Why People Say Illustrations Can't Be Turned into 3D | Cases Even AI Struggles With

The claim that "illustrations can't be turned into 3D" still lingers all over the internet. Some of you have probably tried it yourselves and ended up with something quite different from what you pictured.

To give the conclusion up front: there is a range that works and a range that doesn't, and there are technical reasons for where that line falls. It is neither "everything works now" nor "nothing works yet." In this article we lay out, without exaggeration, the shapes AI struggles with, the mechanism that makes the output drift from what you expected, and the conditions under which things stall at the output stage. We cover the conditions that tend to work with the same level of detail.

What you'll learn in this article

• The three different meanings mixed into the word "can't"
• The shapes AI struggles with, and the technical reasons why
• The mechanism that produces a 3D form different from the one you had in mind
• Concrete criteria for shapes that stall at the 3D printing stage
• The conditions under which a drawing tends to work relatively well

Note: Figmee currently lets you generate and download figurine-style images and 3D model data (GLB / 3MF) from drawings and illustrations you upload. Sales of physical 3D-printed items are Coming Soon.

"Can't" mixes three different meanings

Let's start by separating the terms. When people say "this illustration can't be turned into 3D," they are actually mixing three issues of a very different nature.

1. AI has trouble reproducing it as a shape — a weak spot on the technology side
2. The resulting 3D form is different from what you expected — a problem of interpretation
3. The data was created, but it can't be produced as a physical object — a constraint on the fabrication side

These three have different causes and different remedies. If you lump them together as "can't," you end up giving up even on problems that could actually be solved. Let's go through them in order.

(1) Shapes AI struggles with

Thin protrusions and thin structures

This is the clearest weak spot. Thin structures such as hair tips, slender antennae, fingers, the point of a sword, or threads and chains are hard to reconstruct in 3D; they break off partway or fuse with whatever is around them.

What's interesting is that this "thinning" phenomenon has been pointed out to share the same root as the problem of flat coloring, which we explain next. It is a difficulty inherent in the very problem of estimating a 3D form from a single image, not the kind of issue that goes away by switching tools.

Flat-colored illustrations

This may come as a surprise, but flat-colored hand-drawn illustrations are harder to turn into 3D than photographs.

The reason is the amount of shading information. Photos contain gradations created by how light falls on the subject, and the bumps and hollows of the surface can be inferred from them. Drawings built from solid fills and minimal shadows, such as anime-style cel shading, offer far fewer clues about the 3D form.

Research has tackled this point head-on. One paper reports that existing image-to-3D methods "all break down on flat-colored images and output abnormally thin structures." The cause it identifies is a distribution shift: the data these methods were trained on contained plenty of 3D cues such as shadows and plausible depth structure, while flat-colored drawings lack them (source: Art3D: Training-Free 3D Generation from Flat-Colored Illustration, published April 2025).

In other words, the "thinning" phenomenon from the previous section and flat coloring are not separate problems; they stem from the same cause. And this means that "turning illustrations into 3D is fundamentally harder than turning photos into 3D." Services that deal with illustrations work on this area precisely because this is where the hard part lies.

What you can do on the input side is to settle on a single light source, organize light and dark into three tones, and draw cast shadows and contact shadows distinctly. Shading techniques for giving an illustration a sense of depth cover this process step by step.

Compositions where front-to-back order is hard to read

Arms folded in front of the body, a hand over the face, a cape wrapped around the body, hair overlapping an arm. In compositions with this much overlap, it is hard to tell which part is in front and which is behind, and the parts can end up fused together in the 3D form.

Sides that aren't drawn

When you start from a single front view, the back has to be estimated. The hairstyle at the back of the head, the design on the back, the soles of the feet. The drawing contains no information about these, so the AI fills them in with "a plausible shape." When this differs from your intent, it is less an AI failure than an inevitable result of missing information.

Transparent and semi-transparent elements

Glass, water, flames, light effects, see-through fabric. From an image it is hard to judge whether there is a surface there or not, which makes these difficult to treat as 3D. In compositions where effects overlap heavily, the effect itself can be turned into a solid lump. When working with heavily decorated illustrations, preparing a version with the effects removed for 3D conversion makes the results more stable.

Summary of the weak spots

(2) The mechanism behind "this isn't what I expected"

Sometimes nothing has technically broken down, yet the result feels off the moment you look at it. This pattern has a different cause.

A drawing can work even when it is contradictory as a 3D object. In illustration, artists deliberately shift the position of the eyes for appeal, pile on volumes of hair in shapes that couldn't physically exist, or exaggerate the proportions of body parts. These are expressions that work because the image is flat, and when converted to 3D they are reproduced "as is," which creates the sense that something is off.

The face is where this is most pronounced. Anime- and manga-style faces are often designed to look cute from the front and are not meant to be seen in profile. The awkwardness you notice when you view the 3D form from the side is not an AI error; it is the flat-only expression in the original drawing becoming visible.

Here are some concrete examples. Large manga-style eyes look like they fit inside the face on a flat page, but in 3D they don't match the curve of an eyeball and look strange. A tuft of hair flicking upward is an easy-to-read directional symbol in 2D, but in 3D it becomes a floating shape with nothing supporting it. A slender line from jaw to neck can look beautiful from the front but turn out too thin when seen from the side. All of these are expressions that are correct as a drawing but hard to make work as a 3D object.

This phenomenon is actually useful information. Knowing where things break down in 3D means knowing which parts of the design need rethinking. If you read the generated result not as a "failure" but as "a visualization of the 3D contradictions in the original drawing," it becomes useful in a different way. In fact, in professional prototype sculpting, flat character sheets are never translated directly into 3D; there is always a step of adjusting the design so it works as a solid object. The places where AI output feels off broadly match the places that need that adjustment.

(3) Shapes that stall at the 3D printing stage

Even when there is nothing wrong with the data, things can stall once you get to producing a physical object. The criteria are concrete.

For resin (stereolithography), a wall thickness of roughly 0.5 mm to 2.0 mm is required depending on the size of the surface, and engraved and embossed details are recommended to be at least 0.8 mm deep and wide (source: JLC3DP "3D Printing Design Guidelines", as of July 2026). Another guide explains that although a wall thickness of about 1 mm can be printed, it breaks easily when support material is removed, so for strength it is safer to secure at least 1 mm (source: DMM.make "A Complete Guide to Stereolithography (SLA) 3D Printing", as of July 2026).

Illustrated characters are highly likely to have parts that fall below these thresholds. Thin locks of hair, the ends of ribbons, accessory chains, slender legs. Most cases of "I could make the 3D data but couldn't turn it into a physical object" happen here. To deal with this, you need to make decisions such as thickening parts in modeling software, splitting them into separate pieces to glue together later, or boldly leaving them out. We go over which materials suit which purposes in Which materials suit 3D printing from illustrations?.

Conversely, conditions that tend to work

Since we've written mostly about weak spots, here are the conditions that tend to succeed, with the same level of detail.

• A full-body drawing in a neutral standing pose. A composition where the arms are away from the body and parts overlap little
• A drawing with a clear silhouette. Strong lines and clear boundaries between surfaces
• Strongly stylized characters. Designs with few fine details and a clear overall mass
• A white background with only the subject. Detailed backgrounds interfere with detecting the outline
• A drawing with some shading. Compared with completely flat coloring, shadows provide more cues about the 3D form

Put the other way around, one option is to draw a separate image specifically for 3D conversion that meets these conditions. Rather than feeding in a finished piece with an elaborate composition as is, preparing a separate rough sketch in a neutral standing pose quite often gives results that better serve your purpose.

Checklist before you try

• Is the whole body inside the frame?
• Do the arms and legs avoid overlapping heavily with the torso?
• Are there any extremely thin parts in the hair or accessories? (If so, expect the results to be unstable.)
• Do you care a lot about the design of the back? (If so, plan for fixes after generation.)
• Is your final goal the data, or a physical print? (If the latter, you need to check wall thickness.)

Frequently asked questions

So, can illustrations be turned into 3D or not?

It depends on the conditions. Creating a 3D form as data is possible with today's tools, but there are limits in reproducing thin protrusions, estimating sides that aren't drawn, and turning flat-only expressions into 3D. A more realistic way to see it is not "a finished product comes out as is" but "you get a 3D base in a short time."

If I regenerate many times, will I get the ideal shape?

Results change with each generation, so it is worth trying different conditions. However, the structural weak spots listed above won't disappear through repetition. If the same part keeps breaking no matter how many times you try, you'll make faster progress by switching to either changing the input drawing or fixing it by hand after generation.

Can I fix it myself after generation?

Yes. If you import the data exported as GLB / 3MF into 3D software such as Blender, you can correct the shape and add parts. For how to think about surface detail, see What Is Polygon Count?.

Does a three-view drawing improve accuracy?

A three-view drawing explicitly provides front, side, and back information from a person, so it does reduce the parts that rely on estimation. However, the current mainstream approach to generating 3D from illustrations takes a single image as input, and it is not guaranteed that a three-view drawing can be used as input. Three-view drawings really shine in the process where a person builds the model by hand in modeling software. The expectation that "if you give AI a three-view drawing, you'll get exactly the shape you want" does not match reality at this point.

Why do photos work better?

Because photos contain gradations from how light falls, from which the orientation of surfaces and their relief can be inferred. As mentioned above, flat-colored drawings lack these cues. This is also why, even for the same character, a drawing with carefully added shadows tends to produce more stable results.

Why is outdated information still around?

This field changes quickly, and descriptions written a few years ago may no longer match the current situation. At the same time, the structural difficulties described in this article still remain. If you read with attention to the date of the information, and to whether a statement refers to "the limits of the technology" or "the state of services at the time," you are less likely to make the wrong call.

Summary

Turning illustrations into 3D is not a binary of "can" or "can't." Thin structures, flat coloring, compositions with heavy overlap, sides that aren't drawn: these are weak spots with technical reasons behind them, and they won't vanish by switching tools. On the other hand, with a full-body drawing in a neutral standing pose or a design with a clear silhouette, you can get a 3D base in a short time.

And most cases of "this isn't what I expected" are the flat-only expressions in the original drawing becoming visible through 3D conversion. Whether you see this as failure or as design information for sculpting makes a big difference to how you can use this technology.

Related guides

• How Image-to-3D AI Works — A Beginner's Guide to Turning Drawings and Photos into 3D
• How to Turn an Illustration into 3D (2026 Edition): A Complete Guide to Turning Your Lines Directly into 3D with AI
• Image-to-3D AI Tools Compared (2026 Edition): Choosing by How Faithfully They Reproduce Illustrations
• How to 3D Print an Illustration: The Basics of Data Preparation, Thickness, and Material Choice