@BiuBoomFlow Thanks for the reply. I tried, but now the output is just always 1:1 the same as the input. Connected "Load Image" Mask to "LayerUtility: ImageScaleByAspectRatio V2" and from there i connected Mask to "Qwen Edit Config Preparer" but it doesnt work.

@BiuBoomFlow I saved it, this is how it looks like on my end HhFCTuR.png (1847×885)

@BiuBoomFlow alright. well thank you lots for this work flow creation though =D . i havent seen any others out there that do what this one does. i have been enjoying using it for sure. just needa find a img2img model that can do what i want it do and as easy to use right on load up as this one.

@BiuBoomFlow add more detail to skin. but as i play more with this and prompts and such. im also noticing that the image itself is also a factor. cause some results will be pretty good. and others will not be as good. example around the waist will be a clear line and a weird color change where the panties were. like a 3d model and its edges not being connected and its texture was off. but another image will have just perfect waist generation. if i wanted to get deep id learn photoshop. but this is just for fun and messing around. in a month i probably wont even be messing this AI stuff anymore. so i wont even worry much about it anymore really.

@BiuBoomFlow 没有进行过任何节点的改动，重新下载了工作流也还是不行o(╥﹏╥)o

@BiuBoomFlow 谢谢老大，刚才让gemini写了个自动修复脚本居然突然能跑了，困扰了我好几天

import json

import sys

import os

def fix_workflow(file_path):

if not os.path.exists(file_path):

print(f"❌ 找不到文件: {file_path}")

return

print(f"📖 正在读取: {file_path} ...")

try:

with open(file_path, 'r', encoding='utf-8') as f:

data = json.load(f)

except Exception as e:

print(f"❌ JSON 读取失败: {e}")

return

nodes = data.get('nodes', [])

links = data.get('links', [])

# 1. 建立索引

# 找到所有的 Set 节点和 Get 节点

set_nodes = {} # Key: 变量名, Value: {node_id, source_link_id}

get_nodes = [] # List: {node_id, var_name, output_slot_links}

# 辅助：Link 查找表 {link_id: [origin_id, origin_slot, target_id, target_slot, type]}

link_map = {l[0]: l for l in links if l}

print("🔍 正在分析节点逻辑...")

for node in nodes:

node_type = str(node.get('type', '')).lower()

widgets = node.get('widgets_values', [])

# 提取变量名 (通常是 widget 的第一个字符串)

var_name = None

if widgets and isinstance(widgets, list):

for w in widgets:

if isinstance(w, str) and w.strip():

var_name = w

break

if not var_name:

continue

# 判断是否为 Set 节点 (KJNodes / RGThree 等)

if "set" in node_type and ("node" in node_type or "var" in node_type or "image" in node_type):

# 找到 Set 节点的输入源

inputs = node.get('inputs', [])

source_link_id = None

if inputs:

# 假设第一个输入就是数据源 (大部分 SetNode 只有一个主输入)

source_link_id = inputs[0].get('link')

if source_link_id:

if var_name not in set_nodes:

set_nodes[var_name] = []

set_nodes[var_name].append({

'id': node['id'],

'source_link_id': source_link_id

})

# 判断是否为 Get 节点

elif "get" in node_type and ("node" in node_type or "var" in node_type or "image" in node_type):

get_nodes.append({

'id': node['id'],

'var_name': var_name,

'outputs': node.get('outputs', [])

})

# 2. 生成新的 Link ID

# 找到当前最大的 link id，防止冲突

max_link_id = 0

for l in links:

if l and isinstance(l[0], int):

if l[0] > max_link_id:

max_link_id = l[0]

new_link_id_counter = max_link_id + 1

# 3. 执行修复 (重连线)

fixed_count = 0

print("-" * 50)

for get_node in get_nodes:

var_name = get_node['var_name']

# 检查是否有对应的 Set 节点

if var_name in set_nodes:

targets = set_nodes[var_name]

# 如果有多个同名 Set，默认取第一个（通常是最近修改的或唯一的）

target_set = targets[0]

source_link_id = target_set['source_link_id']

# 获取数据源头的节点信息

source_link_data = link_map.get(source_link_id)

if not source_link_data:

continue

origin_node_id = source_link_data[1]

origin_slot = source_link_data[2]

link_type = source_link_data[5]

# 遍历 Get 节点的所有输出，找到它连向的下游节点

for output in get_node['outputs']:

if not output.get('links'):

continue

# 复制一份列表遍历，因为我们要修改 links

current_links = list(output['links'])

for link_id in current_links:

link_data = link_map.get(link_id)

if not link_data:

continue

target_node_id = link_data[3]

target_slot = link_data[4]

# --- 核心操作 ---

# 1. 删除旧的 Link (Get -> 下游)

if link_data in links:

links.remove(link_data)

# 2. 创建新的 Link (源头 -> 下游)

# 格式: [id, origin_id, origin_slot, target_id, target_slot, type]

new_link = [

new_link_id_counter,

origin_node_id,

origin_slot,

target_node_id,

target_slot,

link_type

]

links.append(new_link)

# 3. 更新下游节点的 inputs 数据，让它指向新 Link ID

# 这一步至关重要，否则 UI 上线连上了，但逻辑没变

for node in nodes:

if node['id'] == target_node_id:

for inp in node.get('inputs', []):

# 只有 slot_index 匹配时才更新

# ComfyUI input 没有存 slot index，通常按顺序，或者需要名字匹配

# 这里我们通过旧 link_id 匹配最稳妥

if inp.get('link') == link_id:

inp['link'] = new_link_id_counter

print(f"🔧 修复: 变量 '{var_name}' | 下游节点 ID {target_node_id} 现已直接连接到源头 ID {origin_node_id}")

new_link_id_counter += 1

fixed_count += 1

# 4. 保存文件

if fixed_count > 0:

base, ext = os.path.splitext(file_path)

new_file_path = f"{base}_fixed{ext}"

# 更新数据

data['links'] = links

data['nodes'] = nodes

with open(new_file_path, 'w', encoding='utf-8') as f:

json.dump(data, f, indent=2)

print("-" * 50)

print(f"✅ 成功! 共修复了 {fixed_count} 处连接。")

print(f"📁 已保存修复文件为: {new_file_path}")

print("💡 请在 ComfyUI 中加载这个 '_fixed' 文件。连线可能比较乱，但逻辑是通的。")

else:

print("⚠️ 未发现可修复的 Set/Get 对应关系。请确认 Set 节点有输入连线，且变量名完全一致。")

if name == "__main__":

# 支持拖拽文件

target_file = ""

if len(sys.argv) > 1:

target_file = sys.argv[1]

else:

target_file = input("请输入 .json 工作流文件路径: ").strip().strip('"')

fix_workflow(target_file)

input("\n按回车键退出...")

不过这个gemini修改过的工作流虽然能正常跑了，但是速度慢好多不知为何