One of the most common frustrations people face when using AI image generators like Nano Banana, DALL·E, or similar tools is trying to swap outfits between two people. For example, you may want to take a casually dressed lady and put her into a Yoruba traditional attire worn by another lady. But instead of blending them, the AI just keeps reproducing the second image exactly as it is — same face, same body, same person. Why does this happen, and how can you fix it?

In this article, we’ll explore why the one-step method fails, and then walk through a clever two-step approach that actually works. I’ll also share the exact prompts that solved the problem, plus a few bonus variations you can try for even better flexibility.

Why the One-Step Method Fails

When you provide two human images — one of the person you want to transform, and one of the reference outfit — the AI often struggles.

Failed prompt:

“Transform the woman in the first photo into a traditional Yoruba attire look. Keep her face, skin tone, and natural appearance the same, but dress her in full Yoruba cultural clothing like the second reference photo: a rich purple Aso-Oke wrapper (iro), blouse (buba), and matching head tie (gele). Add traditional coral bead jewelry on her neck, ears, and wrists. Pose her sitting gracefully on a wooden stool against a plain beige background, with a professional studio lighting style. Ensure the final image is elegant, vibrant, and realistic, resembling a formal Yoruba traditional portrait.”

Instead of dressing the first person in the second person’s clothes, it simply recreates the second person. This happens because:

• Reference dominance: The reference image (the Yoruba attire lady) is visually strong and distinctive, so the AI treats it as the main subject.
• No clear identity preservation: Unless you specifically tell the AI to preserve the first lady’s identity, it assumes you want to regenerate the reference lady completely.
• Prompt weighting confusion: With two competing human identities, the AI gets confused about which face, skin tone, and body physics to keep.

As a result, instead of “Lady A in Lady B’s outfit,” you end up with just “Lady B again.”

The Two-Step Solution That Works

The breakthrough is to separate the outfit from the person’s identity. Instead of trying to do everything in one shot, you guide the AI in two clear steps.

Step 1: Turn the Outfit Lady Into a Mannequin

In the first step, you ask the AI to strip away the human features of the reference lady while keeping her clothing intact. That way, you have the Yoruba attire preserved, but without her face, skin tone, or body physics.

Prompt 1:

Using the reference image, turn the lady’s body into a mannequin wearing the same dress and shoe.

This gives you a neutral mannequin dressed in the Yoruba attire — a blank canvas holding only the outfit.

Step 2: Replace the Mannequin With the Casual Lady

Now you bring in the casually dressed lady and tell the AI to map her identity onto the mannequin. Since the mannequin has no identity, there’s no conflict, and the AI easily applies the casual lady’s features to the Yoruba attire.

Prompt 2:

Using the first image as reference, turn the mannequin into the lady in the second image. Preserving lady’s facial looks, skin tone, legs and body physics.

The result? The casual lady’s identity perfectly combined with the Yoruba outfit, without any dominance from the original reference lady.

Why This Two-Step Method Works

Here’s why this approach succeeds where the one-step method fails:

• Step 1 isolates the clothing. By turning the reference lady into a mannequin, you remove her identity and keep only the outfit.
• Step 2 restores the target identity. You reintroduce the casual lady’s face, skin, and body physics, but now the clothes are already in place.

In short: no identity clash, no reference dominance — just a clean outfit transfer.

Bonus: Alternative Prompt Variations

If you want to experiment further, here are some variations you can try for each step. Sometimes small wording changes make the AI follow your instructions more accurately.

Step 1 Variations (Mannequin Creation)

• “Convert the lady in this image into a mannequin, but keep the same outfit, accessories, and shoes exactly as they are.”
• “Remove all human features from the lady, replacing her with a mannequin, while preserving the clothes, pose, and jewelry.”
• “Transform this photo so the model becomes a mannequin but the Yoruba attire remains untouched.”

Step 2 Variations (Identity Transfer)

• “Replace the mannequin with the woman in the first image, preserving her face, skin tone, body shape, and legs while keeping the outfit unchanged.”
• “Turn the mannequin into the first lady, applying her identity and physics but keeping the Yoruba clothes, accessories, and pose intact.”
• “Overlay the first lady’s facial features, skin tone, and body structure onto the mannequin wearing Yoruba attire.”

These variations give you flexibility depending on how literal or creative the AI model is with your wording.

Key Takeaways

• The one-step method fails because the reference image dominates and confuses the AI.
• The two-step method works by separating outfit (via mannequin) and identity (via the casual lady).
• Exact prompts matter, but variations give you flexibility depending on the model.
• This technique works across different outfits and styles, not just Yoruba attire.

Final Thoughts

AI image generators are powerful, but they don’t always interpret complex tasks the way we expect. When you try to do too much in one step, the model gets confused. By breaking the task into smaller steps — first preserving the outfit, then overlaying the person’s identity — you give the AI clarity, and the results come out exactly as you want.

If you’ve been frustrated with outfit swaps that never look right, try this two-step mannequin approach. It’s simple, effective, and it feels like magic once you see the results.