DEV Community

AI-KSK
AI-KSK

Posted on Edited on Originally published at huggingface.co

3D Pose and Lighting Control Without Leaving ComfyUI

Pose control usually means a separate 3D application: pose the mannequin there, export, import. VNCCS Pose Studio moves that inside ComfyUI, and this workflow pairs it with Qwen 2511 for character-consistent rendering.

The graph

Node Value
UNETLoader qwen_image_edit_2511_bf16.safetensors
CLIPLoader qwen_2.5_vl_7b_fp8_scaled.safetensors, type qwen_image
VAELoader qwen_image_vae.safetensors
LoraLoaderModelOnly VNCCS_PoseStudioQIE2511_V2.safetensors at 1.0

25 nodes. Small for what it does.

Why a pose LoRA rather than a ControlNet

A pose ControlNet takes a skeleton image as conditioning. A pose LoRA teaches the model to respond to pose instructions in its own latent space.

The LoRA approach has an advantage for character consistency: the pose is applied on top of an edit rather than replacing the model's structure. Since the base is an edit model (2511), you can supply a character reference and a pose simultaneously, and the model holds identity while changing the pose.

A ControlNet would fight that — it constrains structure, which competes with the character reference.

The _V2 in the filename

VNCCS_PoseStudioQIE2511_V2 — version 2, targeted at Query Image Edit 2511.

Version matters here because the LoRA is trained against a specific base. V2 targets 2511; the 2509 or 2601 versions would not be interchangeable. If you upgrade the base model, check whether the LoRA has a matching release.

What 3D-in-ComfyUI buys

Doing pose inside the graph means:

  • No export/import step — the pose is a node value, not a file
  • Reproducibility — the pose is saved in the workflow, so the workflow alone is sufficient
  • Fast iteration — change the pose, re-run, no round-trip

The cost is that the in-graph pose tool is less capable than a full 3D application. If you need precise finger posing or complex object interaction, you still want the external tool.

Lighting control

The workflow title mentions lighting alongside pose. Lighting control in an edit model means instructing the model on light direction and quality while keeping the subject.

That is harder than pose control because lighting is global — it affects every surface. A pose change is local; a lighting change rewrites the shading of the entire image, which is where drift appears.

Where this breaks

  1. Pose plus character reference compete. The more extreme the pose, the more the model has to invent anatomy, and the more identity drifts.
  2. Lighting changes cause global drift. Expect to need the anti-drift configuration for subtle lighting work.
  3. The LoRA is base-version-specific. V2 for 2511 only.
  4. 25 nodes has minimal validation.
  5. Hands and fingers are outside what a pose LoRA handles well.

Practical approach

Start with a modest pose change — a small rotation, a raised arm — on a character you have a clear reference for. Confirm identity holds. Then increase the pose delta until it breaks; that break point tells you what this workflow can do for your characters.

Workflows

One graph, published on RunningHub with invite code rh-v1111: character-consistent pose and lighting control with VNCCS Pose Studio and Qwen 2511.

Files

https://huggingface.co/AI-KSK/ksk-video-175-comfyui-workflows

Notes

  • Workflow file only. Model weights are not included; each keeps its own licence.
  • Tutorial export, not a product.
  • Node count and loader values quoted above are read from the published JSON file.
  • Independent AI-KSK release. Not affiliated with the model authors or RunningHub.

Optional API platform - 88API (affiliate link)

Explore 88API for access to multiple AI model APIs. Available models, routes and prices are listed on the platform; Alipay top-ups are supported.

https://88api.ai/sign-up?aff=usx3

Disclosure: I may earn a commission from eligible purchases after you register through this referral link.

Top comments (0)