Tutorials Get updates

Kill the Generic AI Look

The problem

You can spot it from across the room: glossy skin, perfect symmetry, teal-and-orange everything, a soft glow on every edge. It is the default look every model falls back to when the prompt does not give it a specific medium to imitate. Words like "cinematic", "epic" and "moody" do not fix it. They are the cause. The model has seen millions of images tagged "cinematic" and averaged them into exactly this look.

The fix is to speak the language the training data was actually labelled with: film stocks, lenses, lighting setups, print processes.

Steps

  1. Name a real medium. Replace "cinematic" with a physical thing: "35mm film still", "medium format photograph", "Super 8 frame", "risograph print", "gouache on paper". The model now has a specific texture, grain and colour response to imitate.
  2. Name a real lens and distance. "85mm portrait, shallow depth of field" or "24mm wide, deep focus". Lens words control distortion and how the background falls off, which is most of what makes an image feel photographed rather than rendered.
  3. Name one light source. Not "dramatic lighting". Say where it comes from and what it is: "single window light from the left", "overhead fluorescent", "bare bulb practical". One source gives you shadows that make sense.
  4. Allow imperfection. Add one flaw the real medium would have: "slight motion blur", "visible grain", "lens flare from the practical", "expired film colour shift". Perfection is the tell.
  5. Turn the model's taste down. In Midjourney, add --raw and drop --stylize to 50 to 150. High stylize is the model adding its own flourishes back in. Raise it only when you want them.
  6. Cut the mood words. Read your prompt once more and delete every adjective that does not describe something a camera or a painter could actually do. Usually that removes a third of the prompt and improves the result.
35mm film still: a woman walks past a tram stop on a city street at dusk under a streetlight.35mm film still
Risograph print in two colours: a woman stands at a tram stop as a tram approaches, with halftone texture.Risograph print
Gouache painting: a woman sits alone in a tram shelter on a cobbled street at dusk under one streetlight.Gouache painting
Super 8 frame: a woman waits at a tram stop at dusk, with a warm light leak at the edge of the frame and heavy grain.Super 8 frame
Same woman, same tram stop, same light. Only the medium word changes. Generated with Adobe Firefly.

Worked example

Generic:

epic cinematic portrait of an old fisherman, dramatic lighting, hyper detailed, 8k, masterpiece

Specific:

old fisherman mending a net on a wooden dock, early morning fog, medium format colour photograph on expired Portra film, soft overcast light from the right, 80mm, waist-up, slight grain and faded blacks --ar 4:5 --raw --stylize 80

Notice that the second prompt has zero quality words. No "detailed", no "8k", no "masterpiece". Those words do not raise quality. They pull toward the average.

Generic result: a glossy, high-contrast portrait of an old fisherman at night with a lantern, lit in teal and orange.Generic prompt
Specific result: an old fisherman mending a net on a wooden dock in soft morning fog, muted colour and fine grain.Specific prompt
The two prompts above, side by side. Generated with Adobe Firefly to show the principle, which holds in any image model.

Words to swap

Instead ofTry

  • cinematic35mm film still, anamorphic frame
  • moodysingle hard source, deep shadows, underexposed by one stop
  • epic24mm wide, low angle, subject small in frame
  • dreamysoft diffusion filter, backlit haze
  • vintageexpired Kodak Gold, faded blacks, slight magenta cast
  • hyper detailed, 8kdelete

Your move

Try this today

Take any prompt you have that contains "cinematic". Replace it with a film stock and a lens. Generate. Then remove every remaining quality adjective and generate again. Compare the three. Most people never go back.

Stop waiting – start creating!

All 16 tutorials