Create AI Selfies with your Reference Image

Inputs (keep short)
- REFERENCE_IMAGE: {optional: upload a clear, well-lit photo of the person to appear in the selfie}
- ASPECT_RATIO: {your ratio}
- PERSON: {short description — if using reference image, describe clothing/accessories/temporary changes only; if no reference, describe full person}
- SETTING: {short description}
- POSE: {short description}
- SELFIE_MODE: {FRONT_CAMERA or MIRROR_SELFIE}

Priority Stack
1) MUST be an actual selfie capture (camera viewpoint = phone position)
2) Realism > everything (unfiltered, imperfect)
3) If REFERENCE_IMAGE is provided, preserve subject identity with high fidelity (see Reference Image Fidelity Gate)
4) Match PERSON, SETTING, POSE
5) Match ASPECT_RATIO exactly

Selfie Authenticity Gate (non-negotiable)
- The image must be taken BY the subject using a smartphone.
- Viewpoint must match selfie capture mechanics:
  - FRONT_CAMERA: camera is the phone's front lens at arm's length. Phone is NOT visible (or only a tiny edge at most).
  - MIRROR_SELFIE: phone CAN be visible, but only as reflection logic (mirror) with correct reflection and perspective.
- If it looks like an external photographer shot, it is WRONG.

Reference Image Fidelity Gate (when REFERENCE_IMAGE is provided)
- Preserve the following from the reference image with high accuracy:
  - Facial bone structure, jaw shape, and face proportions
  - Eye color, eye shape, and eye spacing
  - Skin tone, complexion, and any visible skin features (freckles, moles, scars, birthmarks)
  - Nose shape and size
  - Lip shape and proportions
  - Hairline shape (hair style and color may change only if specified in PERSON field)
  - Ear shape and position
  - Overall body proportions if visible
- Adapt ONLY the following to match the new scene:
  - Lighting and shadows on the face (must match SETTING light source)
  - Expression (must match POSE description)
  - Clothing and accessories (must match PERSON description)
  - Hair styling only if explicitly changed in PERSON field
  - Camera angle perspective distortion (must match selfie mechanics)
- Do NOT blend, morph, or average the reference face with any other identity.
- Do NOT beautify, smooth, or idealize features beyond what appears in the reference.
- The result must be immediately recognizable as the same person in the reference photo.
- If the REFERENCE_IMAGE is filtered or beauty-moded, attempt to see through those filters to the natural face beneath.

FRONT_CAMERA (default) required cues
- Arm-length framing, slight wide-angle distortion at edges.
- Natural hand-held tilt, imperfect centering.
- Face is closest to camera; mild perspective stretch (subtle).
- Eyes sharp; subject looking into or near the lens.
- Do NOT show the whole phone in the foreground.

MIRROR_SELFIE required cues
- Scene includes a mirror; subject + phone visible in reflection.
- Reflections must be physically plausible; background matches mirror space.
- No third-person camera viewpoint.

Generate
Create ONE ultra-photoreal, unfiltered smartphone selfie.
If REFERENCE_IMAGE is provided, use it as the identity anchor for the subject.
Expand the short inputs into realistic specifics (skin texture, hair flyaways, believable clothing, small environment details).
Keep everything plausible and consistent.

Person realism
- If REFERENCE_IMAGE is provided: use the reference face as-is with all its natural features. Apply clothing and temporary changes from PERSON field only.
- If NO REFERENCE_IMAGE: create a NEW non-celebrity identity (do not resemble a famous person).
- Natural skin: pores, minor blemishes, subtle under-eye shadows.
- No beauty filter, no airbrushing, no perfect symmetry.

Setting realism
- Expand SETTING with 2 to 4 concrete details.
- Single main light source that makes sense (window, daylight, lamp, neon).
- Background is real but secondary (light blur ok).

Pose expansion
- Expand POSE into: head angle + expression + arm position holding phone + framing and crop.
- Natural posture (no staged photoshoot posing).

Avoid (hard negatives)
- Third-person or photographer-taken look
- Phone prominently in foreground in FRONT_CAMERA mode
- Studio portrait vibe, cinematic grading, CGI or illustration look
- Text, watermarks, fake UI overlays
- Face morphing, identity blending, or averaging with other faces (when using reference image)
- Beautification or smoothing beyond what exists in the reference image