AI influencer: one consistent face across dozens of photos (without drift)

Create an AI persona you can reuse across many photos — with the techniques that actually maintain consistency over dozens of generations:

STEP 1 — BUILD A CHARACTER SHEET (do this once, save it):
Create a text document with every fixed attribute. This is your 'character DNA' that goes into every prompt:
• Face: [specific eye color, shape], [nose], [lip shape], [jawline], [skin tone + undertone]
• Hair: [color, length, texture, part direction]
• Distinctive: [freckle pattern, birthmark, scar, dimples — 2-3 unique features]
• Body: [build, height relative to frame, posture]
• Age range: [specific, e.g. late 20s]
Save this as a doc. Copy the same description into EVERY prompt.

STEP 2 — GENERATE YOUR ANCHOR IMAGE:
'Photorealistic portrait of [PASTE YOUR FULL CHARACTER SHEET DESCRIPTION], neutral expression, neutral grey background, flat even daylight, shot on 85mm f/1.4 lens, natural skin texture, pores visible, sharp focus on the eyes --ar 4:5 --style raw --no makeup, accessories'

Generate 4, pick the one that best matches your character sheet. This is your ANCHOR — every future reference comes from this image.

STEP 3 — REUSE IN NEW SCENES:
In Midjourney: paste your anchor image URL + --cref [url] --cw [value]. In Flux/Nano Banana: upload the anchor as character reference.

'[PASTE CHARACTER SHEET DESCRIPTION] [doing activity] in [location], [outfit], [time of day lighting], candid and natural, photorealistic, natural skin texture --ar 4:5 --style raw'

THE --CW VALUE MAP (what each range actually controls):
• --cw 100: Maximum face AND style preservation — same face, same 'vibe.' Use for selfie-style content where the person should look identical.
• --cw 75-90: Strong face preservation, moderate style flexibility — same face, different outfits and settings work well. THE SWEET SPOT for most AI influencer content.
• --cw 50-70: Face similarity maintained but allows more variation — good for 'inspired by' or age/style shifts.
• --cw 0-40: Only general characteristics preserved — useful for 'same energy, different person' or family resemblance.

FIXING CHARACTER DRIFT (the biggest problem after 5+ generations):
Drift happens when you reference outputs-of-outputs instead of the original. Each generation compounds small changes until the face has shifted significantly. The fix:
1. ALWAYS reference your original anchor image, never a downstream generation
2. Every 5-10 generations, compare back to the anchor. If drift is visible, regenerate using the anchor directly.
3. If a scene turned out great but the face drifted, use inpainting to fix JUST the face using the anchor as reference
4. Keep a 'consistency gallery' of your best 3-4 outputs that nail the face — use these as multi-reference when available

⚠️ ETHICAL NOTE ON AI PERSONAS:
If you're posting AI-generated 'person' content publicly, major platforms (Meta, TikTok, YouTube) increasingly require disclosure that content is AI-generated. The FTC has also signaled enforcement against undisclosed AI personas used for endorsements. Best practice: include 'AI-generated' in the bio/profile, not buried in hashtags. This protects you legally and builds trust — audiences that know it's AI and follow anyway are more engaged than audiences that feel deceived.

Tips: keep ONE consistent feature list in every prompt — do not paraphrase or abbreviate it between sessions; use the same camera/lens specification across all generations for consistent 'photography style'; if building a following, develop 3-4 signature outfits/settings that become recognizable.