AI pose transfer — copy a pose from reference to generation
Describing a specific pose in a prompt is unreliable — 'arms crossed, slight lean, weight on left leg' renders as a different pose every time. Pose transfer fixes this by extracting the skeletal keypoints from a reference image with DWPose, then conditioning the generation on those exact keypoints so the output character matches the reference pose precisely.
Built on the DWPose preprocessor inside the pipeline — DWPose detects 18+ body keypoints and 21 hand keypoints per side, capturing finger position and torso rotation accurately. The conditioning is applied at inference time so the generation model produces a character matching both your prompt and the reference pose.
DWPose keypoint extraction with 18+ body points and 21 hand keypoints per side — more accurate than OpenPose, captures finger position and torso rotation precisely.
How It Works
- 1
Upload a reference image — any photo or generation with a clear human pose.
- 2
DWPose extracts the skeletal keypoints automatically (no manual rigging required).
- 3
Write your generation prompt describing the character — pose vocabulary is unnecessary.
- 4
The model generates a new character matching your prompt with the reference's exact pose.
Why nocensor.ai?
Pipeline Builder
Chain 9 stages into one submission with bundle discounts — Generate, Face Swap, Animate, Restyle, Upscale and more.
No Restrictions
No content filters. No topic blocks. No suspension risk for adult content.
What You Get
- Copy a specific pose from a reference photo without manual rigging
- Generate the same character in multiple poses by swapping the reference
- Combine pose transfer with Character LoRA to lock both identity and pose
- Apply pose transfer inside the Pipeline Builder for chained workflows
Example Prompts
Copy any prompt below and paste it directly into the generator.
- Photoreal brunette woman, evening dress, studio lighting — pose from reference
- Anime character, school uniform, classroom — exact pose from reference image
- Illustrated fantasy elf, ornate gown — match reference pose precisely
- Photoreal model, lingerie, bedroom setting — copy pose from upload
- Anime girl, beach setting, swimsuit — preserve reference body angle
Frequently Asked Questions
What is AI pose transfer?
A workflow that extracts skeletal keypoints from a reference image with DWPose, then conditions a new AI generation on those exact keypoints — producing a character that matches your prompt with the reference's exact pose.
How is this different from describing a pose in the prompt?
Text descriptions of pose are unreliable — the model interprets 'arms crossed' or 'weight on left leg' differently every generation. DWPose extracts the actual joint positions so the output matches the reference precisely.
Will the character look like the reference person?
No — only the pose transfers. The character's identity comes from your prompt or your trained Character LoRA. The reference image only contributes the skeletal pose.
How accurate is DWPose compared to OpenPose?
DWPose detects 18+ body keypoints plus 21 hand keypoints per side, capturing finger position and torso rotation more accurately than OpenPose, which often misses subtle anatomy and finger detail.
Learn more in the nocensor.ai blog →