| --- |
| license: apache-2.0 |
| tags: |
| - comfyui |
| - comfyui-node |
| - custom-node |
| - controlnet |
| - openpose |
| - pose-editor |
| --- |
| |
| # comfyui_Bonso_Tools |
|
|
|  |
|
|
| Custom nodes for [ComfyUI](https://github.com/comfyanonymous/ComfyUI). |
|
|
| | Node | What it does | |
| | --- | --- | |
| | **BonsoNet** | Five ControlNet preprocessors (Canny, Depth, OpenPose, Scribble, Anime Line Art) with a colour-coded banner and a resolution widget that hits the exact megapixel target. | |
| | **Bonso Pose** | Manual OpenPose editor: place a mannequin over your reference image, drag the bones, export a ControlNet-ready PNG + `.json`, send it to any node. | |
| | **Mask Crop Bonso** | Square mask crop that repositions instead of clamping at the image edges. | |
| | **Easy Model Bonso** | Model inventory manager: scan the active workflow, download what is missing, save/load the inventory as JSON. | |
| | **Bonso_FileName** | Builds a folder path out of up to **6** text boxes joined with `/`. Empty boxes are skipped; if all are empty it returns `Bonso`. | |
| |
| ## Nodes |
| |
| ### BonsoNet |
| Preprocessor for ControlNet with 5 exclusive modes and a color-coded banner on |
| the canvas: Canny, Depth (DepthAnything V1), OpenPose (DWPose), Scribble (XDoG) |
| and Anime Line Art. The `Megapixel` widget (FLOAT) rescales the image |
| proportionally to the exact megapixel target, so the returned map matches the |
| requested resolution. Only `image` and `Megapixel` accept connections. |
| |
| Requires |
| [`comfyui_controlnet_aux`](https://github.com/Fannovel16/comfyui_controlnet_aux) |
| for the best quality detectors; falls back to native OpenCV/PIL processing for |
| any detector that is unavailable. |
| |
| ### Bonso Pose ⛹🏻♂️ |
| |
| Manual OpenPose editor. No detection models — you place a T-pose mannequin and |
| drag the bones over your reference image, tracing the limbs by eye. |
| |
| - **Visual box node**: the reference image is shown inside the node, on a black |
| background, at whatever **opacity** you set — live, as you drag the slider. |
| The preview keeps the reference's aspect ratio and starts **empty — black, with |
| no skeleton** — so you place the mannequin only when you want it. |
| - **Buttons**: `Cargar imagen` (load from file), `Editar pose` (open the drag |
| editor), `Esqueleto` (drop the basic mannequin into the preview, over the |
| black background or over the image), `Eliminar esqueleto` (clear it, so you |
| can place another from scratch) and `ENVIAR A`. |
| - **`Ancho de hueso`** is the bone thickness in pixels at 1 megapixel. It is drawn |
| exactly like that in the preview, in the editor and in the exported pose. You |
| can change it from the node or from inside the editor — both stay in sync. |
| - **Connect an image node** to the `image` input and the box switches to it on |
| its own, no reloading needed. If you change the picture in the `Load Image` |
| above while the workflow is idle, the box follows **instantly**: the node polls |
| its input every 400 ms (the trick `Outpaint Pixaroma` uses) because there is no |
| event that announces a widget changed. |
| - **The window adapts to the picture**: the reference box has a **fixed size in |
| pixels**, derived only from its aspect ratio, and the node resizes itself to |
| fit it plus the buttons. Reloading the tab therefore looks exactly the same as |
| before — dragging the node wider no longer changes the image size. |
| - **The node has a minimum size**: it cannot be shrunk past what the preview, the |
| buttons and the settings need. |
| - **Everything is written to `output\BONSO_POSES`** — `ENVIAR A` poses and the |
| `.json`, plus the copies that `Cargar imagen` uploads. `Multi Image Reference` |
| (Academia SD) can only read from `input`, so for that node a temporary copy is |
| also placed there; the archived pose always stays in `output`. |
| - **No wire outputs.** The result leaves the node only through **ENVIAR A**, |
| which takes a target node name in a text row and hands the rendered pose to |
| it. Works with `Load Image` and with `Multi Image Reference` (Academia SD), |
| using the same convention as `AcademiaSD Image Save & Send`. |
| - **Output keeps the reference image's aspect ratio**, scaled to the requested |
| megapixels (default 1 MP). |
| - **Manual posing**: drag any joint; right-click a joint to hide it (hiding a |
| joint removes every segment attached to it). Pan with |
| <kbd>Space</kbd>/<kbd>Shift</kbd>+drag, zoom with the wheel, <kbd>F</kbd> to fit, |
| <kbd>Ctrl</kbd>+<kbd>Z</kbd> to undo. |
| - **The editor is black past the picture.** Around the reference there is an |
| infinite black canvas, so you can move a joint **outside** the frame — useful |
| when the person is cropped at the waist. The dashed frame shows where the |
| export will cut; the cropped result is exactly the picture's size. |
| - Joints are **not labelled** any more: only the mannequin you place. |
| - Renders in the exact format ControlNet's OpenPose model reads: black |
| background, 60 %-brightness limbs, full-brightness joints, the standard 18 |
| keypoints. |
| - Also writes a ControlNet-compatible `.json` next to the PNG. |
| - **Hands and face are not supported.** Body keypoints only. |
|
|
| ### Mask Crop Bonso |
| Square mask crop that repositions instead of clamping at image edges. Drop-in replacement for WAS Mask Crop Region. |
|
|
| ### Easy Model Bonso |
| Model inventory manager with download capabilities. Track your models, download missing ones, and save/load your inventory as a lightweight JSON file. |
|
|
| - Scan the active workflow to list every referenced model, resolved against your |
| local folders (subfolder-agnostic search, so a workflow pointing at `KREA\` will |
| still find the file in `KREA2\`). |
| - Auto-resolve download links from CivitAI and HuggingFace; the link field stays |
| visible so you can paste one manually when no match is found. |
| - Per-row status LED, background downloads with progress, drag to reorder, and |
| export/import of the inventory as JSON. |
|
|
| ### Bonso_FileName |
| |
| Builds a folder path out of plain words. One word per line, up to **6** lines |
| (`Linea_1` … `Linea_6`): |
| |
| - **Añadir** (green) opens a new empty line below, **Eliminar** (pink) removes |
| the last one. Both buttons always stay visible side by side at a fixed size |
| (68×20) — when they cannot act (6 lines / 1 line) they just dim. |
| - The node grows/shrinks with the lines; resizing the node does not change the |
| button size. |
| - The output is every non-empty line joined with `/`, ready to plug into any |
| text input: |
| |
| `proyecto` + *(empty)* + `escena_01` → `proyecto/escena_01` |
| |
| Empty lines are ignored, and if every line is empty the node returns `Bonso`. |
| No inputs, one `texto` output. |
| |
| ## Installation |
| |
| ```bash |
| cd ComfyUI/custom_nodes |
| git clone https://huggingface.co/OSOBONSO/comfyui_Bonso_Tools.git |
| ``` |
| |
| Restart ComfyUI after cloning. The five nodes show up as **BonsoNet**, |
| **Bonso Pose**, **Mask Crop Bonso**, **Easy Model Bonso** and |
| **Bonso_FileName**. |
| |
| **Requirements**: ComfyUI. |
| [`comfyui_controlnet_aux`](https://github.com/Fannovel16/comfyui_controlnet_aux) |
| is optional — BonsoNet uses it for the best detectors and falls back to native |
| OpenCV/PIL when it is missing. |
| |
| Version: see `pyproject.toml` (currently **1.4.4**). |
| |
| ## Contributing |
| |
| `BITACORA.md` is the handoff document for AI agents. It contains a mandatory |
| protocol (single source of truth, verification rules, commit discipline) and the |
| verified state of the package. Agents should read it before touching anything. |
| |
| ## License |
| |
| Apache 2.0 |
| |
