Jakkanna is an independently governed ComfyUI custom-node suite for 3D posing, OpenPose control, canvas composition, camera direction, regional detailing, and model selection. It is derived from AHEKOT/ComfyUI_VNCCS_Utils and preserves its established workflow identifiers while following an independent release and development path.

The name honours the legendary master sculptor popularly known as Jakanachari or Jakkanna. It expresses the project's aim: precise control of form, movement, framing, and composition.

[!IMPORTANT] Install Jakkanna as a replacement for the upstream vnccs-utils package. Do not install both packages in the same ComfyUI instance. Jakkanna retains the original VNCCS_* node identifiers and /vnccs/ routes so existing workflows continue to load.

[!NOTE] Pose Studio images are captured by the node's browser viewport, not by a server-side renderer. Run Pose Studio workflows from a browser session in which the node is open and its 3D viewer has finished loading; the backend then synchronizes the captures it needs. Headless or API-only execution without synchronized captures stops with a clear error by design — Jakkanna deliberately removed the upstream Python fallback renderer rather than let it produce geometry that differs from the viewport.

Why Jakkanna Exists

Jakkanna begins at version 1.0.0 with a correctness baseline that:

Pose Studio and Pose Studio + OpenPose resolve one effective execution state and render from it exactly once.
Captured images, lighting prompts, and OpenPose keypoints remain aligned across ordinary execution, frontend synchronization, cache fallback, and SAM-driven pose application.
Skin weights are canonicalized, normalized, and limited to four influences per vertex.
Browser and Python MakeHuman morph factors agree for body proportions and ancestry controls.
Pose, capture, lighting, prompt, and total-pixel limits are consistent across browser and server boundaries.
OpenPose validates frame structure and dimensions, exports real face landmarks, and translates people correctly in grid output.
Mixamo sampling is hard-capped without dropping the animation endpoint.
Repository refresh is explicitly user initiated; opening Pose Studio no longer starts an external repository refresh.

See FORK_NOTICE.md for lineage, compatibility, and a comparison with related ComfyUI pose nodes.

Who Jakkanna Is For

Use Jakkanna when a ComfyUI workflow needs an accurate production control source. It is particularly useful for:

existing VNCCS workflows that need the corrections without node migration;
workflows that require rendered frames and body, hand, and face OpenPose data to remain exactly aligned;
batch, grid, Mixamo, SAM-driven, pose-library, and MakeHuman morph workflows; and
workflow authors who need deterministic output when sharing graphs with other users.

If you only need lightweight 2D keypoint editing and do not use VNCCS workflows, a dedicated OpenPose editor may be a smaller fit.

<table> <tr> <td width="33%" align="center"> Upstream VNCCS Community Share results, ask questions, and follow VNCCS updates. <a href="https://discord.com/invite/9Dacp4wvQw" target="_blank">Join the upstream Discord</a> </td> <td width="33%" align="center"> Support Jakkanna Support ongoing compatibility work, testing, fixes, and releases. <a href="https://www.buymeacoffee.com/teenu" target="_blank">Support Jakkanna</a> </td> <td width="33%" align="center"> Support Original VNCCS Support MiuProject's original VNCCS development separately. <a href="https://www.buymeacoffee.com/MIUProject" target="_blank">Support the original project</a> </td> </tr> </table>

Jakkanna Canvas

Jakkanna Canvas is an integrated infinite-canvas image generation and editing workspace inside ComfyUI. It is designed for freeform creative work: generate anywhere, edit any region, build results across layers, and keep refining without being locked to a single fixed image boundary.

Key Features

Infinite Canvas Workflow: Work beyond a single image frame and place generations wherever the composition needs them.
Layer-Based Editing: Build images from separate raster and mask layers with visibility, opacity, selection, movement, and compositing controls.
Generation Anywhere: Use a selected region as the generation target for new images, image edits, inpaint, outpaint, and full-area transformations.
Mask and Object Tools: Paint masks, refine selections, and use SAM-powered object selection to isolate or remove parts of an image.
Preset and Custom Models: Switch between built-in presets or use manual model selection for supported generation backends.
Turbo and LoRA Controls: Use Turbo LoRA cards and a general LoRA Stack directly from the generation panel.
Canvas Editing Tools: Move, transform, resize, snap, undo/redo, and manage generation results without leaving the node.
Progress and Result Handling: Track generation progress and apply results back into the canvas as editable layers.

Krea 2 Community Workflows

The plain Krea 2 workflow provides a text-to-image baseline, while the Pose Studio workflow uses the editable mannequin render as an image-to-image structural reference. Both use raw.safetensors, MysticXXX Stripped at 0.55, and Raw-to-Turbo at 1.00. The plain text-to-image baseline uses Euler/beta for 12 steps at 1024 × 1024. The Pose Studio image-to-image workflow uses the empirically verified 704 × 1280 preset: ER-SDE/simple, 8 steps, CFG 1, denoise 0.60, and a fixed seed. Its raw-capture preview should show a shaded, high-contrast figure; flat grey captures leave too little form information for restyling. Run it with the Pose Studio node open in the browser so captures can synchronize (see the capture note above). With ComfyUI execution caching enabled, the fixed seed and unchanged inputs allow the completed output to be reused; change the seed or another input to force sampling and explore variants. Describe the pose explicitly in the prompt for the strongest adherence. This remains VAE image-to-image conditioning rather than ControlNet: at denoise 0.60, Krea 2 can still reinterpret a pose that is not described.

FBX Animation and SCAIL-2

Pose Studio imports animated Mixamo-compatible FBX files directly in the browser and samples them into deterministic pose sequences. The importer uses the FBX skeleton and animation; an FBX skinned mesh, materials, and textures are not applied to the Jakkanna mannequin.

The animation controls default to 81 frames at 16 fps:

FIT_CLIP samples the complete clip once across all requested frames. It is the safest choice for ordinary short Mixamo clips and changes playback speed without repeating the motion.
REALTIME samples at exact 1 / fps intervals from the selected start time. An 81-frame sequence at 16 fps requires source animation through 5.000 seconds; shorter clips fail instead of being stretched silently.
PRESERVE root motion retains the FBX pelvis translation on the Jakkanna mannequin. Use IN_PLACE only when the downstream animation is intentionally stationary.

The bundled Fast and Production workflows use one Pose Studio node for the prompt, Krea 2 reference frame, and complete 81-frame motion driver. Fast uses the six-step Lightx2v/Euler/simple topology; Production uses the 40-step Flow-UniPC topology. Both validate the input video and SAM 3.1 track and write a reproducibility manifest. The lossless FFV1 master requires ComfyUI-VideoHelperSuite 1.7.9 or newer. Model loaders identify the required Krea 2, SCAIL-2, SAM 3.1, text-encoder, CLIP-Vision, VAE, and LoRA artifacts; Jakkanna does not download those files when the workflow opens.

To animate an existing single-human image instead of authoring the appearance with Krea, use the content-addressed Registered Image + FBX Animation tool. It fits the image with SAM 3D Body, retargets a compatible binary Mixamo FBX onto that body, renders a camera-registered driver and exact masks, and generates a separate hardened SCAIL-2 workflow. This is intentionally a distinct product path: wiring an arbitrary photograph directly into the Krea workflow would discard the reference/driver registration that makes the animation reliable. The bundled Greenfield Image + FBX workflow reduces this path to four intended controls: reference image, FBX selection, camera preset, and customization prompt. Its preparation, retargeting, framing, validation, sampling, and manifests run automatically. SAM 3.1 isolates the person before SAM 3D Body fitting and again after camera registration; the user does not provide or adjust either mask. The current animation-directory compatibility results are recorded in the FBX input audit. The camera preset offers deterministic full, American, medium, close, and Dutch-roll shot choices. It retains a fixed camera across the clip and applies the same transform to the photograph, driver, and masks. Camera angles that would reveal an unseen side of the subject are excluded because they require novel-view reconstruction from a single image. This workflow requires ComfyUI 0.27.0 or newer, an NVIDIA CUDA GPU, the model files listed in the image/FBX guide, and ComfyUI-VideoHelperSuite 1.7.9 or newer.

Posed Mesh Export

tools/export_posed_mesh.py exports a saved Pose Studio workflow or raw pose_data JSON as OBJ or metre-scaled GLB. It can also write posed joint coordinates for downstream tools:

python tools/export_posed_mesh.py workflow.json \
  --output mannequin.glb \
  --joints mannequin-joints.json \
  --apply-model-rotation

Use --all to export every pose and --triangulate when the downstream tool requires triangles. GLB export uses the trimesh dependency installed with Jakkanna.

SCAIL Turntable Asset

The RealityScan Fast and Production workflows can bake their 81 SCAIL-2 replacement frames directly onto the clean posed mannequin:

tools/run_scail_proxy_asset.sh saved-workflow.json \
  /path/to/Jakkanna/RealityScan/fast/run_00001

The tool rebuilds pose 0 and its exact Pose Studio camera, rejects self-occluded samples, locally aligns the generated masks, and writes JakkannaSCAILProxy.glb with a 2048-pixel unlit texture. Use --texture-size 512 for quick tests. RealityScan alignment is optional validation; unrestricted photogrammetry is not used as the asset geometry.

Jakkanna Pose Studio

Example Workflows: Krea 2 Community · QWEN · Klein9b

Jakkanna Pose Studio is a professional 3D posing, framing, lighting, and pose-library environment running directly inside a ComfyUI node. It is designed for building high-quality pose/control references without leaving the graph: adjust the character body, pose bones interactively, frame the camera, tune lights, manage saved poses, and output single images or pose batches.

Key Features

Interactive 3D Viewport: Pose the mannequin directly in the node with selectable joints, bone manipulation, transform controls, and full Undo/Redo support.
Dynamic Body Generator: Fine-tune the character shape with sliders for Age, Gender blending, Weight, Muscle, and Height.
Multi-Pose Tabs: Create multiple independent pose states inside one node, making batch outputs and pose sequences easier to build.
Pose Copy/Paste: Transfer complex poses between tabs without rebuilding them from scratch.
Modal Pose Gallery: Save, browse, load, and delete poses in a focused full-screen gallery instead of cluttering the main workspace.
Pose Import/Export: Batch save and load pose data via JSON for reuse across workflows or projects.
Tracing Support: Load a background reference image and align the 3D character to it for accurate pose matching.
Precision Camera Controls: Set output dimensions, zoom, model rotation, and camera orbit with an integrated radar-style control.
Viewport Frame Preview: Preview the final render boundary directly in the viewport so composition matches the output.
Advanced Environment Lighting: Control Ambient, Directional, and Point Lights, including 2D radar controls for positioning point lights and radius controls for their influence.
Keep Original Lighting Mode: Bypass synthetic lighting for clean flat renders, useful for ControlNet-style pose/reference outputs.
Prompt-Aware Lighting Output: Generate descriptive lighting prompts that can be combined with your scene prompt.
Custom Prompt Templates: Use tag-based templates such as <lighting> and <user_prompt> to control how the final prompt is assembled.
Direct Sidebar Prompting: Add scene details in an auto-expanding prompt field directly inside the Pose Studio UI.
Flexible Export Modes: Output poses as a list or as a grid, with configurable background color.

👉 Detailed Usage Guide

Additional Nodes

Jakkanna Visual Camera Control

Example Workflow

An interactive node with a visual widget for controlling camera position. It is designed for intuitive angle control and prompt generation, especially for multi-angle LoRAs like Qwen-Image-Edit-2511-Multiple-Angles.

Visual Widget: Select azimuth and distance with the mouse.
Elevation Slider: Pick elevation from -30° to 60°.
Trigger Word Toggle: Enable or disable the <sks> trigger from the widget.

Jakkanna QWEN Detailer

Example Workflow

A QWEN-Image-Edit2511 detailer for enhancing detected regions such as faces, hands, and objects with vision-guided instructions.

Smart Cropping: Automatically squares crops and handles padding.
Vision-Guided Enhancement: Uses QWEN-generated instructions or user prompts.
Drift Fix: Helps keep the enhanced area aligned with the original composition.
Quality of Life Tools: Includes color matching, Poisson blending, and upscaling options.
Inpainting Mode: Supports mask-based editing and filling black areas.
QWEN Options: Supports QWEN-Image-Edit2511-specific options such as distortion_fix and qwen_2511 mode.

Jakkanna Model Manager & Selector

Example Workflow

A system for managing and selecting LoRAs and checkpoints directly in ComfyUI, with support for Civitai and HuggingFace.

Jakkanna Model Manager

The backend node that reads a HuggingFace-hosted model_updater.json and manages model downloads.

Repo ID: Point the manager to your HuggingFace model repository.
Downloads: Queue and download models in the background.
Civitai Support: Use API key authentication for restricted Civitai models.

👉 Configuration Guide: How to create your own model repo

Jakkanna Model Selector

The companion UI node for choosing models from the configured repository.

Visual Card UI: Shows model name, version, status, and description.
Smart Search: Opens a searchable modal model list.
Status Indicators: Shows Installed, Update Available, Missing, and Downloading states.
One-Click Install/Update: Install or update models directly from the selector.
Universal Connection: Outputs a standard relative path string compatible with standard ComfyUI nodes.

👉 Usage Guide: How to use Selector with Standard Loaders

Jakkanna BBox Extractor

A helper node for extracting and visualizing crops when you need detected bounding-box regions without running a full face/detailer workflow.

Installation

Recommended: ComfyUI Manager

Open ComfyUI Manager.
Choose Custom Nodes Manager.
Search for Jakkanna.
Click Install.
Restart ComfyUI.

Remove the upstream vnccs-utils package first if it is already installed.

Manual Installation

Open a terminal in your ComfyUI directory and run:

cd ComfyUI/custom_nodes
git clone https://github.com/teenu/ComfyUI-Jakkanna.git
cd ComfyUI-Jakkanna
pip install -r requirements.txt

Restart ComfyUI after installation.

Development

The regression suite uses the standard-library unittest runner and needs no extra test dependencies. Run it with the same Python environment that runs ComfyUI:

python -m unittest discover -s tests

pytest also discovers these tests if it happens to be installed, but it is not required.

Lineage, Licensing, and Maintenance

The original VNCCS implementation remains attributed to MiuProject under the MIT license. MakeHuman-derived code and assets, SAM 3D Body, MHR, Three.js, and other bundled components retain their own licenses and notices; see THIRD_PARTY_NOTICES.md for the package-level inventory.

Jakkanna is maintained independently at teenu/ComfyUI-Jakkanna. It is not affiliated with or endorsed by the upstream VNCCS project. Please report Jakkanna defects through its issue tracker.