A useful fashion flat needs to survive the next edit. A designer may want to change a pocket, adjust a sewing line, or try a different color on the hem. Those tasks depend on how the drawing is organized beneath its appearance.
At AI Tech Packs, we are converting fashion flat sketches from PNG to SVG: scalable vector graphics made from editable paths and shapes. Our goal is a working fashion CAD drawing—an editable technical flat with components a designer can select, revise, and recolor. We combine local algorithms and AI to draw, group, and refine its geometry specifically for fashion flats.
Our SVG beta is live. For this release, we focused on smart grouping of garment components, especially sewing lines, and colorable fills. Here is why we chose that focus, what makes the problem difficult, and what one hoodie example shows about our progress.
Why we are working on this
Fashion design involves repeated decisions. A sketch becomes a clearer flat, construction details change, and the team explores colorways before bringing the design into a tech pack. A drawing that can be revised makes that process easier to continue.
PNG images are useful for communicating an idea. They store the visible result, including the outlines, stitching, folds, and shading. They do not retain the original objects, groups, or relationships that a designer used to create the drawing.
We want to reconstruct useful components and regions from that image, with less manual organization before the next edit.
Why flat-sketch PNG-to-SVG conversion is hard
The image does not tell a converter what each line means. A dark curve might represent a garment edge, a fold, a seam, or a trim. Closely spaced lines might be separate construction details. A series of small marks might belong to one continuous sewing line.
Recovering those roles requires interpretation. Different arrangements of vector objects can produce very similar pictures, while behaving differently when someone tries to edit them.
Connectivity matters too. Two lines that touch visually may need to remain separate objects. A garment section that looks enclosed needs suitable geometry if a designer is going to fill it with color. A tiny gap or incorrectly joined boundary can change which area gets filled.
There is also a balance between detail and simplicity. Following every irregularity in the source can produce awkward geometry. Simplifying too aggressively can change a pocket opening or remove a construction detail. Fashion flats need enough precision to communicate the design and enough organization to support revision.
That is why we evaluate the resulting drawing through editing tasks as well as its resemblance to the reference.
The main approaches to creating an SVG
Traditional tracing
Image tracing turns raster artwork into vector geometry. It follows image boundaries or line structure, fits paths, and offers ways to adjust the result. Adobe Illustrator's Image Trace, for example, provides controls for paths, corners, noise, strokes, and fills. Its results can be expanded into editable objects. Adobe Image Trace options, editing traced results.
Tracing is a useful foundation. The remaining question for a fashion workflow is whether the resulting objects correspond to what the designer wants to select and change. A trace may need further work to organize garment components, separate sewing details, or prepare regions for recoloring.
Generative SVG models
Models such as QuiverAI's Arrow generate vector output from images or prompts. They can make decisions about the geometry and structure of a drawing. QuiverAI already showcases fashion flats and describes improvements in editability, line quality, and component placement. Arrow 1.1 announcement.
For fashion work, we inspect actual outputs for reference fidelity, useful organization, and the edits a designer can make immediately.
Our hybrid approach
We combine local algorithms and AI to draw, group, and fix the SVG. This lets us apply vector processing and garment interpretation within a workflow built around fashion flat sketches.
The approach gives us opportunities to refine the drawing as it takes shape: construct useful geometry, organize related elements, and correct problems that affect editing or filling. Drawing, grouping, and fixing work together toward the final asset.
We believe this is the right fit for our use case because we can tailor the system closely to garment structure and fashion editing tasks. That specialization has a tradeoff. Our SVG system would not work well as a general-purpose converter for other kinds of graphics.
What we focused on in the live beta
Smart grouping makes the drawing easier to navigate. Related elements should form useful editing units. Sewing lines deserve particular attention because a garment can contain many separate stitch details across the hood, sleeves, pocket, cuffs, and hem.
In this beta hoodie, the SVG separates structural lines, trims, sewing lines, and fillable regions. The sewing structure also identifies specific components, including pocket topstitching, cuff stitching, and sleeve stitching. That makes it possible to address a particular sewing detail or work with the sewing group together.
Colorable fills support the next color decision. We want designers to change a garment region's color while keeping its line work available. This requires useful region boundaries and a drawing structure that keeps those fills manageable.
For the hoodie example, we tested the existing SVG geometry directly. We highlighted and hid the sewing-line group, then recolored the pocket region and lower hem. Those operations preserved the underlying paths and left the other line objects unchanged.
Original beta SVG. No edits applied.
This article demonstration edits existing SVG objects. It is not a screenshot of the product editor.
What the Quiver example helps explain
The Quiver SVG used for this example also contains useful editable geometry. Its sewing lines use dashed strokes, and shared styles allow those strokes to be restyled together. Its pocket is a polygon that can accept a fill. We tested both operations successfully.
The organizational difference is visible in the files:
| Editing task or detail | AI Tech Packs beta example | Quiver example |
|---|---|---|
| Select garment components | Named garment and sewing groups | Individual objects and shared styles; no garment-component groups |
| Restyle sewing lines | Sewing group and named stitch components | Dashed strokes share styles and can be restyled together |
| Recolor regions | Dedicated closed fill regions; pocket and hem tested | Pocket polygon accepts a fill; tested successfully |
| Preserve hood and pocket stitching | Hood center-strip and right pocket-side dashes represented | Those details are absent as dashed lines |
We inspected these supplied files and tested edits in a browser on September 11, 2026. The exact Quiver model version, generation settings, and retries were unavailable. This comparison does not establish Illustrator import behavior or performance across other garments.
Overlaying the SVGs on the PNG
Placing each SVG over the generated PNG makes the differences easier to inspect. In this hoodie example, the beta preserves more of the reference's sewing detail. The Quiver rendering omits the dashed stitching alongside the hood's center strip and does not reproduce the right pocket-side stitching as dashed lines. Those details remain represented in the beta drawing.
View still close-ups of the hood and pocket
Both drawings still deserve close inspection. Our beta example has uneven hood details and rough line junctions that need further refinement. The work is moving toward a drawing that remains useful through revision, and visual fidelity is part of that work.
Keep working from the flat
Our research continues around a practical goal: help a designer pick up the converted drawing and make the next change. Smart component grouping, grouped sewing lines, and colorable regions are the priorities of this live beta.
You can start with the AI Tech Packs flat sketch generator and try the beta SVG export. When the drawing is ready for the next stage, continue into the tech-pack workflow, with construction details reviewed by the design team.