Short answer: use a T-pose. It is what automatic rigging, motion-capture retargeting and every engine humanoid system expect, and the shoulder-deformation argument for an A-pose only pays off on a rig you are hand-weighting anyway.
What is a T-pose? A character standing with its arms straight out to the sides, level with the shoulders, so the silhouette reads as a capital T. What is an A-pose? The same stance with the arms lowered to about 45°, so it reads as an A. Both are bind poses: the neutral position a character is modelled and rigged in, never one it is seen in.
| T-pose | A-pose | |
|---|---|---|
| Arms | Straight out, horizontal | Angled down, roughly 45° |
| Automatic rigging | Expected | Often misjudged |
| Motion-capture retargeting | The standard | Needs conversion |
| Shoulder deformation | Stretched at rest | Closer to neutral |
| Engine humanoid systems | Assumed | Usually remapped |
| Mirroring and bone placement | Aligned to world axes | Off-axis |
This matters for [AI rigging](/kb/generate-3d/ai-rigging/) specifically: an A-pose looks riggable and is not. The joints land in the wrong place, and the result is a skeleton that appears fine until the first animation plays.
Characters are modelled in a neutral pose so they can be rigged and animated. Two conventions dominate.
T-pose
Arms straight out to the sides, horizontal. Legs straight, slightly apart. Facing forward.
- Easier to rig. Bones align with world axes, making placement and mirroring straightforward.
- The de facto standard for automatic tools, motion capture retargeting and engine humanoid systems.
- Worse for shoulder deformation. The shoulder is at an extreme of its range, so weights tuned there behave poorly in the middle of the range.
A-pose
Arms angled down, roughly forty-five degrees. Closer to how a body rests.
- Better shoulder deformation. The joint sits nearer the middle of its range, so skinning is accurate across more of it.
- More natural for sculpting, since the musculature is not stretched.
- Harder to rig automatically, because nothing aligns to an axis.
Which to use
If you are hand-rigging a hero character, A-pose is generally the better choice for deformation quality. If you are relying on automatic rigging, retargeting from a motion library, or an engine's humanoid system, T-pose is what those tools expect.
Why auto-rigging requires T-pose
Automatic rigging matches your mesh against an expected layout to place joints. A T-pose puts limbs in known, separated, axis-aligned positions. An A-pose moves every arm joint away from where the matcher looks, so joints land badly — most visibly at the shoulders — even though the model looks perfectly riggable.
That failure is silent rather than loud, which is why it surprises people: the rig is produced, and it is wrong.
Generating a character in T-pose
Far easier than re-posing one afterwards, because re-posing needs a rig, which is what you were trying to get. Ask for the pose in the prompt, or supply T-pose reference images. From concept art, Isolate Parts redraws a character in a T-pose with empty hands before it becomes a model, and AI rigging then places the skeleton on it.
Other conventions
Hands flat with palms down, fingers slightly spread, feet flat and parallel, head level, mouth closed and eyes open. Symmetry matters — an asymmetric neutral pose breaks mirroring.