Textured and Afro Hair in Virtual Try-On — What Fidelity Requires

“Supports curly hair” is not a sufficient quality claim.

Curly, coily and Afro-textured hair contains geometry that is easy to simplify into a generic mass. A system may produce an attractive curly result while losing the exact curl structure of the catalogue reference.

For a wig boutique, that is a product mismatch.

Texture is not style

The Stanford preprint *Detangled* separates persistent strand texture from the overall direction or style of the hair. It proposes parameters including thickness, curl radius, curl wavelength, twist prevalence and porosity.

That distinction is useful for virtual fitting. Two wigs can share the same overall silhouette while differing materially in:

• curl diameter;
• distance between curls;
• visible twist;
• clumping;
• density;
• frizz;
• strand separation.

A result that preserves length and volume but replaces defined ringlets with dense generic curls has not preserved the reference.

Why complex curls are difficult to reconstruct

The paper explains several technical difficulties:

• curly styles are underrepresented in databases;
• multiple curl directions can intersect within one image pixel;
• high-curvature geometry weakens methods based on simple direction gradients;
• curls self-occlude;
• internal twist is difficult to represent.

These challenges concern digital hair more broadly. They do not directly benchmark every commercial wig try-on product. They do explain why a universal claim should require evidence across texture families rather than one successful image.

What the ELLËS test found

An initial ELLËS test placed a lace 3C wig with defined ringlets onto a source image with natural 4C hair.

The first result handled the broad transformation well:

• the original Afro silhouette became a longer falling shape;
• the centre part remained credible;
• the temple showed no obvious hard helmet line;
• the dark colour remained close in a comparable exposed area.

The main failure was texture. Ringlets appeared denser and more aggregated than the catalogue reference. The result moved toward “generic curly.”

After an implementation change, a second result showed more individually defined spirals and more visible negative space between curls. The improvement was visually clear in the examined regions.

Three later tests with the same 3C reference preserved the improved ringlet separation across different source hairstyles and skin tones. This is a useful robustness signal for that reference.

It does not prove performance for 3A, 3B, 4A, 4B or 4C wig references.

Replacing existing hair is a separate test

A fitting system has to account for the source hairstyle beneath the selected wig.

The test set included:

• a natural 4C Afro;
• tight buns;
• cornrows;
• a fully exposed scalp.

In one cornrow case, the output showed no obvious remaining ridge or geometric trace from the source braids in the reviewed crown area. That is a positive observation.

The same image also showed that quality can vary around one hairline: the upper forehead was convincing, while a lateral temple edge appeared more geometric.

This matters because a single score for “hairline quality” can hide a local failure. Review the crown, central hairline, both temples, ears and shoulder transitions separately.

Skin-tone stability must be tested alongside hair

Representation is not solved if the curl is accurate but the face changes.

In the small ELLËS variance set, sampled facial colour differences changed direction across models and facial regions. One darker-skinned model showed the largest measured cheek difference in that campaign. With the small sample and uncontrolled lighting, it is not possible to infer a systematic relationship between skin tone and error.

It is also not responsible to declare the problem solved.

The result supports four actions:

1. Expand the darker-skin test set.
2. Control lighting or record its quality.
3. Measure multiple facial regions.
4. Reject a universal correction that improves one region while worsening another.

The goal is not to force every source photograph into the same brightness. It is to avoid unexplained changes between the source face and the fitting.

A proper texture test matrix

One model per category is not enough. Build a matrix across:

Each cell should be reviewed for:

• catalogue texture identity;
• root and part;
• hairline by zone;
• source-hair residue;
• occlusion;
• face preservation;
• colour confidence;
• repeatability.

This is a demanding matrix. It should grow with real commercial priorities rather than delaying every pilot until every cell is full.

Commercial language before full validation

Avoid:

Perfect for every hair type and skin tone.

The first fully inclusive fitting system.

Certified 4C accuracy.

Use:

Tested first on source images often missed by generic tools, including natural 4C hair and cornrows.

Current 3C tests show stable ringlet definition across several source hairstyles. Broader texture validation continues.

Skin-tone stability remains an active measurement area.

This language turns transparency into evidence of discipline.

Why this work can become defensible

The defensible asset is not a broad promise. It is a growing, traceable test system:

• known catalogue references;
• repeated input pairs;
• labelled source conditions;
• local hairline review;
• texture-specific criteria;
• documented failures and corrections.

Competitors can claim support for many styles. A boutique choosing between systems needs a clearer answer: what exactly was tested, against which reference, in which conditions, and what still fails?

A standard for faithful textured-hair fitting

A textured-hair result should preserve:

1. The curl family, not merely “curliness.”
2. The visible spacing and clumping pattern.
3. The catalogue silhouette and length.
4. A natural hairline around every visible zone.
5. The client’s face and skin appearance.
6. Stable product identity across repeated attempts.

The hardest cases should not be hidden in a gallery. They should shape the test protocol.

For the broader acceptance framework, read what virtual wig try-on should prove. For the colour and lighting limits that interact with skin and fibre appearance, read colour accuracy in virtual wig fitting.