Distorted hands, warped faces, and unexpected poses are some of the most common issues in character generation. This article explains why these problems happen and how to get cleaner, more consistent character results in Meshy.
Pose Input Quality
The quality of your input has the biggest impact on the quality of your character's pose. Meshy interprets whatever pose, silhouette, and proportions are visible in your reference image or implied by your prompt, so ambiguous or low-quality inputs lead to ambiguous or distorted results.
Use a clear, well-lit reference image with the character fully visible and not cropped at the limbs.
Avoid busy backgrounds, overlapping objects, or other people in the frame — these can confuse pose and limb detection.
A front-facing, full-body reference image generally produces more reliable results than an angled or partial view.
Tip: Use Meshy's Image-to-Image feature to clean or optimize your reference image before sending it to Image to 3D. A better-framed, higher-quality input directly improves hand and face results — especially for characters with complex details or accessories.
Using A-Pose, T-Pose, or Custom Pose
If you need a character in a specific, predictable pose (for example, for rigging), don't rely on describing the pose in a text prompt — instead use Meshy's built-in Pose Control in the Image to 3D workspace, which supports A-Pose, T-Pose, and fully Custom Pose adjustment via joint controls.
A-Pose and T-Pose toggles generate a standardized, rig-ready pose with minimal self-intersection.
Custom Pose lets you manually adjust joints before generating, which is useful when neither preset matches what you need.
Simply typing "T-pose" into a Text to Model prompt is not a reliable way to control pose — generate a reference image first, then use Image to 3D with the pose toggle enabled.
Common Hand and Face Issues
Hands and faces are the hardest parts of a character for any 3D generation model to reconstruct accurately, since they involve many small, high-detail geometric features.
Fingers may fuse together, appear too few/many, or clip through each other — this is more likely when hands are small, partially occluded, or blurry in the source image.
Faces can look flattened, asymmetric, or lack fine detail, especially at lower resolution settings or with stylized/non-photorealistic references.
Try a reference image where hands and face are unobstructed and reasonably large in frame, and consider regenerating a few times, as results can vary between generations.
Multi-view inputs reduce ambiguity: Using Multi-view Image to 3D (front, side, and back images) gives the model significantly more information to work with, producing more accurate face proportions and hand geometry — especially when a single-image result looks flat or off-angle.
Stylized and fantasy art styles: Heavily stylized or fantasy characters produce more artifacts on hands, faces, and thin features (weapons, accessories) compared to photorealistic references. If your character uses a non-realistic style, expect more variance between generations and plan for additional retouching.
Rigging Expectations
If your goal is to animate the character afterward, plan your generation with rigging in mind from the start.
Generate in A-Pose or T-Pose rather than a dynamic action pose — auto-rigging tools and skeletons are built around these standard poses.
Keep limbs separated and avoid overlapping arms/legs in the reference image, since touching or intersecting geometry can make automatic rigging and weight painting unreliable.
Review the mesh topology around joints (shoulders, elbows, knees) before rigging, and clean up any obvious artifacts in your DCC tool of choice first.
Retouch Workflow
If a generation is mostly good but has a localized issue — like a distorted hand or a slightly off facial feature — you often don't need to start over.
Try regenerating the model a few times from the same input; results are non-deterministic and a different seed/attempt can resolve the specific defect.
Use any available local editing or retexturing tools to touch up specific regions rather than the whole mesh.
For persistent issues, bring the exported mesh into a DCC tool — Blender, ZBrush, or 3D Builder (Windows, beginner-friendly for quick non-manifold mesh repair) — to manually sculpt or repair problem areas. This is standard practice even with high-quality generations.
Limitations
Character generation quality has inherent limits worth setting expectations around.
Pose Control is designed for humanoid characters; non-humanoid or highly stylized subjects may not respond predictably to pose controls.
Very small or fine details (individual fingers, intricate facial features, accessories) are the most likely areas to show artifacts.
For heavily stylized or fantasy art styles, hands, faces, and thin features (weapons, accessories) are more prone to artifacts than with photorealistic references.
Not every generation will be perfect on the first try — treat generation as an iterative process and check your plan's current feature availability and limits in your dashboard.
FAQ
1. Why does my character's pose look wrong even though I described the pose in my prompt?
Text prompts alone are not a reliable way to control pose. Use the A-Pose/T-Pose toggle or Custom Pose controls in the Image to 3D workspace instead of describing the pose in text.
2. Why are my character's hands distorted or fused?
Hands are one of the most difficult features to reconstruct. Use a reference image where hands are clearly visible, unobstructed, and reasonably sized, and try regenerating if the first result isn't clean.
3. Why does my character's face look flat or asymmetric?
Face quality depends heavily on the clarity and angle of your reference image. A clear, front-facing image with good lighting generally produces better facial detail than an angled, blurry, or heavily stylized one. For best results, consider using Multi-view Image to 3D (front + side + back) to eliminate ambiguity and improve facial accuracy.
4. What pose should I use if I plan to rig and animate my character?
Generate in A-Pose or T-Pose using Pose Control. Standard rigging poses with separated, non-overlapping limbs give auto-rigging tools the best starting point.
5. Can I fix a distorted hand or face without regenerating the whole model?
Yes — try regenerating a few times first, since results vary between attempts. For persistent issues, use a DCC tool like Blender or 3D Builder to manually retouch just the affected area of the exported mesh.
6. Does Pose Control work on any character, including non-humanoid ones?
Pose Control is built specifically for humanoid characters. Non-humanoid or highly abstract characters may not respond predictably to A/T-Pose or Custom Pose adjustments.
7. Why does my stylized or fantasy character look worse than a realistic one?
Meshy's generation models perform best with photorealistic or semi-realistic references. Heavily stylized art styles — including anime, cartoon, and fantasy — introduce non-standard proportions, unusual color palettes, and exaggerated features that increase the likelihood of artifacts on hands, faces, and fine details like weapons or accessories. Use a cleaner, more detailed reference image, consider an Image-to-Image preprocessing pass to improve the input quality, and expect more retouching effort for stylized characters.
8. When should I use Multi-view Image to 3D instead of a single image?
Use Multi-view Image to 3D when a single-image result produces flat faces, inaccurate proportions, or artifacts on the back or sides of the character. Providing front, side, and back views removes ambiguity that a single image creates, and typically results in more accurate face geometry and overall character shape.
9. How can Image-to-Image help improve character quality?
Meshy's Image-to-Image feature lets you refine, clean, or recompose your reference image before passing it to Image to 3D. A better-framed input — with the character well-centered, clearly lit, and free of background clutter — directly improves the quality of hands, faces, and overall pose accuracy. This is especially useful when your original concept art or reference photo has framing, lighting, or composition issues that would carry through into the 3D output.