diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index 2127fcc..5818d7d 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -1,11 +1,21 @@
{
"name": "comfyui-readable-node",
- "owner": { "name": "nomadoor" },
+ "interface": {
+ "displayName": "Readable Node"
+ },
+ "owner": {
+ "name": "nomadoor"
+ },
"plugins": [
{
"name": "comfyui-readable-node",
"source": "./",
- "description": "Rebuilds ComfyUI workflows into a form people can read (Readable Node)."
+ "description": "Rebuilds ComfyUI workflows into a form people can read (Readable Node).",
+ "policy": {
+ "installation": "AVAILABLE",
+ "authentication": "ON_INSTALL"
+ },
+ "category": "Productivity"
}
]
}
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 802bbc8..1fd177a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,17 @@
# Changelog
+## Unreleased (the version number is set when it is published)
+
+- A rail with one target lowers through a Reroute next to the target whenever another column stands between; the canvas no longer wires it straight from the raising Reroute, which its own check called Bad.
+- "A slanted wire one move makes horizontal" no longer counts a move that takes a sampler's column of sources along (a stage's chain on the row below stays there).
+- LATENT_UPSCALE_MODEL counts as a tool wire, not one a node works on.
+- Two stacked nodes aligned only on their left edges are Bad (columns align their right edges); the check names the fix, `put … below` with the right edges aligned. A loader above a sampler is left alone.
+- Rails keep V_GAP from nodes: a rail pushed up in type order stays V_GAP above the node heading the rail below it. New check: a rail closer to a node than V_GAP (the wire from a source to its first Reroute included).
+- Tests: removed the skipped tests and the development notes.
+- English uses American spelling throughout (color, neighbor, center, afterward, backward).
+- README rewritten in English and Japanese, with a Simplified Chinese version (README.zh.md), the intro video, and install steps for the Claude and ChatGPT desktop apps.
+- marketplace.json: the display name, install policy and category used by Codex are back.
+
## 0.0.5 (2026-11-08)
Layout came out worse in 0.0.4: rules lost while untangling sentences, and the canvas telling the agent fixes it could not make. This puts them back and makes the canvas and the rules agree.
@@ -70,10 +82,10 @@ The rules were untangled into one set: sentences that contradicted each other we
- The places the canvas offers for a Note no longer put another node between the Note and the node it explains. A raising Reroute is no longer called unneeded when the rail drawn at the output height would run too close to another rail.
- Export keeps node numbers from one export to the next; a Reroute the canvas made gets a number of its own.
- Fixed: a link from a broadcaster (Anything Everywhere) saved with the type `*` now takes the type of the output it comes from.
-- A Load CLIP whose CLIP reaches no text encoder (one that only feeds an LLM) gets no fixed colour.
+- A Load CLIP whose CLIP reaches no text encoder (one that only feeds an LLM) gets no fixed color.
- A wire may go back one node's width only into Conditioning Zero Out, or from an empty latent into a scheduler.
-- Report page: bypassed nodes are drawn in ComfyUI's bypass colour; the Walkthrough's first page lists the parts, the review and the widened gaps.
-- References: readable-form.md rewritten as eight numbered rules with Why, Bad, Fine and Undecided (the usual form, nodes, wires, flow and position, loaders, value nodes, groups and colour, names); content.md rewritten as two decisions (split; keep, fix, replace or remove); examples.md ordered by those sections; glossary.md defines chain wire, raising and lowering Reroute, gaps, step and more; notes.md: link the Comfy-Org copy of a model, suggest a smaller form of the same model when memory runs short; techniques.md: what becomes of the author's negative text at cfg 1 depends on whether raising cfg is an ordinary choice; replacements.md: `canvas open` replaces Image Comparer, braces in Format Text are doubled.
+- Report page: bypassed nodes are drawn in ComfyUI's bypass color; the Walkthrough's first page lists the parts, the review and the widened gaps.
+- References: readable-form.md rewritten as eight numbered rules with Why, Bad, Fine and Undecided (the usual form, nodes, wires, flow and position, loaders, value nodes, groups and color, names); content.md rewritten as two decisions (split; keep, fix, replace or remove); examples.md ordered by those sections; glossary.md defines chain wire, raising and lowering Reroute, gaps, step and more; notes.md: link the Comfy-Org copy of a model, suggest a smaller form of the same model when memory runs short; techniques.md: what becomes of the author's negative text at cfg 1 depends on whether raising cfg is an ordinary choice; replacements.md: `canvas open` replaces Image Comparer, braces in Format Text are doubled.
## 0.0.2 (2026-10-21)
diff --git a/NOTICE b/NOTICE
index f0549bb..881cb60 100644
--- a/NOTICE
+++ b/NOTICE
@@ -1,13 +1,13 @@
comfyui-readable-node
This skill reads ComfyUI workflows the way the ComfyUI frontend loads them, so its reading code
-reproduces the frontend's behaviour (subgraph expansion, widget value order, legacy group nodes,
-node sizes). That behaviour was checked against prompts produced by the running frontend; no frontend
+reproduces the frontend's behavior (subgraph expansion, widget value order, legacy group nodes,
+node sizes). That behavior was checked against prompts produced by the running frontend; no frontend
source code is included.
skills/comfyui-readable-node/scripts/ue.py reproduces the link rules of the Anything Everywhere custom nodes
(cg-use-everywhere by chrisgoringe, Apache License 2.0). It was written from a description of
-that behaviour; no source code of that project is included.
+that behavior; no source code of that project is included.
skills/comfyui-readable-node/scripts/fonts/PlusJakartaSans-latin.woff2 is Plus Jakarta Sans (Latin subset), embedded in the replay page.
It is licensed under the SIL Open Font License 1.1; see skills/comfyui-readable-node/scripts/fonts/OFL-PlusJakartaSans.txt.
diff --git a/README.ja.md b/README.ja.md
index 4833b41..224973c 100644
--- a/README.ja.md
+++ b/README.ja.md
@@ -1,68 +1,140 @@
-# comfyui-readable-node
+# ComfyUI Readable Node
-[English](README.md)
+[](README.md) [](README.ja.md) [](README.zh.md)
-ComfyUI の workflow を、人が読みやすい形(**Readable Node**)に組み直す、エージェント向けの skill です。
+[](https://comfyui.nomadoor.net/) [](LICENSE)
-1. **ノードを隠さない。** subgraph は展開し、折りたたみは開き、ノードを重ねない。
-2. **線を隠さない。** 無線(Set/Get、Anything Everywhere、線の非表示など)は、線に戻す。
-3. **左から右へ。** 処理の順に左から右へ流す。
-4. **1 task = 1 workflow。** 1 workflow は、使う人のやりたいこと一つ。
-5. **custom node はできるだけ取り除く。** core のノードで組めるものは core で組む。
-6. **適切に色を付ける。** 色は、どの機能に属するかを示す。
+[](https://youtu.be/_Pkw9Mcvkng)
+
+ComfyUI の workflow を、読みやすい形(**Readable Node**)に組み直す skill です。
+
+単にスクリプトでレイアウトを整えるものではありません。
+AI は workflow を一度ばらばらにし、何をしているのかを理解します。その後、タスクがいくつも詰め込まれているものは 1 つずつの workflow に分け、下のルールに従って 0 から組み直します。
+
+元の workflow がやっていることはなるべく保ちつつ、custom node を core のノードに置き換えたり、よりシンプルに組める手段があればそちらで組み直したりして、本質を変えないまま堅牢な workflow にしていきます。
+
+もちろん、0 から AI に workflow を組んでもらうときにも役立ちます。
+
+## Readable Node とは
+
+ここでは「読みやすい」を、何をしているかがスッと頭に入って、安定して動いて、気軽に組み替えられること、と定義しています。
+そのために必要な 6 つのルールが以下です。
+
+1. **ノードを隠さない**
+ subgraph は展開し、折りたたみは開き、ノードを重ねない。
+2. **ワイヤを隠さない**
+ 無線(Set/Get、Anything Everywhere、ワイヤの非表示など)は、ワイヤに戻す。ノードをギチギチに詰めた、いわゆる Bento レイアウトも、ワイヤが隠れるのでやめる。
+3. **左から右へ**
+ 処理の順に左から右へ流す。
+4. **1 つの workflow に、1 つの作業**
+ 1 workflow は、使う人のやりたいこと一つ。
+5. **カスタムノードは最小限**
+ core のノードで組めるものは core で組む。
+6. **適切な色付け**
+ 色は、どの機能に属するかを示す。
## 導入
-下のどれか一つで入れてください。二つ入れると、skill が二重に並びます。
+使っているアプリに合わせて、下のどれかで入れてください。
-### Claude Code
+※ plugin として入れたあとに npx でも入れると、skill が二重に入ってしまうので注意してください。
-Claude Code のセッションの中で、次を実行してください。
+
+Claude デスクトップアプリ
+1. **プラグイン** の画面で **+追加 → マーケットプレイスを追加 → リポジトリから追加** を選ぶ
+2. `nomadoor/comfyui-readable-node` を入力
+3. `comfyui-readable-node` の **+** をクリック
+
+
+
+
+Claude Code(CLI)
+
+ターミナルで次を実行してください。
+
+```bash
+claude plugin install comfyui-readable-node --marketplace nomadoor/comfyui-readable-node
```
-/plugin marketplace add nomadoor/comfyui-readable-node
-/plugin install comfyui-readable-node@comfyui-readable-node
-```
-自動で更新したいときは、`/plugin` の Marketplaces で `comfyui-readable-node` を選び、Enable auto-update を選んでください。
+
+
+
+ChatGPT デスクトップアプリ(Codex)
+
+1. **カスタマイズ → プラグイン** で **追加** を押す
+2. ソースに `https://github.com/nomadoor/comfyui-readable-node.git` を入力。Git ref とスパースパスは空のまま
+3. **個人用** のタブの `comfyui-readable-node` の **+** をクリック
+
+
+
+
+Codex CLI
-### Codex
+ターミナルで次を実行してください。
```bash
codex plugin marketplace add nomadoor/comfyui-readable-node
codex plugin add comfyui-readable-node@comfyui-readable-node
```
-### そのほかのエージェント(Gemini CLI、Cursor、GitHub Copilot など)
+
+
+
+そのほかのエージェント(Gemini CLI、Cursor、GitHub Copilot など)
+
+ターミナルで次を実行してください。Node.js 22.20 以上が必要です。
```bash
npx skills add nomadoor/comfyui-readable-node -g
```
-Node.js 22.20 以上が必要です。
-
-scripts を動かすには Python 3 が必要です(標準ライブラリだけを使います)。
-
-## モデル
-
-長い作業を自分で確かめながら進める skill なので、強いモデルを使ってください。
+
-- いちばん良いのは Claude Opus 5.5 です。
-- GPT 6.1 でも使えますが、配置の崩れが残ることがあります。
-- 少なくとも Sonnet 以上のモデルでないと、うまくいかないかもしれません。
+skill の scripts を動かすには、Python 3 が必要です(標準ライブラリだけを使います)。
## 使い方
-1. ComfyUI を起動しておいてください。skill は、動いている ComfyUI からノードの定義を読みます。
+1. ComfyUI を起動しておいてください。skill は、動いている ComfyUI からノードの情報を読みます。
2. エージェントに頼んでください。たとえば:
- 「Readable Node で `my_workflow.json` を読みやすくして。ComfyUI は http://127.0.0.1:8188 で動いています。」
- 「Readable Node で、Flux の Text to Image の workflow を作って。」
確実に呼びたいときは `/comfyui-readable-node` と打ってください(plugin で入れたときは `/comfyui-readable-node:comfyui-readable-node`)。
-3. 組み直した workflow(`.json`)と報告が出てきます。workflow を ComfyUI で開いて確かめてください。
+3. 出来上がった workflow を ComfyUI で開いて、確かめてください。
+
+時間は、ノード 25 個ほどの workflow で 10 分前後です。大きな workflow だと 20〜40 分かかることもあります。
+
+## できあがるもの
+
+
+
+- **組み直した workflow**: 元のファイル名に `_readable` を付けた `.json` です。元のファイルは上書きしません。
+- **report.html**: ブラウザで開くと、AI が組み直していく様子と、出来上がった workflow の解説が見られます。
+ - Original: 元の workflow と見比べられます。
+ - Rebuilt: AI がノードを一手ずつ動かしていく様子を、タイムラインで再生できます。
+ - Walkthrough: 出来上がった workflow を、部分ごとに解説します。
+
+## おすすめの AI モデル
+
+workflow を読み解いて綺麗に組み直すのは、見た目に反してかなり重い作業なので、できるだけ強いモデルを使ってください。
+
+- 開発に使っていて、いちばんおすすめなのは Claude Opus 5.5 です。
+- GPT 6.1 / 6.0 Astra でも使えますが、配置の崩れが残ることがあります。
+
+## skill が読みに行くもの
+
+- お使いの PC で動いている ComfyUI。どんなノードが入っているかを確かめます。
+- [Comfy with ComfyUI](https://comfyui.nomadoor.net/) の workflow。組み方の参考にすることがあります。
+
+## うまくいかなかったとき
+
+組み直した workflow がおかしい、読みにくいときは、[issue](https://github.com/nomadoor/comfyui-readable-node/issues/new/choose) で教えてください。workflow には、プロンプトやフォルダのパスが入っていることがあります。公開したくないものは消してから添付してください。
+
+## 作者
-報告のページ `report.html` も出ます。元の workflow、組み直しの手順、出来上がった workflow の説明(Walkthrough)を見られます。
+[nomadoor](https://x.com/noma_door)。ComfyUI の基本を解説するサイト [Comfy with ComfyUI](https://comfyui.nomadoor.net/) を運営しています。この skill は、サイトに載せてきた 400 本を超える workflow を土台に作りました。
## ライセンス
-MIT。[LICENSE](LICENSE) と [NOTICE](NOTICE) を見てください。
+MIT
diff --git a/README.md b/README.md
index 0d56ab8..b011aec 100644
--- a/README.md
+++ b/README.md
@@ -1,68 +1,140 @@
-# comfyui-readable-node
+# ComfyUI Readable Node
-[日本語](README.ja.md)
+[](README.md) [](README.ja.md) [](README.zh.md)
-An agent skill that rebuilds ComfyUI workflows into a form people can read: **Readable Node**.
+[](https://comfyui.nomadoor.net/) [](LICENSE)
-1. **Do not hide nodes.** Expand subgraphs, open collapsed nodes, and do not overlap nodes.
-2. **Do not hide wires.** Turn wireless links (Set/Get, Anything Everywhere, hidden links) back into wires.
-3. **Left to right.** Flow from left to right in processing order.
-4. **1 task = 1 workflow.** One workflow is one thing the user wants to do.
-5. **Remove custom nodes as far as possible.** What can be built with core nodes is built with core.
-6. **Colour appropriately.** Colour shows which function a node belongs to.
+[](https://youtu.be/_Pkw9Mcvkng)
+
+A skill that rebuilds ComfyUI workflows so they are easy to read (**Readable Node**).
+
+It doesn't just tidy up the layout with a script.
+The AI takes the workflow apart and works out what it does. A workflow that packs several tasks into a single file is split into separate workflows, one per task, and each is rebuilt from scratch following the rules below.
+
+It preserves what the original workflow does as much as possible, while replacing custom nodes with core nodes and rebuilding parts more simply where it can, making the workflow more robust without changing its purpose.
+
+It also works when you ask an AI to build a new workflow from scratch.
+
+## What Readable Node is
+
+Here, "easy to read" means you can tell at a glance what the workflow does, it runs reliably, and it is easy to rework.
+It follows six rules:
+
+1. **Don't hide nodes**
+ Expand subgraphs, open collapsed nodes, and don't overlap nodes.
+2. **Don't hide wires**
+ Turn wireless links (Set/Get, Anything Everywhere, hidden wires, etc.) back into wires. Avoid so-called Bento layouts, which pack nodes tightly together, since they hide the wires too.
+3. **Left to right**
+ Flow from left to right in processing order.
+4. **One workflow, one task**
+ A workflow does one thing the user wants to do.
+5. **Keep custom nodes to a minimum**
+ What can be built with core nodes is built with core.
+6. **Color appropriately**
+ Color shows which function a node belongs to.
## Install
-Use one of the ways below. Installing two lists the skill twice.
+Choose the method for the app you use.
-### Claude Code
+Note: don't install it both as a plugin and with npx, or you'll end up with two copies of the skill.
-In a Claude Code session, run:
+
+Claude desktop app
+1. In **Settings → Customize → Plugins**, choose **+ Add → Add marketplace → Add from a repository**
+2. Enter `nomadoor/comfyui-readable-node`
+3. Click **+** on `comfyui-readable-node`
+
+
+
+
+Claude Code (CLI)
+
+In a terminal, run:
+
+```bash
+claude plugin install comfyui-readable-node --marketplace nomadoor/comfyui-readable-node
```
-/plugin marketplace add nomadoor/comfyui-readable-node
-/plugin install comfyui-readable-node@comfyui-readable-node
-```
-To get updates automatically, open `/plugin`, go to Marketplaces, select `comfyui-readable-node` and choose Enable auto-update.
+
+
+
+ChatGPT desktop app (Codex)
+
+1. Open **Plugins** and click **Add** to open **Add plugin marketplace**
+2. Enter `https://github.com/nomadoor/comfyui-readable-node.git` in **Source**. Leave **Git ref** and **Sparse paths** empty
+3. On the **Personal** tab, click **+** on `comfyui-readable-node`
+
+
+
+
+Codex CLI
-### Codex
+In a terminal, run:
```bash
codex plugin marketplace add nomadoor/comfyui-readable-node
codex plugin add comfyui-readable-node@comfyui-readable-node
```
-### Other agents (Gemini CLI, Cursor, GitHub Copilot and others)
+
+
+
+Other agents (Gemini CLI, Cursor, GitHub Copilot, etc.)
+
+In a terminal, run the following. Requires Node.js 22.20 or later.
```bash
npx skills add nomadoor/comfyui-readable-node -g
```
-This needs Node.js 22.20 or later.
-
-The scripts need Python 3 (standard library only).
-
-## Models
-
-The skill runs a long job and checks its own work as it goes, so it needs a strong model.
+
-- Claude Opus 5.5 works best.
-- GPT 6.1 works, but may leave some layout problems.
-- With anything below Sonnet, it may not work well.
+The skill's scripts need Python 3 (standard library only).
-## Use
+## Usage
-1. Start ComfyUI. The skill reads the node definitions from the running ComfyUI.
+1. Start ComfyUI first. The skill reads the node information from the running ComfyUI.
2. Ask your agent, for example:
- "Make `my_workflow.json` readable with Readable Node. ComfyUI is running at http://127.0.0.1:8188."
- "Build a Text to Image workflow for Flux with Readable Node."
- To call it directly: `/comfyui-readable-node` (`/comfyui-readable-node:comfyui-readable-node` when installed as a plugin).
-3. You get the rebuilt workflow (`.json`) and a report. Open the workflow in ComfyUI to check it.
+ To make sure the skill is invoked, type `/comfyui-readable-node` (`/comfyui-readable-node:comfyui-readable-node` when installed as a plugin).
+3. Open the finished workflow in ComfyUI and check it.
+
+A workflow of about 25 nodes takes around 10 minutes. A large workflow can take 20 to 40 minutes.
+
+## What you get
+
+
+
+- **The rebuilt workflow**: a `.json` file with `_readable` appended to the original name. The original file is not overwritten.
+- **report.html**: open it in a browser to watch the AI rebuild the workflow and read an explanation of the result.
+ - Original: compare with the original workflow.
+ - Rebuilt: replay the AI's moves on a timeline, one step at a time.
+ - Walkthrough: the finished workflow explained part by part.
+
+## Recommended AI models
+
+Reading a workflow and rebuilding it cleanly is far more demanding than it looks, so use the strongest model you can.
+
+- Claude Opus 5.5: the skill was developed with it, and it's my top recommendation.
+- GPT 6.1 / 6.0 Astra also work, but the layout may still have a few issues.
+
+## What the skill reads
+
+- Your local ComfyUI, to see which nodes are installed.
+- Workflows on [Comfy with ComfyUI](https://comfyui.nomadoor.net/), which it may use as a reference for how to build workflows.
+
+## Reporting problems
+
+If a rebuilt workflow is wrong or hard to read, let me know in an [issue](https://github.com/nomadoor/comfyui-readable-node/issues/new/choose). A workflow may contain your prompts and folder paths, so remove anything you don't want to share publicly before attaching it.
+
+## Author
-You also get a report page, `report.html`: the original workflow, the rebuild step by step, and a walkthrough of the finished workflow.
+[nomadoor](https://x.com/noma_door). I run [Comfy with ComfyUI](https://comfyui.nomadoor.net/), a site that explains the basics of ComfyUI. This skill is built on more than 400 workflows published there.
-## Licence
+## License
-MIT. See [LICENSE](LICENSE) and [NOTICE](NOTICE).
+MIT
diff --git a/README.zh.md b/README.zh.md
new file mode 100644
index 0000000..4b77e3e
--- /dev/null
+++ b/README.zh.md
@@ -0,0 +1,140 @@
+# ComfyUI Readable Node
+
+[](README.md) [](README.ja.md) [](README.zh.md)
+
+[](https://comfyui.nomadoor.net/) [](LICENSE)
+
+[](https://youtu.be/_Pkw9Mcvkng)
+
+一个把 ComfyUI workflow 重新组织成易读形式(**Readable Node**)的 skill。
+
+它不只是用脚本整理布局。
+AI 会先拆解 workflow,弄清楚它在做什么。如果一个 workflow 里塞了好几项任务,就按任务拆成一个个 workflow,再按照下面的规则从零重新搭建。
+
+它会尽量保留原 workflow 所做的事情,同时把自定义节点(custom node)换成原生节点(core 节点),或在有更简单的搭建方式时改用更简单的方式,在不改变本质的前提下,让 workflow 更加稳健。
+
+当然,让 AI 从零搭建 workflow 时,同样可以使用这个 skill。
+
+## 什么是 Readable Node
+
+这里把"易读"定义为:一眼就能明白它在做什么,运行稳定,并且便于修改和重新组合。
+为此需要以下 6 条规则。
+
+1. **不隐藏节点**
+ 展开 subgraph,展开被折叠的节点,节点之间不重叠。
+2. **不隐藏连线**
+ 把不走连线的隐式连接(Set/Get、Anything Everywhere、隐藏的连线等)恢复成连线。把节点挤得满满当当的所谓 Bento 布局,也会把连线遮住,同样不要用。
+3. **从左到右**
+ 按处理顺序从左往右排列。
+4. **一个 workflow,一项任务**
+ 一个 workflow 只做用户想做的一件事。
+5. **自定义节点尽量少**
+ 能用原生节点搭的,就用原生节点搭。
+6. **恰当地上色**
+ 颜色表示节点属于哪项功能。
+
+## 安装
+
+请根据你使用的应用,选择下面的一种方式安装。
+
+注意:以 plugin 方式安装后,如果再用 npx 安装,skill 会被重复安装。
+
+
+Claude 桌面应用
+
+1. 在 **Settings → Customize → Plugins** 中,选择 **+ Add → Add marketplace → Add from a repository**
+2. 输入 `nomadoor/comfyui-readable-node`
+3. 点击 `comfyui-readable-node` 旁边的 **+**
+
+
+
+
+Claude Code(CLI)
+
+在终端中运行:
+
+```bash
+claude plugin install comfyui-readable-node --marketplace nomadoor/comfyui-readable-node
+```
+
+
+
+
+ChatGPT 桌面应用(Codex)
+
+1. 打开 **Plugins**,点击 **Add**,选择 **Add plugin marketplace**
+2. 在 **Source** 中输入 `https://github.com/nomadoor/comfyui-readable-node.git`。**Git ref** 和 **Sparse paths** 留空
+3. 在 **Personal** 标签页中,点击 `comfyui-readable-node` 旁边的 **+**
+
+
+
+
+Codex CLI
+
+在终端中运行:
+
+```bash
+codex plugin marketplace add nomadoor/comfyui-readable-node
+codex plugin add comfyui-readable-node@comfyui-readable-node
+```
+
+
+
+
+其他 agent(Gemini CLI、Cursor、GitHub Copilot 等)
+
+在终端中运行以下命令。需要 Node.js 22.20 或更高版本。
+
+```bash
+npx skills add nomadoor/comfyui-readable-node -g
+```
+
+
+
+skill 中的脚本需要 Python 3(只使用标准库)。
+
+## 使用方法
+
+1. 请先启动 ComfyUI。skill 会从正在运行的 ComfyUI 读取节点信息。
+2. 向 agent 提出请求,例如:
+ - "用 Readable Node 把 `my_workflow.json` 整理得易读一些。ComfyUI 运行在 http://127.0.0.1:8188。"
+ - "用 Readable Node 搭一个 Flux 的 Text to Image workflow。"
+
+ 如果想确保调用这个 skill,请输入 `/comfyui-readable-node`(如果以 plugin 方式安装,则输入 `/comfyui-readable-node:comfyui-readable-node`)。
+3. 在 ComfyUI 中打开完成的 workflow,检查一下。
+
+25 个节点左右的 workflow 大约需要 10 分钟。较大的 workflow 可能需要 20 到 40 分钟。
+
+## 输出内容
+
+
+
+- **重新搭建的 workflow**:文件名为原文件名加 `_readable` 的 `.json` 文件,不会覆盖原文件。
+- **report.html**:在浏览器中打开,可以看到 AI 重新搭建的过程,以及对完成的 workflow 的讲解。
+ - Original:与原 workflow 对照。
+ - Rebuilt:在时间轴上回放 AI 一步一步移动节点的过程。
+ - Walkthrough:分部分讲解完成的 workflow。
+
+## 推荐的 AI 模型
+
+读懂 workflow 并整齐地重新搭建,远比看起来繁重,请尽量使用能力最强的模型。
+
+- 开发时使用的是 Claude Opus 5.5,也最推荐使用它。
+- GPT 6.1 / 6.0 Astra 也可以使用,但可能会留下一些布局上的问题。
+
+## skill 会读取的内容
+
+- 你电脑上正在运行的 ComfyUI,用来确认装了哪些节点。
+- [Comfy with ComfyUI](https://comfyui.nomadoor.net/) 上的 workflow,有时会用作搭建方法的参考。
+
+## 遇到问题时
+
+如果重新搭建的 workflow 有问题或不好读,请通过 [issue](https://github.com/nomadoor/comfyui-readable-node/issues/new/choose) 告诉我。issue 请用英文或日文撰写。workflow 中可能包含提示词和文件夹路径,附上之前请删除不想公开的内容。
+
+## 作者
+
+[nomadoor](https://x.com/noma_door)。运营着一个讲解 ComfyUI 基础知识的网站 [Comfy with ComfyUI](https://comfyui.nomadoor.net/)。这个 skill 以该网站上发布的 400 多个 workflow 为基础制作。
+
+## 许可证
+
+MIT
diff --git a/skills/comfyui-readable-node/SKILL.md b/skills/comfyui-readable-node/SKILL.md
index c45c9eb..9b830a1 100644
--- a/skills/comfyui-readable-node/SKILL.md
+++ b/skills/comfyui-readable-node/SKILL.md
@@ -1,6 +1,6 @@
---
name: comfyui-readable-node
-description: Builds a ComfyUI workflow (JSON) into a form people can read easily (Readable Node). Use it when the user mentions Readable Node, asks to make a ComfyUI workflow readable or tidy, takes an existing workflow apart to rebuild it, or builds a new workflow. Do not hide nodes, do not hide wires (turn wireless links back into wires), left to right, 1 task per workflow, fewer custom nodes, appropriate colouring.
+description: Builds a ComfyUI workflow (JSON) into a form people can read easily (Readable Node). Use it when the user mentions Readable Node, asks to make a ComfyUI workflow readable or tidy, takes an existing workflow apart to rebuild it, or builds a new workflow. Do not hide nodes, do not hide wires (turn wireless links back into wires), left to right, 1 task per workflow, fewer custom nodes, appropriate coloring.
license: MIT
metadata:
version: "0.0.5"
@@ -20,10 +20,10 @@ The items are ordered by how much they matter when a person reads a workflow. Wh
1. **Do not hide nodes.** Expand subgraphs, open collapsed nodes, and do not overlap nodes.
2. **Do not hide wires.** Turn wireless links (Set/Get, Anything Everywhere, hidden links, and the like) back into wires.
-3. **Left to right.** Flow from left to right in processing order. Do not stretch too far towards the lower right.
+3. **Left to right.** Flow from left to right in processing order. Do not stretch too far toward the lower right.
4. **1 task = 1 workflow.** One workflow is one thing the user wants to do.
5. **Remove custom nodes as far as possible.** What can be built with core nodes is built with core.
-6. **Colour appropriately.** Colour shows which function a node belongs to.
+6. **Color appropriately.** Color shows which function a node belongs to.
There are three goals.
- The reader can follow the processing along a single line (can trace the details).
@@ -42,9 +42,9 @@ The rules in the references serve these principles and goals; how to read a rule
## Division of work
-**What you decide (judgement)**: understanding, how to split, what to keep, remove or replace, rebuilding custom nodes, what goes where, what to align, rails, groups, colours for added functions, Notes.
+**What you decide (judgment)**: understanding, how to split, what to keep, remove or replace, rebuilding custom nodes, what goes where, what to align, rails, groups, colors for added functions, Notes.
-**What the canvas does (things with exactly one answer)**: node sizes, calculating positions, the default colours of loaders, LoRAs and Notes, group padding, checks (check), export. When reading the original (`canvas open`), it fixes what has exactly one way of fixing, such as turning wireless links back into wires, and applying the official node-replacement list and replacements that give the same calculation. It returns everything it did.
+**What the canvas does (things with exactly one answer)**: node sizes, calculating positions, the default colors of loaders, LoRAs and Notes, group padding, checks (check), export. When reading the original (`canvas open`), it fixes what has exactly one way of fixing, such as turning wireless links back into wires, and applying the official node-replacement list and replacements that give the same calculation. It returns everything it did.
**How to use the canvas**
- Do not write coordinate numbers. Point at things: "put it to the right of A", "make the wire horizontal", "align the top edges". The canvas does the calculation. Only what you point at moves.
@@ -60,7 +60,7 @@ The rules in the references serve these principles and goals; how to read a rule
## Steps
-`S` is the path of scripts/. Choose one working folder `W` inside the folder where the results go (the one the user named; otherwise next to the original), and type every canvas move as `python3 $S/canvas.py --dir $W `. Below, this is written `canvas`. The list of moves, their syntax and the detailed behaviour (undo, placements calculated only once, block, and so on) are at the top of canvas.py (`python3 $S/canvas.py` prints them): read it once before step 4.
+`S` is the path of scripts/. Choose one working folder `W` inside the folder where the results go (the one the user named; otherwise next to the original), and type every canvas move as `python3 $S/canvas.py --dir $W `. Below, this is written `canvas`. The list of moves, their syntax and the detailed behavior (undo, placements calculated only once, block, and so on) are at the top of canvas.py (`python3 $S/canvas.py` prints them): read it once before step 4.
- Put every file you make while working inside `W`: plan files (build.txt for building, layout.txt for arranging: "the plan" is layout.txt), Note files, workflows or object_info you fetch to read, notes. Only the exported workflows and the report page go outside `W`. `canvas finish` deletes the files of `W` it knows (what it wrote, the plan and Note files, understanding.md) and names the others it left: delete only those, one by one. Do not delete `W` with a recursive delete.
Moves marked "(when there is an original)" are only for rebuilding a given workflow. When building a new one, skip them.
@@ -100,7 +100,7 @@ When the user asks for it fast, skip steps 9 and 10 (the review and the report p
2. Read references/readable-form.md, references/glossary.md and the "Layout" part of references/examples.md, again here if you read them before. Then paste the lines of `canvas core ` at the top of layout.txt. They are a draft of the core's block: adjust them to this workflow (a part the usual form lacks is named in a comment, to be placed by relations).
3. For each grouping of the understanding document, write `block …` and the arrangement inside it.
4. Run `canvas between layout.txt` to see the wires between blocks and the outputs handed out to several blocks. Whether a value handed out to several stages needs a `rail`: readable-form.md, "Wires".
- 5. Write the arrangement between blocks: each block against the block it is wired to (`put left|right wire`). Groups, colours and the titles of value nodes and of nodes loading an image go in the same file (readable-form.md, "Groups and colour", "Names"); the canvas reads every group line before it places anything, so where in the file a group line stands does not change the layout.
+ 5. Write the arrangement between blocks: each block against the block it is wired to (`put left|right wire`). Groups, colors and the titles of value nodes and of nodes loading an image go in the same file (readable-form.md, "Groups and color", "Names"); the canvas reads every group line before it places anything, so where in the file a group line stands does not change the layout.
6. Run `canvas layout layout.txt`.
How to place:
@@ -145,7 +145,7 @@ When the user asks for it fast, skip steps 9 and 10 (the review and the report p
- references/content.md: the rules about content (splitting, replacing, keeping, removing).
- references/techniques.md: building techniques (hard-to-read forms, and forms rebuilt to be readable).
- references/replacements.md: the replacement dictionary (custom nodes and deprecated nodes, things that may be removed, things to keep) and how to find the source.
-- references/readable-form.md: the layout rules (nodes, wires, flow and position, groups and colour, names).
+- references/readable-form.md: the layout rules (nodes, wires, flow and position, groups and color, names).
- references/notes.md: how to write Notes and the Walkthrough, and where to place Notes.
- references/examples.md: examples (how the rules applied in real workflows).
- references/review.md: the review after export (what you do, and the brief for the agent that reviews).
diff --git a/skills/comfyui-readable-node/references/content.md b/skills/comfyui-readable-node/references/content.md
index 29ec12a..9653c7b 100644
--- a/skills/comfyui-readable-node/references/content.md
+++ b/skills/comfyui-readable-node/references/content.md
@@ -10,7 +10,7 @@ Two decisions for each part of the original: which workflow it goes into (Split)
## Split
**One workflow is one thing the user wants to do.** Many workflows are fine: the aim is the principles of SKILL.md, not fewer workflows. Why: 1 task = 1 workflow.
-- One workflow: a generation whose stages pass conditioning or a latent on (base → refiner → Hires.fix; t2i → a redraw at a larger size with the same model). Effects on the finished image (film grain, colour matching, frame interpolation, a 1x deblur) stay in it.
+- One workflow: a generation whose stages pass conditioning or a latent on (base → refiner → Hires.fix; t2i → a redraw at a larger size with the same model). Effects on the finished image (film grain, color matching, frame interpolation, a 1x deblur) stay in it.
- Split: what changes what can be done (t2i, i2i and edit; T2V and I2V); what changes the sampling setup (steps, sigmas, cfg, the sampler): models each with their own settings, a speed-up that changes steps or cfg. Tasks that are not connected (outputs sharing no upstream).
- A bypassed branch in the one workflow: a speed-up that only may make it faster, an addition that only may make it a little better (CFGZeroStar, a guidance adjustment).
- An added function the user may want or not (a ControlNet, a guide image) is split: with it and without it. A few nodes that must be switched together can stay in the one workflow as a group switched with bypass.
diff --git a/skills/comfyui-readable-node/references/examples.md b/skills/comfyui-readable-node/references/examples.md
index 07ade33..50f64aa 100644
--- a/skills/comfyui-readable-node/references/examples.md
+++ b/skills/comfyui-readable-node/references/examples.md
@@ -1,6 +1,6 @@
# Examples
-How the rules applied in real workflows: "Layout" under the section headings of readable-form.md, then "Content" and "Notes" by topic. "Before" is a form an agent built, "After" the corrected form, "Layout" a layout arranged by hand. In the same situation, lean towards this form.
+How the rules applied in real workflows: "Layout" under the section headings of readable-form.md, then "Content" and "Notes" by topic. "Before" is a form an agent built, "After" the corrected form, "Layout" a layout arranged by hand. In the same situation, lean toward this form.
An example marked **Exception ()** is a correct counter-example: there that rule gives way. Every other example is a rule applied as written.
## Layout
@@ -134,7 +134,7 @@ Layout: the image Switches of all stages, the loader and the saver at one height
**Spacing**
```
Before: 150px between Load LoRA and KSampler
-After: no wider space to a wired neighbour without a reason
+After: no wider space to a wired neighbor without a reason
```
### 5. Loaders
@@ -166,11 +166,11 @@ Layout: a single stage: Load Checkpoint just left of the text encoders (its CLIP
- The cfg 1.0 of a distilled model is not changed, so it stays in each KSampler; no cfg value node.
- Sizes typed by hand once are not handed out; a size the author measures from an image comes from Get Image Size.
-### 7. Groups and colour
+### 7. Groups and color
- A two-stage LTX workflow has three groups (stage 1, stage 2, decode). A t2i with IPAdapter has no group.
-- A Detail pass whose every node was coloured green inside its group: the group already guides the eye.
-- Colour split finely (SCAIL 2): the core green; taking the person in the input video pale_blue; taking the person in the reference image cyan; CLIP Vision brown.
+- A Detail pass whose every node was colored green inside its group: the group already guides the eye.
+- Color split finely (SCAIL 2): the core green; taking the person in the input video pale_blue; taking the person in the reference image cyan; CLIP Vision brown.
- In that two-stage LTX workflow, when the Stage 1 result is decoded too, that decode belongs to the stage 1 group; only the decode after upscaling is in the decode group.
### 8. Names
@@ -187,7 +187,7 @@ After: ComfyUI's display name
### Split
```
-One: base → refiner → Hires.fix (conditioning passes on), with the film grain and colour correction on the finished image
+One: base → refiner → Hires.fix (conditioning passes on), with the film grain and color correction on the finished image
Split: nine models with their own sampler settings × t2i and i2i = eighteen workflows
Split: a prompt written by hand or by an MLLM from an image, one side bypassed in the author's workflow: two text-to-image workflows, one with the MLLM and one without
Before: an Anima LLLite guide, its nodes in no group, switched on and off with bypass in one workflow
diff --git a/skills/comfyui-readable-node/references/glossary.md b/skills/comfyui-readable-node/references/glossary.md
index 44ddc69..043e7b8 100644
--- a/skills/comfyui-readable-node/references/glossary.md
+++ b/skills/comfyui-readable-node/references/glossary.md
@@ -12,10 +12,10 @@ The meaning of the words used in the skill's documents. Each is defined here onc
## Kinds of nodes
-- **sampler**: KSampler, KSampler (Advanced), SamplerCustom, SamplerCustomAdvanced, and custom nodes that add processing these core nodes cannot realise.
+- **sampler**: KSampler, KSampler (Advanced), SamplerCustom, SamplerCustomAdvanced, and custom nodes that add processing these core nodes cannot realize.
- **loader**: a node that reads a model file (Load Checkpoint, Load Diffusion Model, Load CLIP, Load VAE, Load LoRA, and so on).
- **value node**: a node that only outputs one value (Int, Float, String, and so on).
-- **core node**: a node that comes with ComfyUI from the start. A **custom node** is a node installed afterwards.
+- **core node**: a node that comes with ComfyUI from the start. A **custom node** is a node installed afterward.
- **deprecated node**: a core node that has been superseded by a newer one ((DEPRECATED) is in its name).
- **source**: an upstream node that connects into an input of a node.
- **chain**: nodes connected in a line by wires.
@@ -35,7 +35,7 @@ The meaning of the words used in the skill's documents. Each is defined here onc
- **row**: a horizontal band of nodes lined up side by side. Rows are separated top to bottom by role, like the model row, the conditioning row, the image row.
- **column**: a vertical band of nodes stacked up and down. The column of sources to the left of a sampler, and so on.
- **eye line (band)**: the height every stage's sampler (KSampler, SamplerCustomAdvanced) stands at; a band with width through which the main flow passes.
-- **main flow**: the flow from the sampler towards the final output (sampler → decode → … → save).
+- **main flow**: the flow from the sampler toward the final output (sampler → decode → … → save).
- **output node**: a node that takes a result and passes nothing on (Save, Preview, Compare).
- **rail**: a value, or several side by side, carried far or to several places with Reroutes (`rail`).
- **Reroute**: a small node that relays a wire. Used for rails.
diff --git a/skills/comfyui-readable-node/references/notes.md b/skills/comfyui-readable-node/references/notes.md
index fbf188c..68dc408 100644
--- a/skills/comfyui-readable-node/references/notes.md
+++ b/skills/comfyui-readable-node/references/notes.md
@@ -19,7 +19,7 @@ The text of the Notes on part of the processing, and the one or two sentences at
**At the very top, what this workflow does, in one or two sentences: what goes in and what comes out. Below it, the list of models.**
- The first heading says that in the user's language. The models heading is always `## models`, in every language (the canvas finds the model Note by it).
-- It has no title: its place at the upper left and its colour say what it is, and its headings divide it.
+- It has no title: its place at the upper left and its color say what it is, and its headings divide it.
- The list of models is in exactly this form (the canvas `models` gives a draft):
````markdown
@@ -63,8 +63,8 @@ Generates an image from a text prompt with Qwen Image 2.1.
- One Note per place the user looks at (one node, or a cluster looked at together). Several topics there become several headings in that one Note. One thing is said once in the whole workflow: the same point (larger is slower) is not written again in the Note of another node.
- Each topic has a `###` heading that says in one plain phrase what the topic wants to say (a warning calls for attention). Under it, one or two phrases. Indent at most one level. A polite register.
- Point at nodes by their names; to have a value changed, write the node name and the widget name ("set sage_attention on Patch Sage Attention KJ to disabled").
-- Text to be copied exactly is in quotation marks, not inline code (`), which breaks in the legacy renderer and on coloured Notes.
-- Bad: pointing at nodes by their colour (a bypassed node loses its colour).
+- Text to be copied exactly is in quotation marks, not inline code (`), which breaks in the legacy renderer and on colored Notes.
+- Bad: pointing at nodes by their color (a bypassed node loses its color).
## Where to place
diff --git a/skills/comfyui-readable-node/references/readable-form.md b/skills/comfyui-readable-node/references/readable-form.md
index 7e6986c..a731d59 100644
--- a/skills/comfyui-readable-node/references/readable-form.md
+++ b/skills/comfyui-readable-node/references/readable-form.md
@@ -1,6 +1,6 @@
# The Readable Node form
-What you decide when arranging a workflow, and what a finished workflow is checked against. The canvas does the rest (node sizes and widths, the fixed colours, rail heights, Reroute positions); its moves and checks are at the top of canvas.py.
+What you decide when arranging a workflow, and what a finished workflow is checked against. The canvas does the rest (node sizes and widths, the fixed colors, rail heights, Reroute positions); its moves and checks are at the top of canvas.py.
How to read a rule: the bold line says what to do. **Why** names the principle or goal of SKILL.md it serves; when a case is not covered, decide by it. **Bad** is the form that breaks the rule, as seen in a finished workflow. **Fine** is a form that looks like Bad but is not; a rule without Fine has no exception. Cases are in examples.md, "Layout", under the same section headings.
@@ -32,7 +32,7 @@ How to read a rule: the bold line says what to do. **Why** names the principle o
- Fine: back by at most the target's width plus one column gap, into Conditioning Zero Out or from an empty latent into a scheduler.
**A wire is drawn directly. Reroutes are used only when, without them, one could not see where the wire goes (a value carried far or to several places: a rail). When a direct wire would be hidden behind a node, move its source or its target so it shows, before bending it with Reroutes.** Why: Reroutes cut one wire into pieces, and the reader must trace them to see where a value comes from.
-- A rail is one value, or several side by side, carried with Reroutes. Write a `rail` line; it runs at the height the line writes (`above `, `below `, `level .