Anima Control Pose Preview-2: Native Pose Control for ComfyUI

news

Anima Control Pose Preview-2 brings pose-controlled image generation to Anima with native ComfyUI custom nodes, multi-resolution support at 512/768/1024, and improved pose accuracy

A new pose control adapter for the Anima v1.0 image model has been released — Anima Control Pose Preview-2 brings skeleton-based pose conditioning to Anima with native ComfyUI custom nodes, multi-resolution support, and substantially improved pose-following accuracy.

What It Does

Anima Control Pose is a channel-concat control-LoRA that conditions Anima image generation on a skeleton pose map. Given a reference photo, the ComfyUI node detects the pose and renders a skeleton, which the model follows during generation — enabling precise control over character positioning, gestures, and body language in AI-generated images.

Preview-2 is a major update over Preview-1 with:

  • Multi-resolution support: 512, 768, and 1024 px (Preview-1 was 512-only)
  • Larger training corpus — cleaner anatomy and steadier poses
  • New ComfyUI node "Anima Pose Control" — auto-detects pose from a reference photo with multiple skeleton rendering styles (thin, thick, puppet, heatmap, with/without hands and face)
  • Body-PCK@0.1: ~0.83 at 1024 resolution (vs ~0.59 for Preview-1 at 512)

Method

The control adapter uses a zero-initialized ControlEmbedder that produces control tokens added to the frozen base patch-embed output, plus a rank-16 low-rank adapter on the transformer blocks. The base Anima model, text encoder, and VAE remain frozen.

At strength = 0, the adapter is exactly the base model — the control contribution grows only as it earns loss during training.

SettingValue
Resolution512 + 768 + 1024, aspect-ratio bucketed
Adapter rank16
Control dropout0.1
Precisionbf16

Availability

Anima Control Pose Preview-2 is available on Hugging Face under a non-commercial license:

  • HF Repo: Claquasse/Anima-Control-Pose
  • Main file: anima_pose_preview2.safetensors (place in ComfyUI/models/loras/)
  • Custom nodes: Included in the repo under comfyui/
  • Base model: Anima v1.0 by Circlestone Labs (required)

Usage in ComfyUI

Preview-2 uses two small custom nodes: Anima Control Apply (AnimaControlApply) applies the adapter, and Anima Pose Control (AnimaPoseControl) detects the pose from a photo and renders the skeleton.

Install

  1. Download anima_pose_preview2.safetensors into ComfyUI/models/loras/.
  2. Copy both folders from comfyui/ in the HF repo into ComfyUI/custom_nodes/:
    • anima_control_lora/
    • ComfyUI-anima-pose-control/
  3. Restart ComfyUI. ComfyUI-Manager auto-installs the second node's requirements.txt; otherwise run: 
    pip install -r ComfyUI-anima-pose-control/requirements.txt
    (rtmlib, opencv-python, onnxruntime, numpy; torch and Pillow come with ComfyUI.)
  4. Load a workflow from the menu (see below).

The safetensors file holds both the low-rank adapter (lora.* keys) and the control embedder (control_embedder.* keys). A single LoraLoaderModelOnly reads the LoRA weights, while Anima Control Apply reads the control embedder — both point at the same file.

Workflows

Four .json workflows are included in the HF repo:

  • pose_control_demo.json (easiest) — control vs no-control side by side
  • pose_control.json — simple single-generation workflow
  • pose_control_edit.json — single node with skeleton style picker
  • pose_control_compare.json — one pose rendered across every skeleton style at once

Strength

0.0 produces the base model with no control; 1.0 follows the skeleton (range 0–2). Higher values track the pose more closely but can reduce image quality.

Pose detector (first run)

The Anima Pose Control node needs a pose detector (rtmlib). On first use it downloads two ONNX files (~316 MB total) from the HF repo's detector/ folder and caches them under ~/.cache/rtmlib/hub/checkpoints/; after that it runs offline.

If you see urllib ... getaddrinfo failed, the machine couldn't reach the download host. Download the two ONNX files manually from the HF repo's Files tab and place them into that cache folder (create it if missing), then restart ComfyUI:

  • detector/yolox_m_8xb8-300e_humanart-c2c7a14a.onnx
  • detector/rtmw-dw-x-l_simcc-cocktail14_270e-256x192_20231122.onnx

The author notes that Preview-2 is still experimental — while significantly improved, it can still miss poses or produce artifacts on complex poses. A Preview-3 is in development with a focus on better skeleton representation and a purpose-built anime pose detector.

Comments

Sign in with GitHub to join the discussion.

Loading comments…
Anima Control Pose Preview-2: Native Pose Control for ComfyUI | ComfyUI Wiki