Skip to main content
A pose is a body position drawn on two faceless grey mannequins, a woman and a man, from the front and from other angles. Write its @handle after a character in a prompt and the character takes that position: the mannequin’s picture goes with the mention as a reference image, and the pose’s description is added to the prompt. The character keeps their own face, build and clothes.
Every pose in the library is one a real body can do: each one is checked for joints, balance and what holds the body up before it joins. You’ll find everyday poses, athletic ones (splits, handstands, yoga) and glamour poses; contortion and sex positions are not in the library. The library is shared, so every pose works for every user, and you can design new poses that join it.

The pose library

Browse every pose by shelf, see its handle and its pictures from every angle.
Available on every paid plan, like the rest of the API. Over MCP, your agent finds poses with search_poses and get_pose, designs new ones with design_pose and check_pose_design, makes new views with make_pose_view and check_pose_view, and poses a character with the same @mention in generate_image or generate_video.

​
The two mannequins

Every pose is drawn twice: on a woman and on a man, both faceless and grey, with adult proportions. A mention sends the one that matches the character’s gender, so a female character gets the woman and a male character the man. When the gender isn’t known, the woman is sent. Each figure is drawn from the front and from other angles: three-quarter, side, from above and from below, and from behind on poses that are about the back. A designed pose has the front and three-quarter views. Anyone can add a view from another place around the figure, and it joins the library for everyone: see Views from other angles.

​
Posing a character

  • Write the character, then the pose: "@rosa in @sitting_knee_up_pose on a beach at sunset". A pose goes with the nearest character written before it, so with two people, write a pose after each of them. Up to 4 poses per generation; their pictures count as reference images.
  • Add a view to the mention to choose the angle: @sitting_knee_up_pose.three_quarter. front is the default. Every picture in a pose’s pictures carries its own mention, so the pictures show the views you can ask for. A view the pose doesn’t have yet uses its nearest picture; make the view for the exact angle.
  • Only the position carries over: where the torso, head, arms, hands, legs and feet are, which way the body faces, and what touches the floor or a seat. The mannequin’s build, blank face, lines and grey colour never do.
  • Dress the character too with an outfit: "@barry wearing @camel_belted_overcoat_outfit in @sitting_knee_up_pose".
  • A pose works on the models that take characters (see supported_models on GET /v1/characters/{handle}), in image and video prompts. The mannequin’s picture is a reference image, like a character’s picture.

​
Find a pose

GET /v1/poses searches the library. Filter with q (words that must all appear in the name, handle, description or tags), shelf and fit, and page with limit (20 by default, at most 50) and offset: total counts every match and has_more says whether there are more. Each pose also says how hard it is (difficulty: everyday or athletic), whether a person can hold it still for a photo (holdable; false for a moment in motion, like a jump), what holds the body up (support), the props it needs, and fit: who it is made for (women, men or unisex). fit=women (or men) lists the poses made for women (or men) and the unisex ones. Every pose works for any character, because the mannequin follows the character’s gender.

​
One pose in full

GET /v1/poses/{handle} returns { "pose": { … } }: the description word for word (the text a mention adds to the prompt) and every mannequin picture by figure (woman, man) and place, each with url, thumbnail_url, width, height, its place on the globe (azimuth, elevation and a label such as "Back left · from above") and the mention that asks for it. The handle works with or without the @. A list carries each pose’s own pictures and a view_count. This answer carries every view members made too, each with view: true and the username of the member who made it in made_by.

​
Views from other angles

A pose can be seen from any place on the Camera Angles globe. When the view you need isn’t in a pose’s pictures yet, ask for it: the mannequin’s front picture is turned to that place, a check confirms it is still the same pose and where the camera landed, and the view joins the library for everyone, credited to you. There are no private views.
figure is the mannequin to turn (woman or man; a mention sends the one that matches the character’s gender). azimuth is where the camera stands around the figure, and elevation how high: 0 eye level, 30 from above, 60 from high above. In front, .above asks for elevation 30 and .top for 60. The other raised places are named by a code: @seiza_pose.az135_el30. Each view’s mention in the pose’s pictures is always the right one to write.
  • POST /v1/poses/{handle}/views answers 200 with status: "made" and the picture when the place already has a view, for free. Otherwise it answers 202 with status: "making" and the job.
  • Follow the job with GET /v1/poses/views/{id} every 10 to 15 seconds. A view usually takes 30 to 90 seconds, through queued and working to done, with the view in picture. A missed job didn’t come out right after two tries, and a failed one says what happened in message; neither is charged.
  • One job per place: when someone is already making the view you ask for, you join their job (joined: true) and pay nothing.
  • A try that lands on another empty place is kept there for everyone, free, in landed_picture.
  • 40 credits per view, charged only when it lands where you asked; the second try is free. Up to 50 new views a day (429 with error_code: daily_cap past that); 402 with error_code: insufficient_credits when your credits can’t cover it.

​
Design a pose

When the library doesn’t have the pose you need, design it: describe it in words, or start from a library pose and say what to change. Every pose made joins the shared library for everyone, with you as its creator; there are no private poses.
  • words describe the pose head to toe, in up to 600 characters. With parent_pose_id (a library pose’s handle or id), they say what to change: "the left arm raised higher".
  • When library poses already look like the request, POST /v1/poses/designs answers 200 with them in matches and starts nothing: use one, or send the request again with "confirm": true.
  • Otherwise it answers 201 with the design. Follow it with GET /v1/poses/designs/{id} every 20 to 30 seconds. It takes about 3 to 6 minutes, through queued and working (stage_label says what it is doing) to done, with the new pose and its handle. A refused design says why in message and offers a doable alternative in suggestion; a failed one says what happened.
  • Every pose must be one a real body can do: poses a body can’t hold, contortion, sex positions, minors, real people and violence are refused.
  • 100 credits per pose, charged only when it is made: a refused or failed design is free. Paid plans only, up to 20 designs a day (429 past that); 402 with error_code: insufficient_credits when your credits can’t cover it.

​
Over MCP

Connect the MCP server and ask in plain words: “Put @rosa in a sitting pose with one knee up, on a beach at sunset.” The agent calls search_poses (by words, shelf or fit) and gets a short list of poses, each with its mention, shelf, difficulty, a line of description and its front picture. get_pose reads one pose in full, with every picture and how to write it in a prompt. Then the agent writes "@rosa in @sitting_knee_up_pose …" in generate_image or generate_video, with an angle when you ask for one. When nothing fits, design_pose designs a new pose (showing the library poses that already look like it first), and check_pose_design follows it until its handle is ready. When a pose lacks the angle you want, make_pose_view makes that view for everyone and check_pose_view follows it until its mention is ready: “Show @rosa in the seiza pose from behind and a little above.”

​
On the web

Browse the pose library by shelf. Pose on your character’s page, or in Try-on, puts the character in a library pose and keeps the result as one of the character’s pictures, tagged with the pose: @mina.sitting_knee_up_pose then casts that exact picture, like any of a character’s tagged pictures.

​
Pricing and limits

  • Listing and reading poses is free: 300 list calls and 200 detail calls per day per account, at most 30 of each a minute. The API and MCP share these limits; past one, the answer is 429 with Retry-After.
  • Using a pose adds no fee: you pay the model’s normal price for the image or video.
  • Designing a pose costs 100 credits, charged only when the pose is made; paid plans, up to 20 designs a day.
  • A new view costs 40 credits, charged only when it lands where you asked; joining a view someone is already making is free. Up to 50 new views a day. Following a view job is free.