Your images look fake because you asked for perfect cinematic light, plastic skin, and movie framing. Make them less perfect. Five checks you run on every prompt before you publish.
Real phone photos rarely have perfect cinematic contrast. If your prompt says cinematic lighting, volumetric rays, rim light, golden hour, the model will paint a movie still. Direct the actual source: window on the left, overhead kitchen LED, harsh noon sun, one warm lamp. Add mixed color temperature or slight underexposure. Paste the checklist at the bottom onto your next prompt and remove every lighting cliché it flags.
02
02Skin
Skin: pores beat plastic
Most AI skin is too clean. Ask for visible pores, natural texture, slight redness, real specularity. Ban beauty filter, airbrushed, poreless, doll skin, plastic in the negatives. If you are directing a face, add one unflattering true thing (a crease, a shine on the nose, a stray hair). Perfect skin is the tell. Run the checklist's skin line before you generate.
03
03Framing
Framing: a little awkward crop reads human
Centered, full-body, rule-of-thirds-perfect frames look like stock. Direct a crop a real person would take: head slightly clipped, too much floor, subject off-center, phone tilt. Leave a quiet edge for type — never cover a face. If the shot looks like a poster, recrop in the prompt, not in a later editor.
04
04Not a movie
Stop forcing cinematic
The realism trick: look less like a movie still. Delete cinematic, anamorphic, film still, movie lighting, 8k, masterpiece, award winning. Replace with the camera that would actually be there: iPhone, 50mm at f/2.8, on-camera flash, available light. If it could be a paused Netflix frame, it will not pass as a creator photo.
05
05QC
Spot it: A/B yourself before you publish
Before the post goes out, sit two frames next to each other: the default AI version and the directed one. Ask: which one looks like a phone photo from this week? If you cannot tell in two seconds, the directed frame is still slop — rerun the checklist, change one variable, regenerate. Do this on the next three images you were about to post. That is the 30-minute drill.
Paste into Claude (or your image model) · run on every prompt
# ANTI-SLOP DIRECTION CHECKLIST — rewrite the prompt, then generate You are my image director. Rewrite the prompt I paste so it looks like a real photo, not AI slop. Do not add cinematic, 8k, masterpiece, volumetric light, or beauty-filter skin.
1. LIGHTING — Name a real source (window, overhead LED, noon sun, one lamp). Ban: perfect cinematic contrast, rim light, god rays, HDR glow. Prefer: mixed color temperature, slight underexposure, available light. 2. SKIN — Keep pores, texture, redness, real shine. Ban: plastic, poreless, doll, airbrushed, beauty filter. 3. FRAMING — Allow an awkward crop (clip a head, extra floor, off-center, slight tilt). Leave a quiet edge for type. NEVER place text or graphics on a face. 4. NOT A MOVIE — Delete cinematic / anamorphic / film still / movie lighting. Name the camera that would actually be there (iPhone, 50mm, on-camera flash). 5. NEGATIVES — Always end with: cartoon, CGI, plastic, bad anatomy, text, watermark. Add beauty filter + plastic skin for faces. Add over HDR for interiors. 6. GRAIN — Keep film grain. Do not add glow or sharpening that reads as HDR.
Return: (A) the rewritten prompt, (B) the negative line, (C) the three words you deleted that were slop. Then I generate. If I paste two images, tell me which is directed and why in one sentence.
Want this running for you?
Build it step by step inside AI Creative OS, with me in the thread.
