ComfyUI 多重扩散放大

ComfyUI 多重扩散放大自动化工具。当用户想要放大 AI 生成的图片、处理高清图像、放大小图、提高图片分辨率时使用。支持文生图和图生图两种基础工作流，自动执行三重放大（潜空间放大→区块放大→模型放大）。使用场景包括：(1) 将低分辨率 AI 绘画放大到 4K/8K (2) 增强图片细节和清晰度 (3) Com...

Security Scan

The skill's files, scripts, and runtime instructions are coherent with its stated purpose (generating ComfyUI triple-upscale workflows); it does not request credentials, does not download remote code, and its actions are proportional to the task.

Purpose & Capability

Name/description (ComfyUI upscaler) match the included artifacts: SKILL.md describes generating/importing a ComfyUI workflow and the scripts check for ComfyUI and emit a JSON workflow. Required functionality (model paths, upscaler names, tile sizes) is consistent with an upscaling workflow.

Instruction Scope

SKILL.md instructs the agent/user to run local Python scripts under the skill directory and to load the produced JSON into ComfyUI. The scripts only inspect local filesystem paths (common ComfyUI locations, custom_nodes, models directories), check Python dependencies, and write a JSON workflow. They do not access unrelated system files, network endpoints, or environment variables.

Install Mechanism

No install spec is present. This is instruction-only with two small helper scripts. Nothing is downloaded or executed from arbitrary URLs and nothing is written outside the created output JSON and normal skill directory.

Credentials

The skill requests no environment variables or credentials. The scripts suggest Python packages (torch, numpy, PIL) that are reasonable for local image/model work. Model download recommendations appear in docs but are not automated and reference public model hubs.

Persistence & Privilege

Flags are default (always: false). The skill does not modify other skills or system-wide configs. It reads/writes only within the skill folder and any output directories the user specifies.

Assessment

This skill appears coherent and low-risk for its stated purpose. Before installing or running it: 1) Ensure you have ComfyUI installed in one of the checked locations or pass correct paths; the scripts only detect common install locations. 2) The helper scripts will create the output JSON (and its parent dirs) and check for local Python packages — they will not auto-download models but the docs recommend model sources (HuggingFace/CivitAI); only download models from sources you trust and respect licensing. 3) You will need substantial GPU/VRAM and disk space for large models; adjust tile_size and settings if you get OOMs. 4) No credentials are requested and no network exfiltration is present, but always inspect third-party models before use. If you want extra assurance, review the two included scripts yourself (they are small and readable) or run them in a controlled environment.

ComfyUI 多重扩散放大 Skill

本 Skill 帮助你在 ComfyUI 中快速搭建并运行「三重超清放大」工作流。

工作流原理

三重放大 = 潜空间重采样 → 分区模型放大 → 全局锐化

步骤	节点	作用
1	Latent Scale + K采样器	在潜空间加噪声重采样，增加细节
2	SD Upscale (分区放大)	分块放大避免显存爆炸，局部增强
3	Image Upscale with Model	全局锐化，让模糊变清晰

快速开始

检查环境

cd ~/.openclaw/skills/comfyui-upscaler
python3 scripts/check_env.py

生成工作流

python3 scripts/generate_workflow.py \
  --base_model "your-sd-model.safetensors" \
  --upscale_model "4x-UltraSharp.pth" \
  --tile_size 1024 \
  --output "upscaled_workflow.json"

在 ComfyUI 中加载

1. 打开 ComfyUI
2. 点击 "Load" 加载生成的 upscaled_workflow.json
3. 拖入你的低分辨率图片
4. 点击 "Queue"

脚本说明

generate_workflow.py

生成可导入 ComfyUI 的工作流 JSON 文件。

参数：

• --base_model: 基础模型路径（SD1.5/SDXL/Pony等）
• --upscale_model: 放大模型（4x-UltraSharp 等）
• --tile_size: 分块大小（默认1024，SDXL建议1024以上）
• --scale: 放大倍数（默认2）
• --denoise: 重采样去噪强度（默认0.35）
• --output: 输出JSON路径

check_env.py

检查 ComfyUI 环境是否满足要求：

• ComfyUI 是否安装
• 必要的节点是否安装（LatentScale、SDUpscale等）
• 放大模型是否存在

参考资料

• 工作流详解 - 各节点参数详解
• 模型推荐 - 推荐使用的放大模型
• 常见问题 - 常见问题和解决方案