AI pose generator
Write a pose as a physical action—not a mood—and generate one square anime concept. Use the result to test silhouette, balance, gesture, and staging before you redraw or approve a production pose.
The live generator is in the first viewport. Name one original adult character, the action phase, line of action, weight-bearing foot or hand, free limbs, camera view, crop, clothing, and ground contact. Each run returns one square raster image on this page.
Higher-detail generation for paid plans
Example outputs
What one focused input can produce
Standard image
Advanced image
Actual output
one newly generated 1024×1024 flattened square raster image
Useful for
gesture exploration, character key-pose ideas, storyboard references, action thumbnails, illustration planning, and pose discussions
Not provided
pose skeleton input, OpenPose maps, exact character identity, rigged models, editable joints, transparent sprites, animation, batch views, or anatomy guarantees
Product and guidance reviewed 2026-07-22
Capability wording was checked against the current text-only image route, sign-in return, one-image 1024×1024 result, provider and credit handling, and in-page failure state. The page does not claim pose-map, reference-image, rig, seed, or identity-lock controls that the interface does not expose.
“Dynamic pose” leaves the important choices unresolved. A better brief says what force acts on the body, where the weight goes, which joints compress or extend, and what the viewer must read before costume detail.
01
A jump contains preparation, takeoff, suspension, landing, and recovery. Each phase has different joint angles and balance. Name the instant you need so the model is not forced to combine incompatible body mechanics into one figure.
02
State which foot carries weight, which hand grips a prop, and where the body touches the ground or wall. Contact points make the physical problem testable; accessories can be added after the action reads.
03
Shrink the result until fingers and facial details disappear. If the action still reads from the torso angle, limb spacing, and support shape, the pose has a useful graphic foundation.
01
Write what the character is doing, why the moment matters, and whether the pose comes before, during, or after the main force.
02
Name the weight-bearing limb, external force, counterweight, contact points, and direction of travel.
03
Choose front, side, rear, or three-quarter view; state camera height and keep required hands, feet, and props inside frame.
04
Submit the focused brief and wait for one square raster result. Treat it as a proposal, not measured anatomy.
05
Check balance, joints, hands, feet, costume tension, and context; then redraw or annotate the selected idea for production.
“Adult rooftop courier at the recovery phase of a landing; left leg compressed and weight-bearing, right foot lightly behind, forward torso, arms counterbalancing, full body in frame, three-quarter view, clear ground shadow.”
Why this pose brief is reviewable: The prompt identifies phase, support, counterweight, crop, and camera. Those are visible decisions that can be accepted or corrected after generation.
Use the section that matches the current decision: action phase, line of action, balance, hands and feet, camera, clothing, anatomy review, variation, rights, or production handoff.
Begin with a verb that has a physical object and a precise phase: catching a falling case just before impact, pushing a jammed door after the first step, settling into a landing, or turning after hearing a sound. “Running, jumping, fighting, dynamic” combines several moments and gives the generator no single arrangement to solve. A pose is one instant selected from an action. The phase determines whether muscles compress, extend, suspend, or recover, and it changes where the center of mass must sit relative to the support.
Write one sentence of story pressure. A courier lands quietly because they do not want to be heard; a mechanic pulls because the door is stuck; a dancer reaches because the partner has moved out of frame. This pressure creates visible choices without requiring a long biography. It can affect gaze, speed, tension, and asymmetry. If the story note cannot alter the body, leave it out of the prompt and keep the limited text budget for evidence the viewer can actually inspect.
This page accepts text only. You cannot drag joints, upload OpenPose maps, read joint coordinates, lock bone lengths, or export a pose rig.
Hands, feet, shoulders, hips, foreshortening, overlaps, and hidden limbs can fail even when the overall gesture looks energetic. Inspect before reuse.
The result is a flattened square still. It does not provide in-between frames, temporal continuity, a sprite sheet, transparent layers, or motion data.
Repeating hair, clothing, palette, and prop improves direction but does not guarantee the same face, proportions, costume construction, or accessory placement across runs.
Reviewed 2026-07-22 · Anime Maker product and editorial team
Page illustrations are editorial examples, not live account task logs. Judge the tool by the result shown in your own output panel.
It outputs one newly generated 1024×1024 square raster anime image. It does not output joint coordinates, a pose map, a rig, a layered file, or an animation.
Not on this page. The current interface is text-only. Describe the action, phase, support, view, crop, and contact points instead of implying that an unprovided image will be followed.
Choose an action phase, establish a line of action, separate the limbs in silhouette, show where weight is supported, and let the torso and free limbs counterbalance the main force. “Dynamic” alone is not enough information.
Keep the character original, define the contact points, generate one frame, and redraw anything that fails the anatomy or storytelling check.
Generate an anime poseRepresentative single-output proof: one square image shows a complete landing pose, support foot, counterbalancing arm, cast shadow, and readable silhouette without implying a rig, pose map, or multi-view sheet.
“Adult greenhouse worker reaching above shoulder height, side view, left foot planted under the hip, right heel lifted, pelvis shifts toward support, ribcage extends, free hand steadies a vine, fingers visible.”
Why this pose brief is reviewable: The quiet action still needs weight distribution. Naming the planted foot and heel lift prevents a generic upright mannequin with a raised arm.
“Adult workshop mechanic pulling a stuck sliding door, both hands on the handle, staggered stance, rear leg braced, torso rotates away, elbows flex under tension, coat and bag lag behind the motion, full body low three-quarter view.”
Why this pose brief is reviewable: The door supplies an external force. Foot placement, torso rotation, grip, and trailing fabric should all agree with that force.
Separate the pose request from finish language. First solve action, phase, support, force, camera, and crop. Then add the original character anchors, clothing, environment, light, and rendering. If the first generation fails, this order helps diagnosis: you can tell whether the body mechanics were vague or whether costume and scenery simply overwhelmed them. Keep a copy of the short mechanics-only brief so later variations can preserve the same physical question.
The line of action is a design abstraction that describes the dominant flow through the figure. It may run from the head through the spine into a support leg, or connect a reaching hand through the torso to the opposite foot. It is not a literal bone, and it should not force every limb onto one curve. Use one broad direction for the main energy, then let the arms and free leg support, resist, or counterbalance it. A clear primary flow is more useful than asking every joint to bend dramatically.
Balance can be reviewed with a simple question: where would the body fall if the support disappeared? In a quiet standing pose, the center of mass usually needs to project over the base formed by the feet. In running or falling, it may move outside that base because momentum carries the body toward a new contact. State whether the pose is stable, temporarily unstable, or actively recovering. That single distinction helps separate an intentional lean from a figure that appears to tip accidentally.
Asymmetry often makes a pose feel observed rather than assembled. One shoulder can rise while the opposite hip carries weight; one arm acts while the other stabilizes; the head can resist the torso turn. Do not add asymmetry at random. Tie it to load, attention, or balance. After generation, cover costume detail mentally and look only at head, ribcage, pelvis, hands, and feet. If their relationship does not explain the action, rendering polish is not a reason to accept the pose.
Hands and feet become easier to judge when they have a job. “Hand on railing” should specify whether the palm bears weight, fingers hook over the top, or fingertips only guide movement. “Foot on ground” should identify the support foot, heel state, direction, and whether the shoe slides, pushes, or lands. These details do not guarantee correct anatomy, but they create observable criteria. A hand that floats beside a handle can then be rejected for a clear reason rather than described as vaguely strange.
A prop introduces mass, leverage, and resistance. A heavy case pulls the shoulder down and may move the pelvis in the opposite direction. A long staff changes spacing and can create a strong diagonal. A bicycle, instrument, or tool also establishes scale and contact geometry that the image has to respect. Keep one primary prop in early tests. Several objects create more opportunities for missing grips, fused forms, inconsistent scale, and accidental redesign.
Clothing should respond to the body after the pose works. Compression folds collect at bending elbows, knees, and hips; tension lines can follow a pulling arm; loose fabric and straps may lag behind a turn. Do not treat folds as anatomy evidence because a polished coat can hide an impossible shoulder. Review the unclothed construction conceptually first, then ask whether fabric, hair, and accessories reinforce the action without obscuring the support limbs or silhouette gaps.
A side view can clarify stride length and forward lean; a front view can expose stance width and symmetry; three-quarter views show depth but create more overlaps; high and low viewpoints can support story power while increasing foreshortening risk. Pick the camera because it reveals the pose question, not because it sounds cinematic. If hand placement is critical, do not hide both hands behind the torso. If foot contact matters, leave the floor visible and avoid a crop that ends at the ankles.
Write the crop explicitly: full body with head, both hands, and both shoes inside the frame; knee-up with the gripping hand visible; or waist-up focused on a turn and expression. A square output has finite room. Full-body action needs more surrounding margin and a simpler background than a portrait. Combining an extreme low angle, very close camera, wide environmental view, and full-body requirement creates competing scale instructions that may lead to cropped limbs or weak staging.
Foreshortening should have a purpose. A hand near the camera can emphasize urgency, but it can also become the largest shape and conceal the face or torso. Name which body part enters the foreground, what it overlaps, and which features must remain readable. Treat perspective language as descriptive because the page does not expose lens, sensor, depth map, or camera calibration. If exact spatial measurement matters, build the pose in a 3D tool or draw a construction grid.
Start with count and connection: one head, one torso, two arms, two hands, two legs, and two feet unless the original character concept intentionally differs. Trace each limb through overlaps. Check whether shoulders connect to the ribcage, hips support the legs, elbows and knees bend in plausible directions, and hands actually attach at the wrists. Extra fingers are visible failures, but missing joints, duplicated forearms, or a leg that changes thickness through an overlap can be more consequential for a redraw.
Next check weight and ground. Look for a coherent support foot, compressed or extended joints that match the action phase, and a shadow or contact cue consistent with the floor. A figure may have attractive anatomy in isolation but still float, slide, or lean without support. Compare the pelvis and ribcage to the direction of force. If a character pushes left while every counterweight also moves left, the pose may not communicate resistance even when individual limbs look plausible.
Finally check proportion and context. Perspective can legitimately enlarge a near hand or shorten a limb, but the change should agree with the camera and overlaps. Costume pieces should attach consistently; props should keep scale; hair should emerge from a coherent skull; and the background should not cut confusing tangents through joints. Mark problems on a copy and decide whether to regenerate, redraw, photobash for internal planning, or discard. Do not quietly pass structural errors into later production.
When the overall action reads but one contact fails, keep the successful phase, camera, and character anchors. Rewrite only the support or grip instruction. When the silhouette is crowded, change limb spacing or view before rewriting the costume. Controlled revision makes outputs comparable. Replacing the entire prompt after every result may create novelty, but it does not teach you which instruction affected balance, framing, or continuity.
For a set of key poses, write a small fixed character block and a separate action block. The fixed block can include adult age, silhouette, hairstyle, palette, one accessory, and costume shape. The action block changes phase, force, support, view, and crop. This can improve art-direction consistency, but it is not an identity lock. Select one approved image as a visual anchor and expect to redraw faces, proportions, clothing construction, and props across the final set.
Keep a review table with prompt version, intended change, result path, pass or fail, and the specific reason. Useful reasons include readable action, support foot missing, hand fused to prop, camera hid the contact, face drifted, or background tangent weakened silhouette. Avoid notes such as “bad” or “more dynamic.” Specific observations shorten the next brief and create a traceable record for collaborators who need to understand why one pose was selected.
Place the selected result beside your own gesture thumbnail rather than tracing it without review. Reconstruct the line of action, ribcage, pelvis, support, and contact points. Correct perspective and anatomy with observation, life-drawing knowledge, authorized references, or a 3D block-in. Then design costume folds and details around the corrected structure. This keeps the generator in an idea role and makes the final drawing depend on human decisions you can explain.
If the pose enters a storyboard, check whether the previous and next shots make the selected phase understandable. A landing pose needs a plausible approach and recovery; a handoff needs matching screen direction and prop position; an emotional turn needs eye line and staging that connect to the cut. One appealing still cannot solve continuity by itself. Record which details are critical and carry them into neighboring frames deliberately.
Export the generated square as a reference file, not a hidden source of truth. Preserve the prompt, generation date, provider context available to you, selected result, annotations, and final redraw. If the production has client, employer, school, competition, or platform rules about generative tools, check them before submission. A polished redraw does not remove obligations attached to how the concept was created or to unauthorized source material.
Describe an original character rather than requesting a protected franchise figure or the style of a named living artist. Roles, materials, eras, palettes, line qualities, and lighting are more useful production language than imitation. Keep records of source references and permissions. A pose itself may be common, but a specific copyrighted image, character design, costume, or photograph can still raise rights and consent questions.
Do not use a generated pose as evidence of a real event, injury, capability, or person. The figure and mechanics may be invented. For medical, ergonomic, athletic, safety, or legal decisions, use qualified sources and people rather than an anime image generator. If a production depicts dangerous action, present it as fiction and do not treat visual plausibility as instruction for performing the act.
Before release, review the final crop for accidental text, logos, signatures, culturally specific symbols, sexualized age ambiguity, and unintended resemblance. Add appropriate alt text that describes the visible action and context rather than listing every costume detail. If the image is decorative, use the destination platform’s empty-alt convention. Revisit rights and disclosure when a private concept moves into advertising, merchandise, education, or another public context.
The page does not expose a calibrated lens or 3D scene. Extreme foreshortening and low angles may change body proportions or hide important contact points.
A run requires sign-in, available credits or access, and a working provider. Safety rejection, timeout, or service failure may require a revised request or another attempt.
No identity lock is offered. Reuse a short anchor list for hairstyle, palette, costume silhouette, and one prop, then manually correct drift when continuity matters.
Ask for full body, head and both shoes inside frame, generous margin, and hands visible. A square crop still has limited space, so avoid combining an extreme close camera with a full-body requirement.
Use it as an idea to inspect, not anatomical authority. Compare balance and joints with trusted life drawing, your own observation, or a qualified reviewer before relying on it.
The supplied examples use original adults. For age-sensitive subjects, keep the context appropriate, avoid sexualized framing, and follow the product safety rules and local requirements.
Read the displayed error. Sign-in, insufficient credit, unavailable provider, safety rejection, and timeout are different failures. Preserve the prompt in the URL during authentication, then retry only after the relevant issue is resolved.
