Why should you generate the model face before adding the costume in AI fashion ads?

Generating the face before the costume keeps the model from averaging facial features into the wardrobe — when face and costume are generated together, the model trades off detail between them and you get generic, plastic-looking faces. Locking the face first gives you a clean identity anchor, then the agent auto-combines that locked face with the right lookbook outfit per scene.

Cast the faces without costumes first, pick the ones you want, and lock them by version number. Then point the invideo agent at your lookbook and let it auto-assign the right wardrobe per scene — you stop prompting costume assignment shot by shot, and the face you approved is the face that shows up in every frame.

The reason this order matters is mechanical. A single generation that has to solve face identity AND costume detail at the same time splits the model's attention across both, and faces are the part that suffers — you get the uncanny, slightly-plastic look that's the most common failure mode in AI fashion stills. Generating the face on its own lets the image model spend its budget on skin, eyes, and bone structure; the costume gets added against an already-locked identity, so the wardrobe iterates without disturbing the face. The face-first ordering is the same logic recommended across the AI image space — establish identity, then layer wardrobe, pose, and lighting on top.

For casting itself, Recraft tends to give better skin texture for portrait-style generations than alternatives — the invideo agent routes there automatically when you ask for casting options. Once you have faces you like, lock them by their exact version number in your lock command so the agent uses those precise outputs for every downstream step. From there, multi-angle character sheets (front, side, back) get generated against the locked face, and those sheets become the reference lock for every storyboard frame and every video clip.

Downstream this pays off twice. First, the same face holds across an entire campaign — multiple products, multiple scenes, multiple shots — which is what makes a fashion ad feel like one shoot instead of a stitched-together montage. Second, when you localize or swap products later, you're swapping wardrobe against a stable face, not regenerating the person. As Hridaye, invideo's creative director, puts it: "The thing that makes a localized ad land isn't translation. It's casting someone the market actually sees themselves in." That casting decision is exactly what you're protecting by generating the face first.

One practical note: if your ad is single-character and single-location, you can collapse the face-then-costume sequence into direct keyframe iteration — but the moment you have multiple characters or multiple outfits, face-first casting is the only order that holds identity across the campaign.

Watch some of these to see what works for you:

The thing that makes a localized ad land isn't translation. It's casting someone the market actually sees themselves in.

— Hridaye, invideo's creative director