta']['title'] = title
|
| 35 |
+
return node
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
def inject(assembler, chain_definition, chain_items):
|
| 39 |
+
"""Inject Qwen-Image-2.1 prompt enhancer nodes into the target generation workflow."""
|
| 40 |
+
if not chain_items:
|
| 41 |
+
return
|
| 42 |
+
|
| 43 |
+
is_enabled = False
|
| 44 |
+
thinking_enabled = False
|
| 45 |
+
max_length = 512
|
| 46 |
+
for item in chain_items:
|
| 47 |
+
if isinstance(item, dict):
|
| 48 |
+
if item.get("enable", False) or item.get("enabled", False) or item.get("value", False):
|
| 49 |
+
is_enabled = True
|
| 50 |
+
if item.get("thinking", False) in (True, "true", "on", "yes", "1", 1):
|
| 51 |
+
thinking_enabled = True
|
| 52 |
+
if "max_length" in item and item["max_length"]:
|
| 53 |
+
try:
|
| 54 |
+
max_length = int(item["max_length"])
|
| 55 |
+
except (ValueError, TypeError):
|
| 56 |
+
pass
|
| 57 |
+
elif item is True or str(item).lower() in ("on", "true", "yes", "1"):
|
| 58 |
+
is_enabled = True
|
| 59 |
+
break
|
| 60 |
+
|
| 61 |
+
if not is_enabled:
|
| 62 |
+
return
|
| 63 |
+
|
| 64 |
+
target_node_name = chain_definition.get('target_node', 'prompt')
|
| 65 |
+
target_node_id = assembler.node_map.get(target_node_name)
|
| 66 |
+
|
| 67 |
+
if not target_node_id or target_node_id not in assembler.workflow:
|
| 68 |
+
for node_id, node in assembler.workflow.items():
|
| 69 |
+
if isinstance(node, dict) and node.get('class_type') == 'TextEncodeQwenImage21':
|
| 70 |
+
target_node_id = node_id
|
| 71 |
+
break
|
| 72 |
+
|
| 73 |
+
if not target_node_id or target_node_id not in assembler.workflow:
|
| 74 |
+
print(f"Warning: Target node '{target_node_name}' (TextEncodeQwenImage21) not found for Qwen-Image-2.1 Prompt Enhancer. Skipping.")
|
| 75 |
+
return
|
| 76 |
+
|
| 77 |
+
target_node = assembler.workflow[target_node_id]
|
| 78 |
+
original_prompt = target_node.get('inputs', {}).get('prompt', '')
|
| 79 |
+
|
| 80 |
+
# 1. Check if any LoadImage nodes exist in the workflow
|
| 81 |
+
load_image_nodes = [
|
| 82 |
+
node_id for node_id, node in assembler.workflow.items()
|
| 83 |
+
if isinstance(node, dict) and node.get('class_type') == 'LoadImage'
|
| 84 |
+
]
|
| 85 |
+
has_load_image = len(load_image_nodes) > 0
|
| 86 |
+
|
| 87 |
+
target_image_conn = None
|
| 88 |
+
|
| 89 |
+
if not has_load_image:
|
| 90 |
+
# Pure text-to-image (txt2img) task without reference images
|
| 91 |
+
clip_model_name = "qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors"
|
| 92 |
+
else:
|
| 93 |
+
# Task with reference images or img2img / inpaint / outpaint / hires.fix
|
| 94 |
+
clip_model_name = "qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors"
|
| 95 |
+
|
| 96 |
+
# Collect image connections
|
| 97 |
+
# Prioritize extracting all images.image_* slots from target_node['inputs'] (injected by reference image injectors)
|
| 98 |
+
indexed_slots = []
|
| 99 |
+
target_inputs = target_node.get('inputs', {})
|
| 100 |
+
for slot_key, slot_val in target_inputs.items():
|
| 101 |
+
if slot_key.startswith('images.image_') and isinstance(slot_val, (list, tuple)):
|
| 102 |
+
try:
|
| 103 |
+
idx = int(slot_key.split('_')[-1])
|
| 104 |
+
except Exception:
|
| 105 |
+
idx = 999
|
| 106 |
+
indexed_slots.append((idx, slot_val))
|
| 107 |
+
|
| 108 |
+
indexed_slots.sort(key=lambda x: x[0])
|
| 109 |
+
ref_image_conns = [s[1] for s in indexed_slots]
|
| 110 |
+
|
| 111 |
+
# If no reference image slots exist (e.g. pure img2img / inpaint / outpaint), retrieve from LoadImage nodes
|
| 112 |
+
if not ref_image_conns:
|
| 113 |
+
for lid in load_image_nodes:
|
| 114 |
+
scaled_conn = None
|
| 115 |
+
for nid, node in assembler.workflow.items():
|
| 116 |
+
if isinstance(node, dict) and node.get('class_type') == 'ImageScaleToTotalPixels':
|
| 117 |
+
img_input = node.get('inputs', {}).get('image')
|
| 118 |
+
if isinstance(img_input, (list, tuple)) and str(img_input[0]) == str(lid):
|
| 119 |
+
scaled_conn = [nid, 0]
|
| 120 |
+
break
|
| 121 |
+
if scaled_conn:
|
| 122 |
+
ref_image_conns.append(scaled_conn)
|
| 123 |
+
else:
|
| 124 |
+
scale_id = assembler._get_unique_id()
|
| 125 |
+
scale_node = create_node(assembler, "ImageScaleToTotalPixels", "Scale Input Image")
|
| 126 |
+
scale_node['inputs'] = {
|
| 127 |
+
"upscale_method": "nearest-exact",
|
| 128 |
+
"megapixels": 1,
|
| 129 |
+
"resolution_steps": 1,
|
| 130 |
+
"image": [lid, 0]
|
| 131 |
+
}
|
| 132 |
+
assembler.workflow[scale_id] = scale_node
|
| 133 |
+
ref_image_conns.append([scale_id, 0])
|
| 134 |
+
|
| 135 |
+
# Organize multiple reference images into grid stitching
|
| 136 |
+
if len(ref_image_conns) == 1:
|
| 137 |
+
target_image_conn = ref_image_conns[0]
|
| 138 |
+
elif len(ref_image_conns) >= 2:
|
| 139 |
+
num_cols = 2 if len(ref_image_conns) <= 4 else 3
|
| 140 |
+
|
| 141 |
+
rows = []
|
| 142 |
+
for i in range(0, len(ref_image_conns), num_cols):
|
| 143 |
+
rows.append(ref_image_conns[i:i + num_cols])
|
| 144 |
+
|
| 145 |
+
# Horizontally stitch images in each row
|
| 146 |
+
stitched_rows = []
|
| 147 |
+
for r_idx, row_imgs in enumerate(rows):
|
| 148 |
+
if len(row_imgs) == 1:
|
| 149 |
+
stitched_rows.append(row_imgs[0])
|
| 150 |
+
else:
|
| 151 |
+
curr_stitch = row_imgs[0]
|
| 152 |
+
for c_idx in range(1, len(row_imgs)):
|
| 153 |
+
st_id = assembler._get_unique_id()
|
| 154 |
+
st_node = create_node(assembler, "ImageStitch", f"Stitch Images (Row {r_idx+1}-{c_idx})")
|
| 155 |
+
st_node['inputs'] = {
|
| 156 |
+
"direction": "right",
|
| 157 |
+
"match_image_size": True,
|
| 158 |
+
"spacing_width": 0,
|
| 159 |
+
"spacing_color": "white",
|
| 160 |
+
"image1": curr_stitch,
|
| 161 |
+
"image2": row_imgs[c_idx]
|
| 162 |
+
}
|
| 163 |
+
assembler.workflow[st_id] = st_node
|
| 164 |
+
curr_stitch = [st_id, 0]
|
| 165 |
+
stitched_rows.append(curr_stitch)
|
| 166 |
+
|
| 167 |
+
# Vertically stitch multiple rows
|
| 168 |
+
if len(stitched_rows) == 1:
|
| 169 |
+
final_stitched = stitched_rows[0]
|
| 170 |
+
else:
|
| 171 |
+
curr_vertical = stitched_rows[0]
|
| 172 |
+
for r_idx in range(1, len(stitched_rows)):
|
| 173 |
+
st_id = assembler._get_unique_id()
|
| 174 |
+
st_node = create_node(assembler, "ImageStitch", f"Stitch Images (Vertical {r_idx})")
|
| 175 |
+
st_node['inputs'] = {
|
| 176 |
+
"direction": "down",
|
| 177 |
+
"match_image_size": True,
|
| 178 |
+
"spacing_width": 0,
|
| 179 |
+
"spacing_color": "white",
|
| 180 |
+
"image1": curr_vertical,
|
| 181 |
+
"image2": stitched_rows[r_idx]
|
| 182 |
+
}
|
| 183 |
+
assembler.workflow[st_id] = st_node
|
| 184 |
+
curr_vertical = [st_id, 0]
|
| 185 |
+
final_stitched = curr_vertical
|
| 186 |
+
|
| 187 |
+
# Scale final stitched image to ~1024 level via ImageScaleToTotalPixels
|
| 188 |
+
final_scale_id = assembler._get_unique_id()
|
| 189 |
+
final_scale_node = create_node(assembler, "ImageScaleToTotalPixels", "Scale Stitched Image")
|
| 190 |
+
final_scale_node['inputs'] = {
|
| 191 |
+
"upscale_method": "nearest-exact",
|
| 192 |
+
"megapixels": 1,
|
| 193 |
+
"resolution_steps": 1,
|
| 194 |
+
"image": final_stitched
|
| 195 |
+
}
|
| 196 |
+
assembler.workflow[final_scale_id] = final_scale_node
|
| 197 |
+
target_image_conn = [final_scale_id, 0]
|
| 198 |
+
|
| 199 |
+
# 2. Ensure model is downloaded in CPU stage before workflow execution
|
| 200 |
+
if ensure_file_downloaded:
|
| 201 |
+
try:
|
| 202 |
+
ensure_file_downloaded(clip_model_name)
|
| 203 |
+
except Exception as e:
|
| 204 |
+
print(f"Warning: Failed to ensure '{clip_model_name}' downloaded: {e}")
|
| 205 |
+
|
| 206 |
+
# 2. Create CLIPLoader node
|
| 207 |
+
clip_node_id = assembler._get_unique_id()
|
| 208 |
+
clip_node = create_node(assembler, "CLIPLoader", "Load CLIP")
|
| 209 |
+
clip_node['inputs'] = {
|
| 210 |
+
"clip_name": clip_model_name,
|
| 211 |
+
"type": "qwen_image",
|
| 212 |
+
"device": "default"
|
| 213 |
+
}
|
| 214 |
+
assembler.workflow[clip_node_id] = clip_node
|
| 215 |
+
|
| 216 |
+
# 3. Create TextGenerate node
|
| 217 |
+
text_gen_node_id = assembler._get_unique_id()
|
| 218 |
+
text_gen_node = create_node(assembler, "TextGenerate", "Generate Text")
|
| 219 |
+
text_gen_inputs = {
|
| 220 |
+
"prompt": original_prompt,
|
| 221 |
+
"max_length": max_length,
|
| 222 |
+
"sampling_mode": "on",
|
| 223 |
+
"sampling_mode.temperature": 0.7,
|
| 224 |
+
"sampling_mode.top_k": 20,
|
| 225 |
+
"sampling_mode.top_p": 0.95,
|
| 226 |
+
"sampling_mode.min_p": 0.0,
|
| 227 |
+
"sampling_mode.repetition_penalty": 1.05,
|
| 228 |
+
"sampling_mode.seed": 0,
|
| 229 |
+
"sampling_mode.presence_penalty": 0.0 if has_load_image else 1.5,
|
| 230 |
+
"thinking": thinking_enabled,
|
| 231 |
+
"use_default_template": True,
|
| 232 |
+
"mtp": "auto",
|
| 233 |
+
"clip": [clip_node_id, 0]
|
| 234 |
+
}
|
| 235 |
+
if target_image_conn is not None:
|
| 236 |
+
text_gen_inputs["image"] = target_image_conn
|
| 237 |
+
|
| 238 |
+
text_gen_node['inputs'] = text_gen_inputs
|
| 239 |
+
assembler.workflow[text_gen_node_id] = text_gen_node
|
| 240 |
+
|
| 241 |
+
# 4. Insert Replace Text (Regex) node if thinking is enabled to strip reasoning tags
|
| 242 |
+
if thinking_enabled:
|
| 243 |
+
regex_node_id = assembler._get_unique_id()
|
| 244 |
+
regex_node = create_node(assembler, "RegexReplace", "Replace Text (Regex)")
|
| 245 |
+
regex_node['inputs'] = {
|
| 246 |
+
"string": [text_gen_node_id, 0],
|
| 247 |
+
"regex_pattern": "^.*?</think>\\s*",
|
| 248 |
+
"replace": "",
|
| 249 |
+
"case_insensitive": True,
|
| 250 |
+
"multiline": False,
|
| 251 |
+
"dotall": True,
|
| 252 |
+
"count": 0
|
| 253 |
+
}
|
| 254 |
+
assembler.workflow[regex_node_id] = regex_node
|
| 255 |
+
target_node['inputs']['prompt'] = [regex_node_id, 0]
|
| 256 |
+
print(f"[Injector] Qwen-Image-2.1 Prompt Enhancer applied successfully with RegexReplace ({'I2I' if has_load_image else 'T2I'}, clip='{clip_model_name}', thinking=True).")
|
| 257 |
+
else:
|
| 258 |
+
target_node['inputs']['prompt'] = [text_gen_node_id, 0]
|
| 259 |
+
print(f"[Injector] Qwen-Image-2.1 Prompt Enhancer applied successfully ({'I2I' if has_load_image else 'T2I'}, clip='{clip_model_name}', thinking=False).")
|
core/model_manager.py
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
import gc
|
| 2 |
from typing import List
|
| 3 |
import gradio as gr
|
| 4 |
-
from utils.app_utils import _ensure_model_downloaded
|
| 5 |
-
from core.settings import ALL_MODEL_MAP
|
| 6 |
|
| 7 |
class ModelManager:
|
| 8 |
_instance = None
|
|
@@ -20,13 +20,20 @@ class ModelManager:
|
|
| 20 |
|
| 21 |
def ensure_models_downloaded(self, required_models: List[str], progress):
|
| 22 |
print(f"--- [ModelManager] Ensuring models are downloaded: {required_models} ---")
|
| 23 |
-
for i,
|
|
|
|
|
|
|
| 24 |
if progress and hasattr(progress, '__call__'):
|
| 25 |
-
progress(i / max(len(required_models), 1), desc=f"Checking file: {
|
| 26 |
try:
|
| 27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
except Exception as e:
|
| 29 |
-
raise gr.Error(f"Failed to download model '{
|
| 30 |
print(f"--- [ModelManager] ✅ All required models are present on disk. ---")
|
| 31 |
|
| 32 |
model_manager = ModelManager()
|
|
|
|
| 1 |
import gc
|
| 2 |
from typing import List
|
| 3 |
import gradio as gr
|
| 4 |
+
from utils.app_utils import _ensure_model_downloaded, ensure_file_downloaded
|
| 5 |
+
from core.settings import ALL_MODEL_MAP, ALL_FILE_DOWNLOAD_MAP
|
| 6 |
|
| 7 |
class ModelManager:
|
| 8 |
_instance = None
|
|
|
|
| 20 |
|
| 21 |
def ensure_models_downloaded(self, required_models: List[str], progress):
|
| 22 |
print(f"--- [ModelManager] Ensuring models are downloaded: {required_models} ---")
|
| 23 |
+
for i, item in enumerate(required_models):
|
| 24 |
+
if not item:
|
| 25 |
+
continue
|
| 26 |
if progress and hasattr(progress, '__call__'):
|
| 27 |
+
progress(i / max(len(required_models), 1), desc=f"Checking file: {item}")
|
| 28 |
try:
|
| 29 |
+
if item in ALL_MODEL_MAP:
|
| 30 |
+
_ensure_model_downloaded(item, progress)
|
| 31 |
+
elif item in ALL_FILE_DOWNLOAD_MAP:
|
| 32 |
+
ensure_file_downloaded(item, progress)
|
| 33 |
+
else:
|
| 34 |
+
_ensure_model_downloaded(item, progress)
|
| 35 |
except Exception as e:
|
| 36 |
+
raise gr.Error(f"Failed to download model '{item}'. Reason: {e}")
|
| 37 |
print(f"--- [ModelManager] ✅ All required models are present on disk. ---")
|
| 38 |
|
| 39 |
model_manager = ModelManager()
|
core/pipelines/pipeline_input_processor.py
CHANGED
|
@@ -457,6 +457,15 @@ def process_pipeline_inputs(ui_inputs: Dict[str, Any], progress: gr.Progress, wo
|
|
| 457 |
"x": int(xs[i]), "y": int(ys[i]), "strength": float(strengths[i])
|
| 458 |
})
|
| 459 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 460 |
return {
|
| 461 |
"active_loras_for_gpu": active_loras_for_gpu,
|
| 462 |
"active_loras_for_meta": active_loras_for_meta,
|
|
@@ -477,5 +486,6 @@ def process_pipeline_inputs(ui_inputs: Dict[str, Any], progress: gr.Progress, wo
|
|
| 477 |
"active_qwen_image_edit": active_qwen_image_edit,
|
| 478 |
"active_reference_images": active_reference_images,
|
| 479 |
"active_conditioning": active_conditioning,
|
|
|
|
| 480 |
"temp_files_to_clean": temp_files_to_clean
|
| 481 |
}
|
|
|
|
| 457 |
"x": int(xs[i]), "y": int(ys[i]), "strength": float(strengths[i])
|
| 458 |
})
|
| 459 |
|
| 460 |
+
pe_enable = ui_inputs.get('qwen_image_2_1_prompt_enhancer_enable', False)
|
| 461 |
+
active_qwen_image_2_1_prompt_enhancer = []
|
| 462 |
+
if 'qwen_image_2_1_prompt_enhancer' in enabled_chains and pe_enable:
|
| 463 |
+
active_qwen_image_2_1_prompt_enhancer.append({
|
| 464 |
+
"enable": True,
|
| 465 |
+
"thinking": bool(ui_inputs.get('qwen_image_2_1_prompt_enhancer_thinking', False)),
|
| 466 |
+
"max_length": int(ui_inputs.get('qwen_image_2_1_prompt_enhancer_max_length', 512))
|
| 467 |
+
})
|
| 468 |
+
|
| 469 |
return {
|
| 470 |
"active_loras_for_gpu": active_loras_for_gpu,
|
| 471 |
"active_loras_for_meta": active_loras_for_meta,
|
|
|
|
| 486 |
"active_qwen_image_edit": active_qwen_image_edit,
|
| 487 |
"active_reference_images": active_reference_images,
|
| 488 |
"active_conditioning": active_conditioning,
|
| 489 |
+
"active_qwen_image_2_1_prompt_enhancer": active_qwen_image_2_1_prompt_enhancer,
|
| 490 |
"temp_files_to_clean": temp_files_to_clean
|
| 491 |
}
|
core/pipelines/sd_image_pipeline.py
CHANGED
|
@@ -122,6 +122,7 @@ class SdImagePipeline(BasePipeline):
|
|
| 122 |
active_qwen_image_edit = processed.get("active_qwen_image_edit", [])
|
| 123 |
active_reference_images = processed.get("active_reference_images", [])
|
| 124 |
active_conditioning = processed["active_conditioning"]
|
|
|
|
| 125 |
|
| 126 |
loras_string = f"LoRAs: [{', '.join(active_loras_for_meta)}]" if active_loras_for_meta else ""
|
| 127 |
|
|
@@ -182,6 +183,7 @@ class SdImagePipeline(BasePipeline):
|
|
| 182 |
"krea2_style_reference_chain": active_krea2_reference_edit,
|
| 183 |
"qwen_image_edit_chain": active_qwen_image_edit,
|
| 184 |
"reference_image_chain": active_reference_images,
|
|
|
|
| 185 |
"vae_chain": [ui_inputs.get('vae_name')] if (ui_inputs.get('vae_name') and 'vae' in enabled_chains) else [],
|
| 186 |
"hidream_o1_smoothing_chain": hidream_o1_smoothing_data,
|
| 187 |
"pid_chain": [ui_inputs.get('pid_settings', 'OFF')] if is_pid_enabled else [],
|
|
@@ -213,6 +215,18 @@ class SdImagePipeline(BasePipeline):
|
|
| 213 |
workflow_inputs['height'] = ui_inputs['height']
|
| 214 |
|
| 215 |
workflow = assembler.assemble(workflow_inputs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 216 |
|
| 217 |
progress(1.0, desc="All models ready. Requesting GPU for generation...")
|
| 218 |
|
|
|
|
| 122 |
active_qwen_image_edit = processed.get("active_qwen_image_edit", [])
|
| 123 |
active_reference_images = processed.get("active_reference_images", [])
|
| 124 |
active_conditioning = processed["active_conditioning"]
|
| 125 |
+
active_qwen_image_2_1_prompt_enhancer = processed.get("active_qwen_image_2_1_prompt_enhancer", [])
|
| 126 |
|
| 127 |
loras_string = f"LoRAs: [{', '.join(active_loras_for_meta)}]" if active_loras_for_meta else ""
|
| 128 |
|
|
|
|
| 183 |
"krea2_style_reference_chain": active_krea2_reference_edit,
|
| 184 |
"qwen_image_edit_chain": active_qwen_image_edit,
|
| 185 |
"reference_image_chain": active_reference_images,
|
| 186 |
+
"qwen_image_2_1_prompt_enhancer_chain": active_qwen_image_2_1_prompt_enhancer,
|
| 187 |
"vae_chain": [ui_inputs.get('vae_name')] if (ui_inputs.get('vae_name') and 'vae' in enabled_chains) else [],
|
| 188 |
"hidream_o1_smoothing_chain": hidream_o1_smoothing_data,
|
| 189 |
"pid_chain": [ui_inputs.get('pid_settings', 'OFF')] if is_pid_enabled else [],
|
|
|
|
| 215 |
workflow_inputs['height'] = ui_inputs['height']
|
| 216 |
|
| 217 |
workflow = assembler.assemble(workflow_inputs)
|
| 218 |
+
|
| 219 |
+
post_assembled_models = []
|
| 220 |
+
for node_id, node in workflow.items():
|
| 221 |
+
if not isinstance(node, dict):
|
| 222 |
+
continue
|
| 223 |
+
inputs = node.get('inputs', {})
|
| 224 |
+
for k, v in inputs.items():
|
| 225 |
+
if isinstance(v, str) and v in ALL_FILE_DOWNLOAD_MAP:
|
| 226 |
+
if v not in required_models and v not in post_assembled_models:
|
| 227 |
+
post_assembled_models.append(v)
|
| 228 |
+
if post_assembled_models:
|
| 229 |
+
self.model_manager.ensure_models_downloaded(post_assembled_models, progress=progress)
|
| 230 |
|
| 231 |
progress(1.0, desc="All models ready. Requesting GPU for generation...")
|
| 232 |
|
core/pipelines/workflow_recipes/_partials/conditioning/qwen-image-2.1.yaml
CHANGED
|
@@ -70,6 +70,10 @@ dynamic_reference_image_chains:
|
|
| 70 |
vae_node: "vae_loader"
|
| 71 |
max_images: 10
|
| 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
ui_map:
|
| 74 |
positive_prompt: "prompt:prompt"
|
| 75 |
negative_prompt: "prompt:negative_prompt"
|
|
|
|
| 70 |
vae_node: "vae_loader"
|
| 71 |
max_images: 10
|
| 72 |
|
| 73 |
+
dynamic_qwen_image_2_1_prompt_enhancer_chains:
|
| 74 |
+
qwen_image_2_1_prompt_enhancer_chain:
|
| 75 |
+
target_node: "prompt"
|
| 76 |
+
|
| 77 |
ui_map:
|
| 78 |
positive_prompt: "prompt:prompt"
|
| 79 |
negative_prompt: "prompt:negative_prompt"
|
mcp_tools/common.py
CHANGED
|
@@ -670,6 +670,16 @@ def _execute_imagegen_pipeline(task_id: str, params: dict):
|
|
| 670 |
if isinstance(is_enabled, str):
|
| 671 |
is_enabled = is_enabled.upper() in ("ON", "TRUE", "1")
|
| 672 |
ui_inputs["pid_settings"] = "ON" if is_enabled else "OFF"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 673 |
elif itype == "krea2_identity_edit":
|
| 674 |
img = _parse_image_param(item.get("image"))
|
| 675 |
if img:
|
|
@@ -773,6 +783,19 @@ def _execute_imagegen_pipeline(task_id: str, params: dict):
|
|
| 773 |
else:
|
| 774 |
ui_inputs["pid_settings"] = "OFF"
|
| 775 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 776 |
_TASKS_DB[task_id]["progress"] = 50
|
| 777 |
|
| 778 |
# Execute Pipeline
|
|
|
|
| 670 |
if isinstance(is_enabled, str):
|
| 671 |
is_enabled = is_enabled.upper() in ("ON", "TRUE", "1")
|
| 672 |
ui_inputs["pid_settings"] = "ON" if is_enabled else "OFF"
|
| 673 |
+
elif itype == "qwen_image_2_1_prompt_enhancer":
|
| 674 |
+
is_enabled = item.get("enable", item.get("enabled", True))
|
| 675 |
+
if isinstance(is_enabled, str):
|
| 676 |
+
is_enabled = is_enabled.upper() in ("ON", "TRUE", "1")
|
| 677 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_enable"] = bool(is_enabled)
|
| 678 |
+
is_thinking = item.get("thinking", False)
|
| 679 |
+
if isinstance(is_thinking, str):
|
| 680 |
+
is_thinking = is_thinking.upper() in ("ON", "TRUE", "1")
|
| 681 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_thinking"] = bool(is_thinking)
|
| 682 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_max_length"] = int(item.get("max_length", 512))
|
| 683 |
elif itype == "krea2_identity_edit":
|
| 684 |
img = _parse_image_param(item.get("image"))
|
| 685 |
if img:
|
|
|
|
| 783 |
else:
|
| 784 |
ui_inputs["pid_settings"] = "OFF"
|
| 785 |
|
| 786 |
+
pe_val = params.get("qwen_image_2_1_prompt_enhancer") or params.get("qwen_image_2_1_prompt_enhancer_enable")
|
| 787 |
+
if pe_val is not None:
|
| 788 |
+
if isinstance(pe_val, bool):
|
| 789 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_enable"] = pe_val
|
| 790 |
+
elif isinstance(pe_val, dict):
|
| 791 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_enable"] = bool(pe_val.get("enable", pe_val.get("enabled", True)))
|
| 792 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_thinking"] = bool(pe_val.get("thinking", False))
|
| 793 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_max_length"] = int(pe_val.get("max_length", 512))
|
| 794 |
+
elif str(pe_val).upper() in ("ON", "TRUE", "1"):
|
| 795 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_enable"] = True
|
| 796 |
+
else:
|
| 797 |
+
ui_inputs["qwen_image_2_1_prompt_enhancer_enable"] = False
|
| 798 |
+
|
| 799 |
_TASKS_DB[task_id]["progress"] = 50
|
| 800 |
|
| 801 |
# Execute Pipeline
|
ui/events/change_handlers.py
CHANGED
|
@@ -18,7 +18,7 @@ from .config_loaders import (
|
|
| 18 |
load_ipadapter_config
|
| 19 |
)
|
| 20 |
|
| 21 |
-
def make_update_fn(m_comp, cat_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, krea2_identity_edit_acc=None, krea2_reference_edit_acc=None, qwen_image_edit_acc=None, ref_img_acc=None, sensenova_ref_acc=None, task_type=None, type_comp=None):
|
| 22 |
def update_fn(*args):
|
| 23 |
arch = args[0]
|
| 24 |
category = args[1]
|
|
@@ -90,6 +90,7 @@ def make_update_fn(m_comp, cat_comp, ar_comp, width_comp, height_comp, cn_types,
|
|
| 90 |
if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
|
| 91 |
if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
|
| 92 |
if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
|
|
|
|
| 93 |
|
| 94 |
if ar_comp:
|
| 95 |
res_key = arch_model_type
|
|
@@ -157,7 +158,7 @@ def make_update_fn(m_comp, cat_comp, ar_comp, width_comp, height_comp, cn_types,
|
|
| 157 |
return update_fn
|
| 158 |
|
| 159 |
|
| 160 |
-
def make_model_change_fn(cat_comp_ref, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, arch_comp_ref, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, krea2_identity_edit_acc=None, krea2_reference_edit_acc=None, qwen_image_edit_acc=None, ref_img_acc=None, sensenova_ref_acc=None, task_type=None, type_comp=None):
|
| 161 |
def change_fn(*args):
|
| 162 |
model_name = args[0]
|
| 163 |
idx = 1
|
|
@@ -237,6 +238,7 @@ def make_model_change_fn(cat_comp_ref, ar_comp, width_comp, height_comp, cn_type
|
|
| 237 |
if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
|
| 238 |
if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
|
| 239 |
if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
|
|
|
|
| 240 |
|
| 241 |
if ar_comp:
|
| 242 |
res_key = arch_model_type
|
|
|
|
| 18 |
load_ipadapter_config
|
| 19 |
)
|
| 20 |
|
| 21 |
+
def make_update_fn(m_comp, cat_comp, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, krea2_identity_edit_acc=None, krea2_reference_edit_acc=None, qwen_image_edit_acc=None, ref_img_acc=None, sensenova_ref_acc=None, task_type=None, type_comp=None, qwen_pe_acc=None):
|
| 22 |
def update_fn(*args):
|
| 23 |
arch = args[0]
|
| 24 |
category = args[1]
|
|
|
|
| 90 |
if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
|
| 91 |
if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
|
| 92 |
if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
|
| 93 |
+
if qwen_pe_acc: updates[qwen_pe_acc] = gr.update(visible=('qwen_image_2_1_prompt_enhancer' in enabled_chains))
|
| 94 |
|
| 95 |
if ar_comp:
|
| 96 |
res_key = arch_model_type
|
|
|
|
| 158 |
return update_fn
|
| 159 |
|
| 160 |
|
| 161 |
+
def make_model_change_fn(cat_comp_ref, ar_comp, width_comp, height_comp, cn_types, cn_series, cn_filepaths, anima_cn_types, anima_cn_series, anima_cn_filepaths, diffsynth_cn_types, diffsynth_cn_series, diffsynth_cn_filepaths, krea2_cn_types, krea2_cn_series, krea2_cn_filepaths, arch_comp_ref, ipa_preset, lora_acc, cn_acc, anima_cn_acc, diffsynth_cn_acc, krea2_cn_acc, ipa_acc, sd3_ipa_acc, flux1_ipa_acc, style_acc, embed_acc, cond_acc, ref_latent_acc, hidream_o1_ref_acc, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp, pid_acc=None, vae_acc=None, joyai_ref_acc=None, krea2_identity_edit_acc=None, krea2_reference_edit_acc=None, qwen_image_edit_acc=None, ref_img_acc=None, sensenova_ref_acc=None, task_type=None, type_comp=None, qwen_pe_acc=None):
|
| 162 |
def change_fn(*args):
|
| 163 |
model_name = args[0]
|
| 164 |
idx = 1
|
|
|
|
| 238 |
if ref_img_acc: updates[ref_img_acc] = gr.update(visible=('reference_image' in enabled_chains))
|
| 239 |
if pid_acc: updates[pid_acc] = gr.update(visible=('pid' in enabled_chains))
|
| 240 |
if vae_acc: updates[vae_acc] = gr.update(visible=('vae' in enabled_chains))
|
| 241 |
+
if qwen_pe_acc: updates[qwen_pe_acc] = gr.update(visible=('qwen_image_2_1_prompt_enhancer' in enabled_chains))
|
| 242 |
|
| 243 |
if ar_comp:
|
| 244 |
res_key = arch_model_type
|
ui/events/main.py
CHANGED
|
@@ -80,6 +80,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 80 |
ref_img_accordion = ui_components.get(f'reference_image_accordion_{prefix}')
|
| 81 |
pid_accordion = ui_components.get(f'pid_accordion_{prefix}')
|
| 82 |
vae_accordion = ui_components.get(f'vae_accordion_{prefix}')
|
|
|
|
| 83 |
|
| 84 |
ipa_preset_list = ui_components.get(f'ipadapter_final_preset_{prefix}')
|
| 85 |
|
|
@@ -121,6 +122,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 121 |
if ref_img_accordion: outputs.append(ref_img_accordion)
|
| 122 |
if pid_accordion: outputs.append(pid_accordion)
|
| 123 |
if vae_accordion: outputs.append(vae_accordion)
|
|
|
|
| 124 |
if ipa_preset_list: outputs.append(ipa_preset_list)
|
| 125 |
|
| 126 |
outputs.extend(valid_extra_comps)
|
|
@@ -134,7 +136,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 134 |
ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
|
| 135 |
ref_latent_accordion, hidream_o1_ref_accordion, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
|
| 136 |
pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, krea2_identity_edit_acc=krea2_identity_edit_accordion, krea2_reference_edit_acc=krea2_reference_edit_accordion, qwen_image_edit_acc=qwen_image_edit_accordion, ref_img_acc=ref_img_accordion, sensenova_ref_acc=sensenova_ref_accordion,
|
| 137 |
-
task_type=task_type, type_comp=type_comp
|
| 138 |
)
|
| 139 |
inputs = [arch_comp, cat_comp]
|
| 140 |
if aspect_ratio_comp:
|
|
@@ -174,6 +176,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 174 |
if ref_img_accordion: outputs2.append(ref_img_accordion)
|
| 175 |
if pid_accordion: outputs2.append(pid_accordion)
|
| 176 |
if vae_accordion: outputs2.append(vae_accordion)
|
|
|
|
| 177 |
if ipa_preset_list: outputs2.append(ipa_preset_list)
|
| 178 |
|
| 179 |
outputs2.extend(valid_extra_comps)
|
|
@@ -192,7 +195,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 192 |
arch_comp, ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
|
| 193 |
ref_latent_accordion, hidream_o1_ref_accordion, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
|
| 194 |
pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, krea2_identity_edit_acc=krea2_identity_edit_accordion, krea2_reference_edit_acc=krea2_reference_edit_accordion, qwen_image_edit_acc=qwen_image_edit_accordion, ref_img_acc=ref_img_accordion, sensenova_ref_acc=sensenova_ref_accordion,
|
| 195 |
-
task_type=task_type, type_comp=type_comp
|
| 196 |
)
|
| 197 |
if type_comp:
|
| 198 |
inputs2.append(type_comp)
|
|
@@ -286,6 +289,7 @@ def attach_event_handlers(ui_components, demo):
|
|
| 286 |
f'krea2_reference_edit_accordion_{prefix}': 'krea2_style_reference',
|
| 287 |
f'qwen_image_edit_accordion_{prefix}': 'qwen_image_edit',
|
| 288 |
f'reference_image_accordion_{prefix}': 'reference_image',
|
|
|
|
| 289 |
f'pid_accordion_{prefix}': 'pid',
|
| 290 |
f'vae_accordion_{prefix}': 'vae',
|
| 291 |
}
|
|
@@ -323,7 +327,9 @@ def attach_event_handlers(ui_components, demo):
|
|
| 323 |
'hidream_o1_reference_accordion_imagegen', 'sensenova_reference_accordion_imagegen', 'joyai_reference_accordion_imagegen',
|
| 324 |
'krea2_identity_edit_accordion_imagegen', 'krea2_reference_edit_accordion_imagegen',
|
| 325 |
'qwen_image_edit_accordion_imagegen',
|
| 326 |
-
'reference_image_accordion_imagegen',
|
|
|
|
|
|
|
| 327 |
]
|
| 328 |
for key in accordion_keys:
|
| 329 |
if key in ui_components:
|
|
|
|
| 80 |
ref_img_accordion = ui_components.get(f'reference_image_accordion_{prefix}')
|
| 81 |
pid_accordion = ui_components.get(f'pid_accordion_{prefix}')
|
| 82 |
vae_accordion = ui_components.get(f'vae_accordion_{prefix}')
|
| 83 |
+
qwen_pe_accordion = ui_components.get(f'qwen_image_2_1_prompt_enhancer_accordion_{prefix}')
|
| 84 |
|
| 85 |
ipa_preset_list = ui_components.get(f'ipadapter_final_preset_{prefix}')
|
| 86 |
|
|
|
|
| 122 |
if ref_img_accordion: outputs.append(ref_img_accordion)
|
| 123 |
if pid_accordion: outputs.append(pid_accordion)
|
| 124 |
if vae_accordion: outputs.append(vae_accordion)
|
| 125 |
+
if qwen_pe_accordion: outputs.append(qwen_pe_accordion)
|
| 126 |
if ipa_preset_list: outputs.append(ipa_preset_list)
|
| 127 |
|
| 128 |
outputs.extend(valid_extra_comps)
|
|
|
|
| 136 |
ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
|
| 137 |
ref_latent_accordion, hidream_o1_ref_accordion, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
|
| 138 |
pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, krea2_identity_edit_acc=krea2_identity_edit_accordion, krea2_reference_edit_acc=krea2_reference_edit_accordion, qwen_image_edit_acc=qwen_image_edit_accordion, ref_img_acc=ref_img_accordion, sensenova_ref_acc=sensenova_ref_accordion,
|
| 139 |
+
task_type=task_type, type_comp=type_comp, qwen_pe_acc=qwen_pe_accordion
|
| 140 |
)
|
| 141 |
inputs = [arch_comp, cat_comp]
|
| 142 |
if aspect_ratio_comp:
|
|
|
|
| 176 |
if ref_img_accordion: outputs2.append(ref_img_accordion)
|
| 177 |
if pid_accordion: outputs2.append(pid_accordion)
|
| 178 |
if vae_accordion: outputs2.append(vae_accordion)
|
| 179 |
+
if qwen_pe_accordion: outputs2.append(qwen_pe_accordion)
|
| 180 |
if ipa_preset_list: outputs2.append(ipa_preset_list)
|
| 181 |
|
| 182 |
outputs2.extend(valid_extra_comps)
|
|
|
|
| 195 |
arch_comp, ipa_preset_list, lora_accordion, cn_accordion, anima_cn_accordion, diffsynth_cn_accordion, krea2_cn_accordion, ipa_accordion, sd3_ipa_accordion, flux1_ipa_accordion, style_accordion, embedding_accordion, conditioning_accordion,
|
| 196 |
ref_latent_accordion, hidream_o1_ref_accordion, prompt_comp, neg_prompt_comp, steps_comp, cfg_comp, sampler_comp, scheduler_comp,
|
| 197 |
pid_acc=pid_accordion, vae_acc=vae_accordion, joyai_ref_acc=joyai_ref_accordion, krea2_identity_edit_acc=krea2_identity_edit_accordion, krea2_reference_edit_acc=krea2_reference_edit_accordion, qwen_image_edit_acc=qwen_image_edit_accordion, ref_img_acc=ref_img_accordion, sensenova_ref_acc=sensenova_ref_accordion,
|
| 198 |
+
task_type=task_type, type_comp=type_comp, qwen_pe_acc=qwen_pe_accordion
|
| 199 |
)
|
| 200 |
if type_comp:
|
| 201 |
inputs2.append(type_comp)
|
|
|
|
| 289 |
f'krea2_reference_edit_accordion_{prefix}': 'krea2_style_reference',
|
| 290 |
f'qwen_image_edit_accordion_{prefix}': 'qwen_image_edit',
|
| 291 |
f'reference_image_accordion_{prefix}': 'reference_image',
|
| 292 |
+
f'qwen_image_2_1_prompt_enhancer_accordion_{prefix}': 'qwen_image_2_1_prompt_enhancer',
|
| 293 |
f'pid_accordion_{prefix}': 'pid',
|
| 294 |
f'vae_accordion_{prefix}': 'vae',
|
| 295 |
}
|
|
|
|
| 327 |
'hidream_o1_reference_accordion_imagegen', 'sensenova_reference_accordion_imagegen', 'joyai_reference_accordion_imagegen',
|
| 328 |
'krea2_identity_edit_accordion_imagegen', 'krea2_reference_edit_accordion_imagegen',
|
| 329 |
'qwen_image_edit_accordion_imagegen',
|
| 330 |
+
'reference_image_accordion_imagegen',
|
| 331 |
+
'qwen_image_2_1_prompt_enhancer_accordion_imagegen',
|
| 332 |
+
'pid_accordion_imagegen', 'vae_accordion_imagegen',
|
| 333 |
]
|
| 334 |
for key in accordion_keys:
|
| 335 |
if key in ui_components:
|
ui/events/run_handlers.py
CHANGED
|
@@ -37,6 +37,13 @@ def create_run_event(prefix: str, task_type: str = None, ui_components: dict = N
|
|
| 37 |
if ui_components.get(f'pid_settings_{prefix}'):
|
| 38 |
run_inputs_map['pid_settings'] = ui_components[f'pid_settings_{prefix}']
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
# Generic / Txt2Img inputs
|
| 41 |
if ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width'):
|
| 42 |
run_inputs_map['width'] = ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width')
|
|
|
|
| 37 |
if ui_components.get(f'pid_settings_{prefix}'):
|
| 38 |
run_inputs_map['pid_settings'] = ui_components[f'pid_settings_{prefix}']
|
| 39 |
|
| 40 |
+
if ui_components.get(f'qwen_image_2_1_prompt_enhancer_enable_{prefix}'):
|
| 41 |
+
run_inputs_map['qwen_image_2_1_prompt_enhancer_enable'] = ui_components[f'qwen_image_2_1_prompt_enhancer_enable_{prefix}']
|
| 42 |
+
if ui_components.get(f'qwen_image_2_1_prompt_enhancer_thinking_{prefix}'):
|
| 43 |
+
run_inputs_map['qwen_image_2_1_prompt_enhancer_thinking'] = ui_components[f'qwen_image_2_1_prompt_enhancer_thinking_{prefix}']
|
| 44 |
+
if ui_components.get(f'qwen_image_2_1_prompt_enhancer_max_length_{prefix}'):
|
| 45 |
+
run_inputs_map['qwen_image_2_1_prompt_enhancer_max_length'] = ui_components[f'qwen_image_2_1_prompt_enhancer_max_length_{prefix}']
|
| 46 |
+
|
| 47 |
# Generic / Txt2Img inputs
|
| 48 |
if ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width'):
|
| 49 |
run_inputs_map['width'] = ui_components.get(f'width_{prefix}') or ui_components.get(f'{prefix}_width')
|
ui/imagegen_ui.py
CHANGED
|
@@ -9,7 +9,7 @@ from .shared.ui_components import (
|
|
| 9 |
create_model_architecture_filter_ui, create_category_filter_ui,
|
| 10 |
create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
|
| 11 |
create_reference_latent_ui, create_hidream_o1_reference_ui, create_sensenova_reference_ui, create_joyai_reference_ui, create_reference_image_ui, create_krea2_identity_edit_ui, create_krea2_reference_edit_ui, create_qwen_image_edit_ui,
|
| 12 |
-
create_pid_ui
|
| 13 |
)
|
| 14 |
|
| 15 |
default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
|
|
@@ -190,6 +190,7 @@ def create_ui():
|
|
| 190 |
components.update(create_krea2_reference_edit_ui(prefix))
|
| 191 |
components.update(create_qwen_image_edit_ui(prefix))
|
| 192 |
components.update(create_reference_image_ui(prefix))
|
|
|
|
| 193 |
components.update(create_vae_override_ui(prefix))
|
| 194 |
components.update(create_pid_ui(prefix))
|
| 195 |
components[f'accordion_wrapper_{prefix}'] = accordion_wrapper
|
|
|
|
| 9 |
create_model_architecture_filter_ui, create_category_filter_ui,
|
| 10 |
create_sd3_ipadapter_ui, create_flux1_ipadapter_ui, create_style_ui,
|
| 11 |
create_reference_latent_ui, create_hidream_o1_reference_ui, create_sensenova_reference_ui, create_joyai_reference_ui, create_reference_image_ui, create_krea2_identity_edit_ui, create_krea2_reference_edit_ui, create_qwen_image_edit_ui,
|
| 12 |
+
create_pid_ui, create_qwen_image_2_1_prompt_enhancer_ui
|
| 13 |
)
|
| 14 |
|
| 15 |
default_vals = MODEL_DEFAULTS_CONFIG.get('Default', {})
|
|
|
|
| 190 |
components.update(create_krea2_reference_edit_ui(prefix))
|
| 191 |
components.update(create_qwen_image_edit_ui(prefix))
|
| 192 |
components.update(create_reference_image_ui(prefix))
|
| 193 |
+
components.update(create_qwen_image_2_1_prompt_enhancer_ui(prefix))
|
| 194 |
components.update(create_vae_override_ui(prefix))
|
| 195 |
components.update(create_pid_ui(prefix))
|
| 196 |
components[f'accordion_wrapper_{prefix}'] = accordion_wrapper
|
ui/shared/ui_components.py
CHANGED
|
@@ -882,4 +882,37 @@ def create_krea2_reference_edit_ui(prefix: str, max_units=3):
|
|
| 882 |
|
| 883 |
components[key('all_krea2_reference_edit_components_flat')] = ref_image_inputs
|
| 884 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 885 |
return components
|
|
|
|
| 882 |
|
| 883 |
components[key('all_krea2_reference_edit_components_flat')] = ref_image_inputs
|
| 884 |
|
| 885 |
+
return components
|
| 886 |
+
|
| 887 |
+
|
| 888 |
+
def create_qwen_image_2_1_prompt_enhancer_ui(prefix: str):
|
| 889 |
+
components = {}
|
| 890 |
+
key = lambda name: f"{name}_{prefix}"
|
| 891 |
+
|
| 892 |
+
with gr.Accordion("Qwen-Image-2.1 Prompt Enhancer Settings", open=False, visible=('qwen_image_2_1_prompt_enhancer' in default_enabled_chains)) as enhancer_accordion:
|
| 893 |
+
components[key('qwen_image_2_1_prompt_enhancer_accordion')] = enhancer_accordion
|
| 894 |
+
gr.Markdown("💡 **Tip:** Automatically selects between [Qwen-Image-2.1-PE-T2I](https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I) and [Qwen-Image-2.1-PE-I2I](https://huggingface.co/Qwen/Qwen-Image-2.1-PE-I2I) for prompt enhancement based on input context.")
|
| 895 |
+
with gr.Row():
|
| 896 |
+
components[key('qwen_image_2_1_prompt_enhancer_enable')] = gr.Checkbox(
|
| 897 |
+
label="Enable Qwen-Image-2.1 Prompt Enhancer",
|
| 898 |
+
value=False,
|
| 899 |
+
interactive=True,
|
| 900 |
+
scale=2
|
| 901 |
+
)
|
| 902 |
+
components[key('qwen_image_2_1_prompt_enhancer_thinking')] = gr.Checkbox(
|
| 903 |
+
label="Enable Thinking (Reasoning CoT)",
|
| 904 |
+
value=False,
|
| 905 |
+
interactive=True,
|
| 906 |
+
scale=2
|
| 907 |
+
)
|
| 908 |
+
components[key('qwen_image_2_1_prompt_enhancer_max_length')] = gr.Number(
|
| 909 |
+
label="Max Length",
|
| 910 |
+
value=512,
|
| 911 |
+
precision=0,
|
| 912 |
+
minimum=16,
|
| 913 |
+
step=16,
|
| 914 |
+
interactive=True,
|
| 915 |
+
scale=1
|
| 916 |
+
)
|
| 917 |
+
|
| 918 |
return components
|
yaml/chain_features.yaml
CHANGED
|
@@ -525,3 +525,25 @@ embedding:
|
|
| 525 |
required:
|
| 526 |
- source
|
| 527 |
- embedding_value
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 525 |
required:
|
| 526 |
- source
|
| 527 |
- embedding_value
|
| 528 |
+
|
| 529 |
+
qwen_image_2_1_prompt_enhancer:
|
| 530 |
+
chains: qwen_image_2_1_prompt_enhancer
|
| 531 |
+
display_name: "Qwen-Image-2.1 Prompt Enhancer Settings"
|
| 532 |
+
description: "Enhance prompts using Qwen-Image-2.1 Prompt Enhancer with automatic model selection (T2I vs I2I/reference images) and optional reasoning CoT thinking mode."
|
| 533 |
+
max_count: 1
|
| 534 |
+
usage_guideline: "Set enable to true to activate prompt enhancement. Enable thinking to allow reasoning CoT (reasoning tags are automatically stripped from the resulting prompt). Set max_length to control generation length."
|
| 535 |
+
parameters_schema:
|
| 536 |
+
type: object
|
| 537 |
+
properties:
|
| 538 |
+
enable:
|
| 539 |
+
type: boolean
|
| 540 |
+
default: false
|
| 541 |
+
description: "Enable Qwen-Image-2.1 Prompt Enhancer."
|
| 542 |
+
thinking:
|
| 543 |
+
type: boolean
|
| 544 |
+
default: false
|
| 545 |
+
description: "Enable thinking (reasoning CoT). Thought text inside <think> tags will be automatically stripped before passing prompt to downstream text encoding."
|
| 546 |
+
max_length:
|
| 547 |
+
type: integer
|
| 548 |
+
default: 512
|
| 549 |
+
description: "Maximum generation length for prompt enhancement."
|
yaml/file_list.yaml
CHANGED
|
@@ -911,6 +911,15 @@ file:
|
|
| 911 |
source: "hf"
|
| 912 |
repo_id: "Comfy-Org/Ideogram-4"
|
| 913 |
repository_file_path: "text_encoders/qwen3vl_8b_nvfp4.safetensors"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 914 |
# PixelDiT
|
| 915 |
- filename: "gemma_2_2b_it_elm_fp8_scaled.safetensors"
|
| 916 |
source: "hf"
|
|
@@ -1006,10 +1015,6 @@ file:
|
|
| 1006 |
source: "hf"
|
| 1007 |
repo_id: "Comfy-Org/Omnigen2_ComfyUI_repackaged"
|
| 1008 |
repository_file_path: "split_files/text_encoders/qwen_2.5_vl_fp16.safetensors"
|
| 1009 |
-
- filename: "qwen3vl_8b_w4a8.safetensors"
|
| 1010 |
-
source: "hf"
|
| 1011 |
-
repo_id: "Comfy-Org/Qwen-Image-2.1"
|
| 1012 |
-
repository_file_path: "text_encoders/qwen3vl_8b_w4a8.safetensors"
|
| 1013 |
vae:
|
| 1014 |
- filename: "wan_2.1_vae.safetensors"
|
| 1015 |
source: "hf"
|
|
|
|
| 911 |
source: "hf"
|
| 912 |
repo_id: "Comfy-Org/Ideogram-4"
|
| 913 |
repository_file_path: "text_encoders/qwen3vl_8b_nvfp4.safetensors"
|
| 914 |
+
# Qwen-Image-2.1 Prompt Enhancers
|
| 915 |
+
- filename: "qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors"
|
| 916 |
+
source: "hf"
|
| 917 |
+
repo_id: "Comfy-Org/Qwen-Image-2.1"
|
| 918 |
+
repository_file_path: "text_encoders/qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors"
|
| 919 |
+
- filename: "qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors"
|
| 920 |
+
source: "hf"
|
| 921 |
+
repo_id: "Comfy-Org/Qwen-Image-2.1"
|
| 922 |
+
repository_file_path: "text_encoders/qwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot.safetensors"
|
| 923 |
# PixelDiT
|
| 924 |
- filename: "gemma_2_2b_it_elm_fp8_scaled.safetensors"
|
| 925 |
source: "hf"
|
|
|
|
| 1015 |
source: "hf"
|
| 1016 |
repo_id: "Comfy-Org/Omnigen2_ComfyUI_repackaged"
|
| 1017 |
repository_file_path: "split_files/text_encoders/qwen_2.5_vl_fp16.safetensors"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1018 |
vae:
|
| 1019 |
- filename: "wan_2.1_vae.safetensors"
|
| 1020 |
source: "hf"
|
yaml/model_architecture_features.yaml
CHANGED
|
@@ -74,6 +74,7 @@ qwen-image-2.1:
|
|
| 74 |
enabled_chains:
|
| 75 |
- lora
|
| 76 |
- reference_image
|
|
|
|
| 77 |
|
| 78 |
qwen-image:
|
| 79 |
enabled_chains:
|
|
|
|
| 74 |
enabled_chains:
|
| 75 |
- lora
|
| 76 |
- reference_image
|
| 77 |
+
- qwen_image_2_1_prompt_enhancer
|
| 78 |
|
| 79 |
qwen-image:
|
| 80 |
enabled_chains:
|
yaml/model_list.yaml
CHANGED
|
@@ -19,7 +19,7 @@ Checkpoint:
|
|
| 19 |
components:
|
| 20 |
unet: "qwen_image_2.1_int8_convrot.safetensors"
|
| 21 |
vae: "qwen_image_2.1_vae_bf16.safetensors"
|
| 22 |
-
clip: "
|
| 23 |
SenseNova-U1.5:
|
| 24 |
latent_type: hidream_o1_latent
|
| 25 |
models:
|
|
@@ -319,7 +319,7 @@ Checkpoint:
|
|
| 319 |
unet: "novaAnimeAM_v40.safetensors"
|
| 320 |
vae: "qwen_image_vae.safetensors"
|
| 321 |
clip: "qwen_3_06b_base.safetensors"
|
| 322 |
-
- display_name: "
|
| 323 |
components:
|
| 324 |
unet: "ramthrustsNSFWPINK_v31RAMSDelicious.safetensors"
|
| 325 |
vae: "qwen_image_vae.safetensors"
|
|
|
|
| 19 |
components:
|
| 20 |
unet: "qwen_image_2.1_int8_convrot.safetensors"
|
| 21 |
vae: "qwen_image_2.1_vae_bf16.safetensors"
|
| 22 |
+
clip: "qwen3vl_8b_nvfp4.safetensors"
|
| 23 |
SenseNova-U1.5:
|
| 24 |
latent_type: hidream_o1_latent
|
| 25 |
models:
|
|
|
|
| 319 |
unet: "novaAnimeAM_v40.safetensors"
|
| 320 |
vae: "qwen_image_vae.safetensors"
|
| 321 |
clip: "qwen_3_06b_base.safetensors"
|
| 322 |
+
- display_name: "Crody/RAMTHRUST'S-NSFW-PINK-ALCHEMY-ANIMA-v3.1"
|
| 323 |
components:
|
| 324 |
unet: "ramthrustsNSFWPINK_v31RAMSDelicious.safetensors"
|
| 325 |
vae: "qwen_image_vae.safetensors"
|
yaml/task_features.yaml
CHANGED
|
@@ -25,6 +25,7 @@ txt2img:
|
|
| 25 |
- krea2_style_reference
|
| 26 |
- qwen_image_edit
|
| 27 |
- reference_image
|
|
|
|
| 28 |
- pid
|
| 29 |
- vae
|
| 30 |
|
|
@@ -50,6 +51,7 @@ img2img:
|
|
| 50 |
- krea2_style_reference
|
| 51 |
- qwen_image_edit
|
| 52 |
- reference_image
|
|
|
|
| 53 |
- vae
|
| 54 |
|
| 55 |
inpaint:
|
|
@@ -74,6 +76,7 @@ inpaint:
|
|
| 74 |
- krea2_style_reference
|
| 75 |
- qwen_image_edit
|
| 76 |
- reference_image
|
|
|
|
| 77 |
- vae
|
| 78 |
|
| 79 |
outpaint:
|
|
@@ -98,6 +101,7 @@ outpaint:
|
|
| 98 |
- krea2_style_reference
|
| 99 |
- qwen_image_edit
|
| 100 |
- reference_image
|
|
|
|
| 101 |
- vae
|
| 102 |
|
| 103 |
hires_fix:
|
|
@@ -122,4 +126,5 @@ hires_fix:
|
|
| 122 |
- krea2_style_reference
|
| 123 |
- qwen_image_edit
|
| 124 |
- reference_image
|
|
|
|
| 125 |
- vae
|
|
|
|
| 25 |
- krea2_style_reference
|
| 26 |
- qwen_image_edit
|
| 27 |
- reference_image
|
| 28 |
+
- qwen_image_2_1_prompt_enhancer
|
| 29 |
- pid
|
| 30 |
- vae
|
| 31 |
|
|
|
|
| 51 |
- krea2_style_reference
|
| 52 |
- qwen_image_edit
|
| 53 |
- reference_image
|
| 54 |
+
- qwen_image_2_1_prompt_enhancer
|
| 55 |
- vae
|
| 56 |
|
| 57 |
inpaint:
|
|
|
|
| 76 |
- krea2_style_reference
|
| 77 |
- qwen_image_edit
|
| 78 |
- reference_image
|
| 79 |
+
- qwen_image_2_1_prompt_enhancer
|
| 80 |
- vae
|
| 81 |
|
| 82 |
outpaint:
|
|
|
|
| 101 |
- krea2_style_reference
|
| 102 |
- qwen_image_edit
|
| 103 |
- reference_image
|
| 104 |
+
- qwen_image_2_1_prompt_enhancer
|
| 105 |
- vae
|
| 106 |
|
| 107 |
hires_fix:
|
|
|
|
| 126 |
- krea2_style_reference
|
| 127 |
- qwen_image_edit
|
| 128 |
- reference_image
|
| 129 |
+
- qwen_image_2_1_prompt_enhancer
|
| 130 |
- vae
|