Sketch to 3D: how does AI turn drawings into 3D garments?
AI is starting to read fashion sketches and produce sewing patterns and draped 3D garments. How the technology works, what research shows, what commercial 3D tools already do and where the limits are.
KEY TAKEAWAYS Summary by the editors
- Turning a sketch into a 3D garment requires inferring a sewing pattern, fabric behaviour and a body to drape on, not just generating a 3D shape.
- The most promising approaches generate garments as structured, parametric data: GarmentCode (SIGGRAPH Asia 2023) describes sewing patterns as programs, and ChatGarment (CVPR 2025) uses a vision-language model to produce such descriptions from images, sketches or text.
- Commercial 3D garment software already simulates drape and fit from patterns; AI is being added to bottleneck steps such as fabric digitisation, as with CLO Virtual Fashion's zFab Kit announced in July 2025.
- Sketch-to-3D results are best treated as fast first drafts for visualisation and design discussion, not as production patterns.
- Accurate fabric data, standard body avatars and pattern-making expertise remain prerequisites for 3D garments that predict real fit.
AI turns a sketch into a 3D garment by inferring the garment's structure (type, panels, proportions), translating it into a sewing pattern and then simulating how that pattern drapes on a 3D body with given fabric properties. In 2026 research systems can do this end to end for many common garment types, while commercial tools mostly apply AI to specific steps. Results are useful for visualising ideas quickly but still need pattern makers to reach production quality.
Why is sketch to 3D harder than it looks?
A sketch is a stylised, two-dimensional drawing that omits most of what a factory needs. It rarely shows the back, the seams, the grain line or exact proportions, and fashion illustration often exaggerates the body. A 3D garment that behaves realistically needs flat pattern pieces, stitching relationships between them and fabric parameters such as stretch, weight and bending. The AI therefore has to infer missing information, which is where errors arise.
Generating a 3D shape that looks like a garment is comparatively easy. Generating one that can be cut from fabric and sewn is much harder, and only the second is useful for product development.
Researchers address this by narrowing the problem. Instead of predicting every point on a garment surface, systems predict a limited set of meaningful parameters, such as garment category, panel shapes and lengths, which a pattern system then turns into precise pieces. This mirrors how pattern makers work from blocks: most new styles are modifications of known structures rather than entirely new constructions.
How do AI systems turn drawings into 3D garments?
Current research converges on a pipeline that treats the garment as data:
- Interpret the input. A vision-language model reads the sketch, photo or text and identifies garment type and design features, such as collar, sleeve length or skirt flare.
- Produce a structured description. The model outputs parameters for a garment template rather than raw geometry.
- Generate the pattern. A parametric pattern system converts these parameters into flat sewing pattern pieces and stitching information.
- Drape and simulate. A physics simulator sews the pieces virtually around a 3D body and drapes them according to fabric properties.
- Edit by instruction. Designers adjust the result, for example lengthening a sleeve, and the pattern updates.
Two research projects illustrate this. GarmentCode, by Maria Korosteleva and Olga Sorkine-Hornung at ETH Zurich and published at SIGGRAPH Asia 2023, is a domain-specific language that builds sewing patterns from reusable components and automates low-level tasks such as placing darts. ChatGarment, presented at CVPR 2025, fine-tunes a vision-language model to read images, sketches or text, output a JSON description of garment type, style and numerical attributes, and drive a refined GarmentCode model to produce sewing patterns that can be draped on a 3D body and animated. It also supports editing through dialogue.
What can commercial 3D tools do today?
Established 3D garment software has long simulated drape and fit from patterns. FashionUnited's August 2025 overview of design and product development software describes capabilities such as 3D garment simulation, automated pattern making, fabric fit simulation, pattern digitisation and virtual try-on across tools in this category. AI is being added at bottlenecks. CLO Virtual Fashion announced the zFab Kit in July 2025, a fabric digitisation system for enterprise clients that it says uses AI to automate analysis of a fabric's physical qualities, so digital fabrics are accurate enough for design and production.
| Aspect | Research systems (e.g. ChatGarment) | Commercial 3D practice |
|---|---|---|
| Input | Images, sketches or text | Patterns created or imported by pattern makers |
| Pattern creation | Generated from parameters | Mostly manual or semi-automated |
| Fabric data | Generic or simplified | Measured fabric properties, increasingly digitised with AI support |
| Output quality | Plausible garments for common types | Production-oriented, depends on operator skill |
| Best use | Fast exploration and visualisation | Fit review, virtual sampling, line presentation |
For brands this means the current commercial value of AI in 3D lies less in generating garments from sketches and more in making existing 3D workflows faster and more accurate. Reliable fabric data, in particular, determines whether a simulated drape can be trusted for fit and proportion decisions, which is why digitisation has become a focus for automation.
What are the limits of AI sketch-to-3D today?
- Template coverage: parametric systems handle garment types their templates describe; unusual constructions may not be representable.
- Hidden information: backs, closures and internal construction must be guessed when the sketch does not show them.
- Fabric realism: drape is only as accurate as the fabric parameters; generic values can mislead fit decisions.
- Fit and grading: a garment that drapes well on one avatar is not a graded size range.
- Integration: generated patterns need to move into the brand's pattern and product development systems to be useful.
Where does sketch to 3D create value for brands?
The clearest value is early in development: aligning designers, merchandisers and sales on a silhouette before physical samples are cut, testing proportions, and creating quick 3D visuals for internal line reviews. Further downstream, value depends on a brand's existing 3D maturity, meaning digitised fabrics, standard avatars and staff who can judge simulated fit. Brands without that base usually gain more by building it first, since AI-generated drafts cannot compensate for missing fabric and pattern data.
What should brands watch in the next two years?
Three developments are worth tracking: whether research approaches that generate garments as parametric data reach commercial 3D and pattern software, whether AI-assisted fabric digitisation becomes routine, and whether generated patterns can be exported cleanly into production workflows. Each would move sketch-to-3D from visualisation towards development, but none removes the need for pattern-making expertise.
Frequently asked questions
Can AI turn a fashion sketch into a 3D model?
Yes, research systems such as ChatGarment, presented at CVPR 2025, can turn sketches, images or text into sewing patterns and draped 3D garments for many common garment types. Commercial 3D tools are adding AI to specific steps, and results still need review by pattern makers.
What is GarmentCode?
GarmentCode is a domain-specific language for programming parametric sewing patterns, developed at ETH Zurich and published at SIGGRAPH Asia 2023. It builds patterns from reusable components so that design parameters and body measurements can be changed while the pattern updates.
Can a 3D garment replace a physical sample?
For some decisions, such as silhouette and proportion reviews, 3D garments can reduce the number of physical samples. Replacing fit samples requires accurate fabric data, standard avatars and skilled operators, and many brands still approve final fit on physical garments.
Do you need a pattern maker if you use AI 3D tools?
Yes. AI can generate draft patterns and 3D visualisations, but production patterns require decisions on construction, fit, tolerances and grading that depend on pattern-making expertise and on the brand's own blocks and fit standards.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- ETH Zurich Interactive Geometry Lab: GarmentCode, programming parametric sewing patterns
- arXiv: ChatGarment, garment estimation, generation and editing via large language models (CVPR 2025)
- PR Newswire Asia: CLO unveils the zFab Kit, an AI-powered fabric digitisation system
- FashionUnited: AI software for design and product development
