You typed something like "a knight in a forest, masterpiece, 8k" and got back a picture that looks like a sticker pasted onto a grey wall. Nothing is wrong with the model. The problem is that almost nobody describes light, and light is what makes a rendered scene read as a place instead of a shape. If you are new to this, start at aiimagegenerators.net and run two or three quick tests so you can see the difference in your own output rather than taking my word for it.
Light has a direction, and most prompts ignore it
Every prompt that omits light direction leaves the model to guess, and it usually guesses "soft light from everywhere." That is why so much beginner output looks the same. Name the direction and the scene changes immediately:
- Side lighting from left or right carves out shape and texture. Great for faces, wood, fabric, and anything with surface detail.
- Backlighting puts the subject between you and the source. It creates rim glow and silhouette, which is dramatic but hides facial detail.
- Top light drops shadows under the eyes and chin. Real, but rarely flattering.
- Front light is flat and honest. Useful for product-style shots, dull everywhere else.
Try the same prompt twice, once with "lit from the left side" and once with "backlit by a low sun." The difference is bigger than swapping half the adjectives in your prompt.
Hard or soft: light quality changes the mood
Direction is half the story. The other half is how big the light source is relative to the subject. A small source gives hard light: crisp shadow edges, strong contrast, a slightly harsh feel. A large source gives soft light: gradual shadow falloff, gentler transitions, a look most people read as "professional."
In prompt language, "hard light, sharp shadows" and "soft diffused light, gentle shadows" are clear instructions. "Dramatic lighting" is not — it means something different to everyone, and the model splits the difference into mush. Vague adjectives are where depth goes to die.
Time of day is really a lighting setting
Instead of describing the light directly, you can describe the hour and let the model do the work:
- Golden hour: warm, low, long shadows. The safest attractive default outdoors.
- Overcast: enormous soft source covering the whole sky. Even, cool, quietly moody.
- Blue hour: dim blue ambient light, city lights starting to read as highlights.
- Midday: harsh top light with short, dark shadows. Honest but rarely pretty.
Time of day also sets colour temperature without you saying anything about colour. That is efficient prompting: one phrase doing two jobs.
Colour temperature is a mood dial
Warm light — candle, fire, tungsten, sunrise — reads as intimate and nostalgic. Cool light — shade, fluorescent, moonlight, storm — reads as clinical or lonely. Mixed temperature is the interesting one: a warm interior seen through a cool window, or a warm fire against a cold blue snowfield. That contrast is what makes painters like Hopper feel alive, and it works in generated art for exactly the same reasons.
Name the source, not just the temperature. "Lit by a single candle" carries direction, quality, colour, and scale at once. "Warm lighting" carries almost nothing.
A formula you can reuse
When a scene comes out flat, rebuild the lighting clause in this order:
- Source: window, candle, neon sign, overcast sky, studio softbox.
- Direction: left, right, behind, above, three-quarter front.
- Quality: soft and diffused, or hard with sharp shadows.
- Time or environment: late afternoon, dusk, interior at night.
A clause like "soft window light from the left, late afternoon" is short, specific, and impossible for the model to misread. Compare it to "beautiful cinematic lighting" and you can see why one produces depth and the other produces wallpaper.
Indoor and outdoor setups behave differently
Outdoors, you are working with one dominant source and a lot of bounce from ground and surroundings. That makes daylight forgiving: point your subject toward the sun and the rest mostly sorts itself out. Indoors, you have a room full of small competing sources, which is why interior scenes so often come out looking flat and slightly yellow. If you want an interior that reads as real, give it one obvious source — a window, a lamp, a fire — and let the rest of the room fall into shadow. Rooms lit evenly from every direction look like furniture showrooms, not photographs.
Three mistakes that flatten everything
Stacking contradictions. "Soft light" plus "sharp dramatic shadows" plus "backlit" plus "even front light" gives the model four conflicting jobs. Pick one plan and commit.
Describing the subject and forgetting the room. Light bounces. If you want bounce, say so: "light bouncing off a pale wall to the right." Without any bounce, shadows go to pure black and the image loses information.
Chasing style words instead of light words. "Cinematic" is a result, not an instruction. Ask for the thing that causes it — anamorphic flare, backlit haze, low-key shadow — and you will get the look far more reliably.
Run the experiment on your next three generations: keep the subject identical, change only the light clause, and line up the results. The pattern will be obvious within ten minutes, and it will improve everything you make after that. Head back to the generator and try a side-lit portrait next to the same portrait backlit — same prompt, one variable.