ComfyUI-Jakkanna Introduction
ComfyUI-Jakkanna is a collection of ComfyUI nodes for controlling poses, camera angles, lighting, image composition, regional edits, and model selection. Its central tool, Jakkanna Pose Studio, lets you pose a 3D mannequin inside a ComfyUI node and turn that pose into visual references for image or animation workflows. Jakkanna Canvas provides a separate, layer-based workspace for generating and editing images beyond a single fixed frame.
The extension is useful when a prompt alone does not give you enough control. For example, you can arrange a character’s limbs and camera framing before generation, use a rendered mannequin as an image-to-image reference, or produce OpenPose body, hand, and face information for a workflow that accepts it. You can also build pose sequences, refine particular image regions, and choose models from a visual list.
Jakkanna is independently maintained and derived from AHEKOT/ComfyUI_VNCCS_Utils. It keeps the original VNCCS_* node identifiers so existing VNCCS workflows can continue to load. Use Jakkanna in place of the upstream vnccs-utils package, not alongside it in the same ComfyUI instance. For its lineage and compatibility details, see the fork notice.
How ComfyUI-Jakkanna Works
Think of Pose Studio as a small virtual photo studio inside your workflow:
- Arrange the subject. Move the mannequin’s joints, adjust its body shape, or load a saved pose.
- Set up the shot. Choose the camera angle, frame, background, and lighting.
- Capture the reference. Pose Studio produces an image of the arranged scene and, where the workflow calls for it, corresponding OpenPose keypoints. The rendered figure and pose data are designed to stay aligned.
- Pass the reference to the rest of your graph. A generation workflow can use the capture as a visual guide, while a compatible pose-control workflow can use the keypoint information.
A rendered reference is not a guarantee that a generation model will reproduce every detail. In the bundled Krea 2 Pose Studio workflow, for example, the mannequin is used as an image-to-image reference rather than ControlNet conditioning. The model can still reinterpret the pose, especially if your text prompt does not describe it clearly.
Pose Studio captures what you see in its browser-based 3D viewer. When running a Pose Studio workflow, keep the node open in a browser session and wait for the viewer to finish loading. The backend synchronizes the captures from that viewer. An API-only or headless run without synchronized captures stops with an error rather than producing a potentially different render.
Other parts of Jakkanna serve different stages of the same creative process: Canvas lets you generate and revise regions on layers; the QWEN Detailer focuses edits on detected areas; and the Model Manager and Selector help you find and load files used by your workflow.
ComfyUI-Jakkanna Features
Pose Studio: shape, pose, and frame your reference
Pose Studio combines a 3D posing workspace with controls for the image it outputs. You can:
- Pose interactively: Select and move joints and bones in the viewport. Undo and redo let you experiment without rebuilding a pose.
- Adjust body proportions: Use controls for age, gender blending, weight, muscle, and height when a different mannequin shape would better match your subject.
- Work with several poses: Keep independent poses in tabs, copy and paste poses between them, and output poses as a list or grid. This is useful for comparing compositions or preparing a sequence.
- Reuse poses: Save and browse poses in the gallery, or import and export pose data as JSON for another project.
- Trace a reference: Place a background image behind the mannequin to help match an existing pose.
- Compose the shot: Set output dimensions, zoom, model rotation, and camera orbit. The viewport frame preview shows the boundary of the final image, helping you keep hands or feet in frame.
- Shape the lighting: Adjust ambient, directional, and point lights. Point-light position and influence can be tuned visually. For a clean, flat reference, Keep Original Lighting bypasses synthetic lighting.
- Build prompts: Add scene text in the sidebar and use lighting descriptions or templates containing tags such as
<lighting>and<user_prompt>to assemble the final prompt. - Choose the background: Set a background color for list or grid output.
For instance, a dramatic image-to-image reference may benefit from lighting that makes the mannequin’s form easy to read. A flat pose-control reference may instead call for Keep Original Lighting. The right choice depends on what the next node in your graph needs.
Start with the Pose Studio usage guide, or explore the Krea 2 Community, QWEN, and Klein9b example workflows.
Canvas: generate and edit across a larger workspace
Jakkanna Canvas is an infinite-canvas workspace inside ComfyUI. Rather than treating every generation as a completely separate picture, you can place and refine content wherever your composition needs it.
- Work on layers: Keep raster content and masks separate, then adjust visibility, opacity, selection, movement, and compositing.
- Generate in a selected region: Target an area for a new image, an edit, inpainting, outpainting, or a broader transformation.
- Select and mask objects: Paint masks, refine selections, or use SAM-powered object selection to isolate or remove part of an image.
- Manage the composition: Move, transform, resize, snap, and undo or redo edits within the node.
- Control generation: Choose supported preset or manually selected models, use Turbo LoRA cards or a general LoRA Stack, follow progress, and apply results back onto editable layers.
For example, you could build a scene in layers, select an empty area to extend the background, then mask one object for a focused revision without replacing the whole composition.
Animation and pose sequences
Pose Studio can import Mixamo-compatible animated FBX files in the browser and sample their skeleton animation into a pose sequence. It uses the FBX skeleton and motion; it does not transfer the FBX character’s skinned mesh, materials, or textures onto the Jakkanna mannequin.
Animation controls default to 81 frames at 16 fps:
| Setting | What it does | When to use it |
|---|---|---|
| FIT_CLIP | Samples the complete source clip once over the requested frames, changing playback speed if necessary. | A convenient choice for an ordinary short Mixamo clip. |
| REALTIME | Samples at exact time intervals from the chosen start time. | Use when preserving the motion’s timing matters. The source must be long enough: 81 frames at 16 fps require animation through 5 seconds. |
| PRESERVE root motion | Keeps the FBX pelvis translation. | Use when the character should travel through the shot. |
| IN_PLACE root motion | Keeps the downstream motion stationary. | Use only when stationary animation is intentional. |
The bundled Fast and Production workflows use one Pose Studio node to supply a prompt, a Krea 2 reference frame, and an 81-frame motion driver. Fast uses a six-step Lightx2v approach; Production uses a 40-step Flow-UniPC approach. Both validate their input video and SAM 3.1 track and write a reproducibility manifest. Their required model files are identified by the workflows but are not downloaded simply by opening them. A lossless FFV1 master requires ComfyUI-VideoHelperSuite 1.7.9 or newer.
If you want to animate the person in an existing single image, use the separate Registered Image + FBX Animation guide, rather than feeding an arbitrary photograph directly into the Krea workflow. That path fits a body to the image, retargets a compatible binary Mixamo FBX, and keeps the photograph, motion driver, and masks registered to the same camera. The Greenfield Image + FBX workflow presents four intended creative controls: reference image, FBX selection, camera preset, and customization prompt. Its camera presets include full, American, medium, close, and Dutch-roll shots. The camera stays fixed across the clip; angles that would expose an unseen side of the subject are excluded. See the FBX input audit for recorded animation-directory compatibility results.
Additional creative nodes
- Visual Camera Control: Choose camera azimuth and distance with a visual widget, set elevation from −30° to 60°, and toggle the
<sks>trigger word. It is intended to make angle prompts easier to control, including with multi-angle LoRAs such as Qwen-Image-Edit-2511-Multiple-Angles. - QWEN Detailer: Improve detected regions such as faces, hands, or objects. It can make square crops with padding, use generated or user-written editing instructions, and apply options for alignment, color matching, Poisson blending, upscaling, and mask-based inpainting. It also provides
distortion_fixandqwen_2511options. - BBox Extractor: Extract and visualize detected bounding-box crops when you need to inspect regions without running a full detailer workflow.
Model Manager and Selector
The Model Loader Showcase workflow demonstrates two companion nodes:
- Model Manager reads a Hugging Face-hosted
model_updater.jsonfrom a repository you specify and manages queued background downloads. It also supports API-key authentication for restricted Civitai models. - Model Selector displays models as searchable cards with names, versions, descriptions, and states such as Installed, Update Available, Missing, and Downloading. You can install or update from the selector and pass its relative-path output to standard ComfyUI loader nodes.
A repository refresh is user initiated; opening Pose Studio does not start one. For configuration and workflow examples, see the Model Manager guide and Model Selector usage guide.
ComfyUI-Jakkanna Models and Example Workflows
Jakkanna is a node suite, not a single image-generation model. It supports different generation and animation workflows, but the relevant checkpoint, LoRA, and other model files depend on the workflow you choose. Do not assume those files are bundled or downloaded automatically.
| Workflow | What it demonstrates |
|---|---|
| Plain Krea 2 Community | A text-to-image baseline without a Pose Studio structural reference. |
| Pose Studio Krea 2 Community | Image-to-image generation guided by a rendered, editable mannequin. |
| Pose Studio QWEN and Pose Studio Klein9b | Alternative example graphs connecting Pose Studio to their respective generation workflows. |
| Krea 2 + SCAIL-2 Fast and Production | Two sampling approaches for an 81-frame animation workflow. |
| Greenfield Image + FBX | Animating a registered single-human image using FBX motion. |
Both plain and Pose Studio Krea 2 Community examples use raw.safetensors, MysticXXX Stripped at 0.55, and Raw-to-Turbo at 1.00. The plain baseline uses 1024 × 1024, Euler/beta, and 12 steps. The Pose Studio example uses a tested 704 × 1280 image-to-image preset with ER-SDE/simple, 8 steps, CFG 1, denoise 0.60, and a fixed seed.
These are example settings, not universal requirements. In the Pose Studio example, look for a shaded, high-contrast figure in the raw-capture preview: a flat grey image gives the image-to-image process less visible form to work from. Describe the pose in your prompt as well as showing it in the reference. With ComfyUI execution caching enabled, unchanged inputs and the fixed seed can reuse a completed result; change the seed or another input when you want a newly sampled variation.
What’s New with ComfyUI-Jakkanna
Jakkanna starts its independent release path at version 1.0.0. The provided project information does not include a later version history. Its initial correctness baseline focuses on making control references more dependable:
- A consistent Pose Studio render: Pose Studio and Pose Studio + OpenPose resolve the same effective execution state and render from it once. This helps the visible capture and its control data describe the same pose.
- Better alignment across workflows: Captures, lighting prompts, and OpenPose keypoints stay aligned across normal execution, browser synchronization, cache fallback, and SAM-driven pose application.
- More reliable body shapes: Browser and Python MakeHuman morph factors agree for body proportions and ancestry controls; skin weights are normalized and limited to four influences per vertex.
- Stricter validation: Browser and server apply consistent pose, capture, lighting, prompt, and total-pixel limits. OpenPose checks frame structure and dimensions, exports actual face landmarks, and positions people correctly in grid output.
- Safer animation sampling: Mixamo sampling is capped without dropping the endpoint of the animation.
- Predictable refresh behavior: Opening Pose Studio no longer initiates an external repository refresh.
Together, these changes are meant to reduce mismatches between what you arrange, what you preview, and what the rest of your ComfyUI graph receives.
Troubleshooting ComfyUI-Jakkanna
Pose Studio says captures are missing, or an API-only run stops.
- Open the workflow in a browser.
- Keep the Pose Studio node open and wait for its 3D viewer to finish loading.
- Run the workflow again so the browser can provide the captures the backend needs.
- Do not rely on a headless run unless the required captures have already been synchronized.
This behavior is intentional: Jakkanna does not substitute a separate server-side render that might disagree with the viewport.
My generated character does not follow the mannequin closely enough.
- Check the raw-capture preview. In the Krea 2 example, make sure the figure has visible, shaded form rather than appearing flat grey.
- Describe the desired pose explicitly in your prompt.
- Check which kind of guidance your workflow uses. The bundled Pose Studio Krea 2 example is VAE image-to-image, not ControlNet, and its denoise value of 0.60 still allows reinterpretation.
I run the Krea 2 example again, but get the same result.
The example uses a fixed seed. If ComfyUI execution caching is enabled and the inputs have not changed, it can reuse the completed output. Change the seed or another input to request a new sample.
REALTIME FBX sampling fails on a short clip.
Check that the source animation extends through the time your frame count and frame rate require. The default 81-frame sequence at 16 fps needs motion through 5.000 seconds. If you want to fit a shorter clip across the requested frames instead, choose FIT_CLIP.
The imported FBX does not look like its original character.
Pose Studio imports the FBX skeleton and animation, not its skinned mesh, materials, or textures. The motion is applied to the Jakkanna mannequin. If your goal is to animate the appearance of a person in an existing image, follow the Registered Image + FBX Animation guide.
A downloaded model is not available when I open an example workflow.
Opening a workflow does not download its required artifacts. Check its model loaders for the expected checkpoint, LoRA, VAE, text-encoder, CLIP-Vision, or other files, as applicable. For models managed through Jakkanna’s selector, check whether the card says Missing or Downloading; the selector guide explains how its output connects to standard loaders.
Can I use Jakkanna alongside the original VNCCS package?
No. Jakkanna preserves the original node identifiers and routes for workflow compatibility, so it is intended to replace vnccs-utils in a ComfyUI instance, not run beside it.
Learn More about ComfyUI-Jakkanna
- Pose Studio usage guide — detailed guidance for posing, framing, and using the node.
- Krea 2 Pose Studio example — a starting point for mannequin-guided image-to-image work.
- Registered Image + FBX Animation guide and FBX input audit — guidance and recorded compatibility results for animating an existing image.
- Model Manager guide and Model Selector guide — instructions for working with a model repository and standard ComfyUI loaders.
- Fork notice — project lineage, compatibility, and comparison with related pose nodes.
- Third-party notices — licenses and notices for bundled components.
- Jakkanna issue tracker — report a Jakkanna-specific defect.
- Upstream VNCCS community Discord (https://discord.com/invite/9Dacp4wvQw) — discuss VNCCS-related workflows with the upstream community. Jakkanna is maintained independently and is not affiliated with that project.
