Comparisons

Best AI Model for Realistic Mirror Selfies: Blind Same-Input Test

One frozen identity photo and one exact mirror-selfie prompt across five models: four visible outputs, one recorded block, and four blinded evaluations.

Seedream V5 Pro mirror-selfie output from a blind same-input model test
Comparison images: Remix.Camera

Each model received the same input bytes, exact prompt, profile, aspect ratio, and comparison run. Four models returned an image. GPT Image 2 Medium blocked the request, and that attempt remains in the record rather than being replaced.

ModelScoreFirst-place votesPanel note
Seedream V5 Pro22.25/253Best combined identity fidelity and full-scene adherence; minor hand and toe artifacts remain.
Grok21.625/250Strong literal pose and scene adherence, with a slightly more generic, polished face.
Seedream V5 Lite21.25/251The most naturally photographic result for one reviewer, though the side-on crouch drifts from the requested pose.
Nano Banana 217.75/250Credible likeness and mirror-photo realism, but the missing heels reduce prompt adherence.

Input, outputs, and recorded block

Comparison 1

Candid bedroom mirror selfie

Blind panel: V5 Pro 22.25/25Swipe to compare →
Candid bedroom mirror selfie, Fixed input output
Fixed input
Candid bedroom mirror selfie, Nano Banana 2 · 17.75/25 output
Nano Banana 2 · 17.75/25
Candid bedroom mirror selfie, Grok · 21.625/25 output
Grok · 21.625/25
Candid bedroom mirror selfie, Seedream V5 Lite · 21.25/25 output
Seedream V5 Lite · 21.25/25
Candid bedroom mirror selfie, Seedream V5 Pro · 22.25/25 output
Seedream V5 Pro · 22.25/25
Prompt blocked by safety filters.
GPT Image 2 Medium

Four blinded visual evaluations placed Seedream V5 Pro first overall with three first-place votes. Grok ranked second on aggregate score; the one-prompt result should not be generalized beyond this scene.

Images: Remix.Camera
Exact prompt and retained controls

Using the person in the uploaded reference image (DO NOT TOUCH FACE), recreate this scene: Ultra-realistic candid bedroom mirror-selfie photo, high-end smartphone camera realism (iPhone 15 Pro look), vertical 4:5. Tight social crop with subject dominant (70–85% of frame), minimal headroom (1–3%), eyes in top third. Full-length mirror selfie captured in a genuine in-between moment (not posed like a studio shoot). SUBJECT An adult Gen Z woman with a slender, petite build and long-limbed proportions. Cinematic facial realism with clear bone structure (orbital depth, cheekbone planes, natural jawline), natural micro-asymmetry, and authentic skin texture. Expression is playful and teasing: penetrating gaze, slight pout with pink-plum glossy lips. She is looking at the phone screen (not directly at the viewer), eyebrows relaxed, eyes sharp and expressive. POSE / BODY LOGIC (STRUCTURAL) She is squatting low and wide in front of a full-length mirror, knees apart, heels grounded, balanced posture with realistic joint angles. Torso slightly forward for stability, hips low, back straight but relaxed. Both hands hold the smartphone in front of her chest/face level, elbows naturally bent. The squat must look physically real and comfortable—not broken anatomy, no twisted limbs, no awkward mannequin stiffness. OUTFIT (PREMIUM + CLEAR) - Off-the-shoulder fitted top (soft knit or structured stretch fabric), snug and clean seams, defined neckline sitting naturally across collarbones and upper arms. - Short skirt (mini length) with premium tailoring: structured waistband, clean hem, fabric identity obvious (black leather OR heavy satin OR textured ponte—choose one coherent material). - High-heeled sandals: elegant, minimal straps, realistic foot pressure and strap tension due to squatting pose, heels stable on the rug. ACCESSORIES (SUBTLE LUXURY) Tasteful jewelry only: small gold hoops or studs + thin bracelet/ring. Optional delicate anklet. No oversized luxury branding. Nails neat with soft neutral/pale pink polish. HAIR / GROOMING Hair remains biologically coherent and stylish: dark /black hair, healthy shine, realistic density. Style can vary slightly for best aesthetics (loose waves or sleek straight), but must remain believable and not costume-like. Natural flyaways and strand separation. MAKEUP (REALISTIC / NO FILTER) Natural glam suited for daylight: soft blush, subtle eyeliner/mascara, clean brows. No heavy contour. Skin shows pores, tonal variation, and natural highlight roll-off—NO glossy plastic skin, NO smoothing. ENVIRONMENT / ROOM Modern bedroom, slightly messy but curated (lived-in): full-length mirror leaning or mounted, bed with neutral bedding, slightly rumpled sheets, pillows not perfectly arranged. White walls with minimalist line art above bed. Soft gray rug in foreground. A nightstand with small decor (plant, candle, scattered items). Background clutter is subtle and realistic (a hoodie on a chair, a bag on floor) but not distracting. LIGHTING Soft natural daylight from a window (side/front 45°), flattering and real. Gentle shadow transitions, no harsh studio flash. Exposure balanced so face remains readable; eyes not crushed into shadow. No cinematic color grading—true-to-life smartphone tones. CAMERA / COMPOSITION Mirror selfie perspective with slight handheld imperfection (minor tilt, not perfectly centered). Sharp focus on face + outfit texture + hands/phone. Background softly blurred with natural smartphone depth separation. No excessive negative space—subject fills the frame. No text, no watermark, no logos. NEGATIVE (DRIFT BLOCKERS) professional studio lighting, DSLR look, cinematic grading, beauty filter, poreless airbrush skin, glossy plastic sheen, over-sharpening, warped hands, extra fingers, broken wrists, twisted joints, floating limbs, face distortion, stiff mannequin pose, subject too small in frame, excessive empty space, heavy makeup, cartoon/anime/CGI/3D render look, readable text, watermark, logos

  • Prompt SHA-256: b4a8c80671d3ed35010e755780090a96e028d92fc8a0b96629129da5fd144422
  • Input SHA-256: 74a0a02df101c0c3993dd14c1881c6aa6dfb8178479696535a6c0affa130db7d
  • Run ID: mirror-selfie-lily-20260717-v1-bedroom-selfie-squat
  • No retries or substitute outputs

What separated the outputs

  • V5 Pro best balanced identity fidelity with the full requested scene, although small hand and toe artifacts remained.
  • Grok followed the pose and room structure closely but shifted the face toward a more generic polished look.
  • V5 Lite produced a convincing photograph but changed the requested squat into a more side-on crouch.
  • Nano Banana 2 kept the person recognizable and the mirror-photo feel, but omitted the requested heels.