Question 1

What is AI Pose to Image?

Accepted Answer

Drop a reference photo of a person or character in a pose you want to copy. The AI extracts the pose skeleton (limb angles, hand positions, body orientation) and renders an entirely new character in that exact pose, in whatever style and setting you describe in the prompt.

Question 2

How does the pose extraction work?

Accepted Answer

OpenPose detector pulls keypoints — head, shoulders, elbows, wrists, hips, knees, ankles, plus hands and face if visible — from your reference image. Those keypoints become a skeleton diagram that the OpenPose ControlNet on SDXL uses as a structural constraint while the diffusion model fills in everything else.

Question 3

When should I use this vs /image/sketch-to-image/?

Accepted Answer

Pose-to-image preserves BODY POSITION. Sketch-to-image preserves OUTLINE / linework. Use pose if you want the same gesture but a different character / outfit / scene. Use sketch if you drew a layout and want it rendered as-is.

Question 4

What kind of reference images work best?

Accepted Answer

Full-body or three-quarter shots with clear visibility of joints. Photos work great. Stylized art works if the figure is anatomically clear. Avoid extreme close-ups (no joint info), heavily cropped figures, or images where the pose is occluded.

Question 5

Is it free?

Accepted Answer

Yes. Each render costs about 1,200 tokens (1,000 base + 20% ControlNet conditioning surcharge). Signed-in users get 5,000 free tokens daily — covers ~4 poses per day at the free tier.

Question 6

Can I copy a pose from a copyrighted image?

Accepted Answer

Pose data alone — joint positions in space — is not copyrightable in most jurisdictions. The output character is your own creation. Avoid copying not just the pose but also the costume / face / specific styling of someone else's character if commercial use is the goal.

Question 7

Does it work for animals or non-human subjects?

Accepted Answer

OpenPose is trained on human anatomy. Animal photos confuse the detector — limbs map poorly to the human keypoint scheme. For animals, /image/sketch-to-image/ with a hand-drawn pose works better.

Question 8

Can I tweak the influence of the pose on the final render?

Accepted Answer

Yes — the "ControlNet strength" slider (default 0.7) controls how strictly the AI follows the pose. 1.0 = exact pose, 0.4 = loose interpretation. Lower it if the output looks stiff; raise it if the pose drifts.

Question 9

What output sizes are supported?

Accepted Answer

512×512 default. Standard SDXL aspect ratios (768×1024, 1024×768, 1024×1024) all work. Larger sizes consume more VRAM and tokens; the H200 supports up to 1024×1024 comfortably.

Question 10

How does it compare to ComfyUI / Automatic1111 with OpenPose?

Accepted Answer

Same model, same quality. ComfyUI / A1111 require local GPU + setup; we run it on shared infrastructure free up to the daily pool, paid beyond that. Output is identical — same SDXL + same xinsir/controlnet-openpose-sdxl-1.0 weights.

Question 11

Are the uploaded reference images stored?

Accepted Answer

Reference images are processed immediately, used for the pose extraction, and discarded after generation completes. Only your generated output + the prompt are kept on /account/?tab=history. Never used for training. Privacy policy at /privacy/.

Question 12

Is there an API?

Accepted Answer

Yes — POST multipart to /v1/image/generate/ with model=sdxl, prompt, control_image (file), control_type=pose. Returns the image URL + share token. Bearer auth, 10K tokens/month free. Curl example at /api/.

Canny → Image (ControlNet)

Kết quả

Canny → Image (ControlNet)

Kết quả

Lấy 10.000 token miễn phí

Chờ đã — Cầm 10K token miễn phí!

Muốn thêm nữa không?