Reference photo vs text-only?

A clear adult reference photo is the steadiest way to keep a face. Text-only works for a look, not a specific person across many shots.

What if the ref looks underage or ambiguous?

Do not use it. Rewrite or replace until age reads clearly adult. Ambiguous age is a stop, not a creative choice.

How many refs before quality drops?

Start with one strong frontal or three-quarter adult face. Extra conflicting refs often hurt more than they help.

Can I reuse the face in video later?

Yes. Lock stills you like, then feed a keeper into image-to-video so motion starts from the same identity.

Keep the same face across adult AI stills

Consistent character AI for adult stills is mostly reference discipline: one clear adult face, a short scene prompt, and small changes between runs. Text-only prompts invent a new person every time; that is fine for mood boards, not for a recurring character.

Adults only. Never use references or language that imply minors.

Pick a reference that can carry identity

Use a sharp photo of a clearly adult subject, face unobstructed, lighting you can live with. Avoid heavy filters, group shots, and crops that hide the jaw or eyes.

If the person in the photo could be read as underage, drop it. Same rule as Prompt tips for realistic adult AI stills: ambiguous age is not fixable with prompt adjectives.

On Lumyn, attach the reference in Studio on models that accept an input image, check the coin price, and generate. Prefer models that already allow adult output; see uncensored vs censored.

Prompt around the face, do not overwrite it

Let the reference own identity. Your text should own place, wardrobe, light, and pose:

> Same woman as reference, hotel windowsill at dusk, open silk robe, waist-up, looking toward camera

Avoid restating a full facial description that fights the photo. If the model drifts, simplify the text and strengthen the ref rather than adding ten face synonyms.

Iterate without breaking the lock

Change one axis per run: crop, outfit, light, or expression. Keep a small bank of keeper stills under the same character. When hands or anatomy break, shorten the pose line; long negative lists rarely restore a face.

One strong reference usually beats a pile of conflicting ones. Add a second angle only when the first lock is already stable.

Promote keepers into video

Once a still matches the character, use it as the start frame for short clips rather than regenerating identity from text. Workflow: Turn a still into short AI video.

Trust and storage of outputs: Does Lumyn keep my generations?.