ComfyUI_Pose-Gallery_ZH Introduction
ComfyUI_Pose-Gallery_ZH brings a searchable pose-reference gallery directly into ComfyUI. You can find a pose, compare its available viewing angles, and send your chosen reference straight to a node in your workflow—without manually downloading and loading each image.
The extension includes:
- 288 pose groups containing 1,439 reference images.
- Full-size images and thumbnails bundled with the extension for offline browsing and output.
- Chinese-keyword and image-ID search, with filters for pose system, category, and viewing angle.
- Single-image and batch nodes, including an option to output every available angle of a pose.
- Image resizing, horizontal flipping, favorites, and recently used references.
- Support for your own reference collection through a custom asset directory. Pose-reference picker inside ComfyUI
For AI art, this gives you a convenient reference shelf beside your workflow. Instead of repeatedly searching folders for a suitable stance or camera angle, you can browse visually, preview the reference, and replace it while keeping your existing node connections.
The extension is based on Anhao’s Chinese pose gallery. The original images and data come from the upstream author, ギア / giaaibusi.
Scope: This extension focuses on pose-reference selection and image output. The upstream project also documents clothing, portraits, depth images, and a three-in-one character studio. Those features belong to the broader upstream project and should not be assumed to be included in ComfyUI_Pose-Gallery_ZH.
How ComfyUI_Pose-Gallery_ZH Works
Think of the extension as an illustrated catalog connected to an image loader.
-
The gallery helps you find a reference. You search or filter the collection and inspect the available views.
-
Your selection identifies an image. Each reference has an image ID. The gallery places that ID into a single-image or batch node.
-
The node prepares the image. When you run the workflow, the node loads the reference from the selected asset library and applies your size and flip settings.
-
The image becomes available to other nodes. Its
IMAGEoutput can connect to an image preview, image-saving node, or compatible image-processing workflow.
Reference images are not pose skeletons
The extension outputs human-figure reference images, not extracted joints, keypoints, or OpenPose skeletons.
If you want the reference to guide the pose of an AI-generated character, a typical workflow is:
text PoseGalleryImage → DWPose / OpenPose → Pose ControlNet → Sampling → Decoding → Save Image
Each stage has a separate job:
- Pose Gallery: supplies the reference image.
- DWPose or OpenPose: reads the image and estimates a skeleton.
- Pose ControlNet: uses that skeleton to guide image generation.
- Your generation workflow: determines the resulting character, appearance, and image. This separation matters: choosing a reference does not by itself generate an image or guarantee an exact pose match.
ComfyUI_Pose-Gallery_ZH Features
Search and filter references
Open ComfyUI_Pose-Gallery_ZH in the left sidebar, then search using a Chinese keyword or image ID.
You can combine filters for:
- Pose system: the broader reference collection.
- Category: a narrower group of poses.
- Viewing angle: the available perspective of a pose. The gallery uses Chinese metadata. If you do not know an appropriate Chinese search term, you can browse visually with filters or return to a known reference using its image ID.
Two browsing modes are available:
- Grouped by pose: useful when you want to compare views of the same pose.
- Individual-image browsing: useful when choosing a specific reference image.
Large preview and expanded picker
Click an image to open its large preview. This lets you inspect the pose and confirm the viewing angle before sending it to a workflow.
Use Expand to open the larger picker. Your current sidebar selection carries over into it, and selections synchronize back when you close it.
You can also:
- Download the original image.
- Copy its image ID.
- Select all available angles of the current pose.
Send a reference to a new or existing node
Click Use This Pose in the preview to create a single-image node by default.
To update a node you already use:
- Choose that node in the Send To selector at the bottom of the gallery, or open the gallery from the node’s Open Pose Gallery button.
- Select a replacement reference.
- Send it to the selected node. Updating an existing gallery node preserves its connections, size settings, and horizontal-flip setting. This makes it easier to compare poses without rebuilding your workflow.
If an input is controlled by a connected wire, the extension prompts you to disconnect that input or create a new node instead.
Batch selection
Select multiple images, then click Send Selected Images to create a batch node by default.
- Images are output in selection order.
- View Selected lets you review and remove references before sending them.
- A batch can contain up to 32 images.
- The total batch output is limited to 32 million pixels. For example, you can collect several candidate poses and preview them together before choosing one for your generation workflow.
A batch is a collection of separate images—not a combined contact sheet.
All available angles of one pose
The batch node includes a Current Pose: All Angles mode. It uses the pose associated with the node’s image ID.
The default view order is:
- Front
- Oblique
- Side
- Top-down
- Low-angle Unavailable views are disabled rather than invented. One pose group in the bundled collection has no low-angle image, so not every pose produces five images.
Image sizing and horizontal flipping
Both nodes offer four size-handling modes:
| Mode | What it does | When it is useful | | --- | --- | --- | | Original size | Keeps the image’s original dimensions. | Inspecting or processing the original reference. | | Fit with padding | Preserves proportions and adds white space to reach the requested size. | Keeping the full reference visible at a consistent output size. | | Center crop | Preserves proportions but removes overflow around the center. | Filling a target frame when losing some outer content is acceptable. | | Stretch | Resizes directly to the requested dimensions without preserving proportions. | Producing an exact size when distortion is acceptable. |
Width and height range from 64 to 4096 pixels and apply only outside Original Size mode.
Additional behavior:
- Bundled originals are 900 × 1200 pixels.
- The single-image node defaults to Original Size.
- The batch node defaults to Fit with Padding at 768 × 1024.
- Transparent images are composited onto a white background.
- Horizontal Flip mirrors the output left to right. For pose work, Fit with Padding is helpful when you want to preserve the complete figure. Center Crop may remove hands, feet, or other details near the edges, while Stretch can change body proportions.
Horizontal Flip is useful when you need a gesture or body direction mirrored without choosing another reference.
Favorites, recent references, and shortcuts
Favorites are saved for individual images and viewing angles, not entire pose groups.
Favorites and recently used references are stored in your current browser. They are not included when you share a workflow.
| Shortcut or gesture | Action |
| --- | --- |
| / | Focus search |
| ← / → in the large preview | Switch viewing angle |
| ↑ / ↓ in the large preview | Switch pose |
| Esc | Close the preview |
| Swipe left or right on a touchscreen | Navigate within the preview |
ComfyUI_Pose-Gallery_ZH Nodes and Parameters
The extension provides two nodes:
| Node | Internal name | Outputs |
| --- | --- | --- |
| Pose Gallery · Single Image | PoseGalleryImage | IMAGE, image ID, pose description |
| Pose Gallery · Batch | PoseGalleryBatch | IMAGE batch, JSON information manifest |
IMAGE is the image format used to pass pictures between compatible ComfyUI nodes. The text outputs help you identify and record which references were used.
Shared parameters
| Parameter | Meaning | | --- | --- | | Asset library | Selects the bundled library or a configured custom library. | | Image ID | Identifies the reference. The picker fills this automatically, or you can enter it manually. | | Size handling | Chooses Original Size, Fit with Padding, Center Crop, or Stretch. | | Width / Height | Sets output dimensions for modes other than Original Size. | | Horizontal flip | Mirrors the output image left to right. |
Batch-specific parameters
The batch node also provides:
-
Image ID list
-
Selection mode
-
Ordering A manually entered ID list accepts:
-
Line breaks
-
Spaces
-
English or Chinese commas
-
English or Chinese semicolons Duplicate IDs are removed automatically.
The batch node’s JSON information manifest is a structured text record corresponding to the output images, one entry per image. It records:
- Image ID
- Pose system
- Pose number and name
- Category
- Viewing angle
- Output dimensions
- Flip setting This is useful when you need to keep track of the references used in a batch.
ComfyUI_Pose-Gallery_ZH Custom Assets and Workflow Sharing
Bundled references
By default, the extension reads its own data/, img/, and thumbs/ directories. It does not depend on folders from the author’s computer.
The bundled assets occupy approximately 659.5 MB, including:
- 1,439 full-size PNG images
- 1,439 JPG thumbnails
Use your own reference library
To select a custom asset directory:
- Open Asset Settings in the gallery.
- Enter the directory’s absolute path—its complete location on the computer.
- Click Check and Save.
- Select Custom Asset Directory at the top of the gallery.
- Set the destination node’s asset library to the custom library as well. The extension checks the manifest structure, IDs, and original-image files before saving. An invalid configuration does not replace your previous valid settings.
To remove the custom directory setting, clear the path and save.
The path must exist on the computer running ComfyUI. If you access ComfyUI through a browser on another device, a folder on that browsing device is not automatically available to the server.
Custom assets are managed through files, folders, and a manifest. See the custom asset guide for the required directory structure and JSON examples.
Settings are stored in the ComfyUI user directory at:
text pose_gallery/settings.json
They are not written into shared workflows or the extension’s source code. Changing the gallery’s browsing source does not automatically change existing nodes.
What a saved workflow remembers
A workflow saves the node’s:
- Image ID
- Asset library
- Size parameters
- Horizontal-flip setting For bundled references, another computer with this extension and its bundled assets can resolve the same image ID.
For custom references, the recipient must configure a suitable local directory and preserve the same asset IDs. Sharing the workflow alone does not supply those custom images.
ComfyUI_Pose-Gallery_ZH Models and Pose-Guided Generation
The gallery itself does not require an inference model. Browsing references, outputting images, resizing, and flipping do not require a diffusion checkpoint or a pose-detection model.
Models become relevant only when you connect the gallery to pose extraction or generation.
| Component | Purpose | When you need it | | --- | --- | --- | | DWPose Estimator | Estimates body, hand, and face keypoints from a reference image. | When converting the selected reference into a pose-control image. | | OpenPose Estimator | Extracts a pose representation, with body, hand, and face detection options. | As another supported approach to creating pose-control images. | | Pose ControlNet weights | Guide generation using the extracted pose. | When using a ControlNet-based generation workflow. | | Base generation model | Produces the final image within your generation workflow. | When generating artwork rather than only previewing references. |
DWPose and OpenPose preprocessing are provided by the separate ComfyUI ControlNet Auxiliary Preprocessors extension. Its first run may download preprocessing models.
Your base model and pose ControlNet must be compatible with one another.
Different pose estimators can interpret a reference differently, especially with:
- Two people
- Overlapping or hidden limbs
- Fingers
- Unusual camera angles Preview the extracted skeleton before generation. A good-looking reference image does not guarantee that every joint will be detected correctly.
ComfyUI_Pose-Gallery_ZH Example Workflows
Drag an example JSON file into ComfyUI to load it.
| Example | What it demonstrates | Additional requirements | | --- | --- | --- | | 01_single_image.json | One reference image connected to a preview. | None beyond this extension and its assets. | | 02_batch_all_angles.json | All available angles of a pose connected to a preview. | None beyond this extension and its assets. | | 03_dwpose_preview.json | Reference image → DWPose → skeleton preview. | ControlNet Auxiliary Preprocessors and preprocessing models. | | 04_controlnet_template.json | DWPose and ControlNet connected into a generation workflow. | Your own compatible base model and pose ControlNet weights. |
Suggested starting point
- Run the single-image example to check reference output.
- Try the all-angles example to understand batch selection.
- Inspect pose extraction with the DWPose preview example.
- Move to the full generation template after verifying the skeleton. The author reports that the first example was tested successfully and the second produced five images successfully.
The third and fourth examples had their node interfaces and connections checked, but model inference and diffusion sampling were not executed.
The fourth example is an SD1.5, 512 × 768 template. Its model names are placeholders: choose compatible models available on your computer before running it.
Troubleshooting ComfyUI_Pose-Gallery_ZH
The sidebar or nodes do not appear
- Restart ComfyUI and refresh your browser.
- Search for the internal node names
PoseGalleryImageandPoseGalleryBatch. - Check the ComfyUI startup log for extension-loading errors.
- Confirm that
__init__.pysits directly inside the extension folder rather than inside an extra nested folder. - If your older frontend does not show the sidebar, try opening the gallery through a node’s Open Pose Gallery button. The author verified the extension with ComfyUI 0.43.0, frontend 1.57.6, and Python 3.12. Other versions were not actually verified. A modern frontend supporting custom sidebars is recommended.
Images or thumbnails are missing
- Check that the bundled
data/,img/, andthumbs/directories are present. - Make sure the full image collection is available, not only the extension code.
- If using custom assets, check that the configured directory is available on the ComfyUI computer.
- Check that the referenced originals exist. For a missing custom thumbnail, the extension falls back to the original image.
For a more detailed asset check, run the following from the extension directory using ComfyUI’s Python environment:
bash python -X utf8 scripts/verify.py
An image ID is reported as invalid
- Check which asset library the node is using.
- Confirm that the ID belongs to that library’s manifest.
- Reselect the reference from the appropriate gallery source.
- For custom libraries, check that IDs have not changed. An invalid ID is not silently replaced with another pose.
A new selection does not update the intended node
- Check the gallery’s Send To selector.
- Select the node you intend to update, or reopen the gallery from that node.
- Check whether the relevant input is controlled by a wire.
- If prompted, disconnect that input or create a new node. Also check the node’s asset-library setting: changing the gallery’s browsing source does not rewrite existing nodes.
A batch fails because image dimensions differ
- Open the batch node.
- Change Size Handling from Original Size to Fit with Padding, Center Crop, or Stretch.
- Choose a shared width and height.
- Run the workflow again. Fit with Padding preserves each image’s proportions and avoids cropping the figure.
A batch exceeds its limits
- Reduce the selection to 32 images or fewer.
- Reduce output width and height if needed.
- Keep the combined output at or below 32 million pixels.
- Split a larger selection into separate runs. For equal-size images, total output pixels are:
text image count × width × height
The default 32-image batch at 768 × 1024 is approximately 25.2 million pixels.
An angle is disabled or only four images are output
- Check which angles exist for the selected pose.
- Use the enabled views.
- Do not expect the extension to generate missing angles. One bundled pose group lacks a low-angle reference.
Pose detection looks incorrect
- Preview the original reference.
- Preview the DWPose or OpenPose output separately.
- Look for missing people, misplaced joints, or hidden limbs.
- Try another available reference angle.
- Confirm that your generation workflow uses a compatible pose ControlNet. Detection quality depends on the external preprocessor, not just the gallery reference.
Frequently asked questions
Can I use the gallery offline? Yes. The bundled references support offline browsing and image output. External preprocessing or generation tools may need model downloads before they can work offline.
Does this extension generate pictures by itself? No. It selects and outputs reference images for other nodes to use.
Are depth images, clothing references, and portraits included? Those are documented features of the broader upstream project, not the documented feature set of this standalone extension.
Why are my favorites missing on another computer? Favorites and recent references are browser-local and are not part of a shared workflow.
Why does my custom library not work for someone receiving my workflow? They need their own valid asset directory with matching IDs. Your local path is not included in the shared workflow.
Can I use the bundled assets commercially? Do not treat the code’s MIT license as permission for commercial use of the assets. The original README states that the bundled images and original data are outside the MIT license and are restricted from commercial use. Review LICENSE.
Learn More about ComfyUI_Pose-Gallery_ZH
- Extension repository — the standalone extension, bundled assets, and project documentation.
- Custom asset guide — directory structure and manifest examples for your own reference collection.
- Testing documentation — detailed verification records and browser-testing information.
- Archived original documentation — historical upstream-site documentation.
- Anhao’s upstream Chinese gallery project — the broader reference-gallery and character-workbench project.
- Upstream online demo — explore the broader project in a browser; its features differ from this extension.
- Original pose-reference gallery (https://ai-pose-gallery.pages.dev/) — the source pose-gallery site.
- ComfyUI ControlNet Auxiliary Preprocessors — documentation for DWPose, OpenPose, and other reference-processing nodes.
ComfyUI_Pose-Gallery_ZH Licensing and Asset Use
The extension’s code and bundled reference materials have different licensing scopes:
- Source code: MIT.
- Images and original data: not covered by the MIT license. The original README states that the original repository prohibits commercial use of these materials. Making the plugin open source does not remove the restrictions on its reference assets. Review LICENSE before using, redistributing, or sharing bundled materials.
