V3.2，修复了错误。请把放大改为euler获得更好的体验。一切改变都在工作流的提示里，请查看！😄
V3.2, bug fixed. Please change the zoom to euler for a better experience. All changes are in the workflow prompts, please check them out! 😄
-------------------------------------------------------------------------
V3
已经更新到v3版本，非常简单并且在4070s12GB下出图快速，请你尝试。如果有问题及时告诉我，请积极在下方画廊投稿！！😄
It is very simple and can produce pictures quickly under 4070s12GB. Please try it. If you have any questions, please let me know in time, and please actively contribute to the gallery below! ! 😄
如果出现显存报错问题，请参考以下代码作为comfyui.bat启动项。
If a video memory error occurs, please refer to the following code as the comfyui.bat startup item.
@echo off
chcp 65001 >nul
title ComfyUI 12GB VRAM Fixed Start
cd /d "D:\ComfyUI"
"D:your comfyUI\python\python.exe" main.py --auto-launch --preview-method latent2rgb --preview-size 512
pause工作流简介 / Workflow Overview
本工作流采用 ZIT + ZIB 双阶段 T2I：先用 ZIT 快速预览构图与姿势，再由 ZIB 接续完成细节，兼顾速度与画面质量。
This workflow uses dual-stage ZIT + ZIB T2I: ZIT quickly previews the composition and pose, then ZIB continues the generation to complete the details.
放大 / Upscaling
内置 Ultimate SD Upscale 2× 分块放大。主流程会自动复用缓存，无需重复运行 ZIT 与 ZIB。
Includes 2× tiled upscaling with Ultimate SD Upscale. Cached ZIT and ZIB results are reused automatically.
四区连续重绘 / 4-Stage Inpainting
提供 4 个连续、独立、可选的手动重绘区域。每段自动继承上一段结果，并可单独设置遮罩、提示词、LoRA、Seed 和重绘参数；不需要的区域可以直接跳过。
Includes four sequential, independent, and optional manual inpainting stages. Each stage inherits the previous result and supports its own mask, prompt, LoRA, seed, and inpainting settings. Unused stages can be skipped.
仅放大未重绘的原图，Only enlarge the original image without redrawing
针对面部，For face
针对胸部，For the chest
针对错误区域左下角，Target the lower left corner of the error area
新手与显存 / Beginner & VRAM
按蓝色按钮依次运行即可：ZIT 预览 → ZIB 完成 → 2× 放大 → 可选局部重绘。针对 12GB VRAM 优化，四个重绘区默认共享基础模型以减少显存占用。
Simply follow the blue buttons: ZIT Preview → ZIB Final Pass → 2× Upscale → Optional Inpainting. Optimized for 12GB VRAM, with all four inpainting stages sharing the base model to reduce memory usage.
使用提示 / Quick Tip
确认 ZIT 预览后，请不要修改主 Seed、提示词、模型或 LoRA，再继续运行 ZIB。局部重绘时，遮罩、提示词与 LoRA 应对应同一目标。
After confirming the ZIT preview, do not change the main seed, prompt, model, or LoRA before running ZIB. During inpainting, keep the mask, prompt, and LoRA focused on the same target.
Description
全新v3版本
New v3 version
FAQ
Comments (1)
显存不足与 OOM 解决方法
Insufficient video memory and OOM solutions
如果运行工作流时出现以下错误：
Allocation on device This error means you ran out of memory on your GPU. CUDA out of memory请依次尝试以下方法：
不要同时开启自动面部修复和手动遮罩修复。
将 Batch Size 保持为 1。
12GB显存建议优先使用 FP8 检查点。
关闭采样过程预览。
关闭其他占用显存的软件。
降低初始画布尺寸或 FaceDetailer 的 Guide Size。
如果 Qwen 文本编码器提供设备选项，可以将其设置为 cpu，但生成速度会稍微降低。
检查 ComfyUI 启动参数，删除：
--disable-cuda-malloc可以尝试使用以下启动参数：
python main.py --preview-method none --disable-smart-memory--disable-smart-memory 会让 ComfyUI 更积极地把暂时不用的模型卸载到系统内存，从而降低显存峰值，但可能增加系统内存占用并稍微降低模型切换速度。
如果使用整合包或启动器，请在启动器的高级设置或额外启动参数中进行修改。修改后需要完全关闭并重新启动 ComfyUI。
VRAM and CUDA Out-of-Memory Troubleshooting
If the workflow produces one of the following errors:
Allocation on device This error means you ran out of memory on your GPU. CUDA out of memoryTry the following solutions in order:
Do not enable automatic and manual face restoration simultaneously.
Keep Batch Size at 1.
FP8 checkpoints are recommended for GPUs with 12GB of VRAM.
Disable sampler previews.
Close other applications using GPU memory.
Reduce the initial canvas resolution or FaceDetailer Guide Size.
If the Qwen text encoder provides a device option, set it to cpu. This may slightly reduce generation speed.
Check the ComfyUI startup command and remove:
--disable-cuda-mallocTry starting ComfyUI with:
python main.py --preview-method none --disable-smart-memory--disable-smart-memory makes ComfyUI aggressively offload inactive models to system RAM, reducing peak VRAM usage. This may increase system memory usage and slightly slow down model switching.
If you are using a launcher or packaged installation, edit these options in its advanced settings or additional launch arguments. Completely restart ComfyUI after changing the startup parameters.