Studio-background-style city street, matte-painting feel - prompt + the depth-separation trick

by @dianaparklab2270 in c/anime-lab · · Background Art

Background Art result. Trying to get that gorgeous painterly anime BACKGROUND look (think a quiet detailed street, the kind the characters walk through) rather than the usual AI mush-behind-the-character. Backgrounds in anime are PAINTINGS and should look painted, not photographed and not cel-shaded.

Prompt direction: anime background art, matte painting, detailed painterly environment, no characters, soft painted light, hand-painted texture, studio background style, atmospheric perspective. 'No characters' is important - it lets the model lavish all its detail on the environment instead of defaulting to a figure.

Depth trick: I generate it in layers - a foreground element (a lamppost, a railing), the midground street, and a hazy background - as separate gens, then composite with atmospheric haze between layers. Real anime BGs have strong atmospheric depth; a single flat gen does not. Stacking layers with haze = instant depth and that lived-in painterly feeling.

440 points · 6 comments

Comments

  • @leochen_art5063 · · 61 points

    'backgrounds are PAINTINGS, should look painted not cel-shaded' is the thing that unlocks good anime BGs. People apply their character cel-shade style to the background and it looks wrong because real anime NEVER cel-shades the background, it is a different artist with a different technique entirely.

  • @hanajohnson_creates6818 · · 48 points

    OP - the character department and the background department are literally different people with different training in a real studio, and they SHOULD look different. The BG is softer, more painterly, more atmospheric. Matching them is the amateur mistake.

  • @lunanguyenlab4622 · · 44 points

    The layer-with-haze depth trick is great. Atmospheric perspective (distant things hazier/bluer/lower-contrast) is the single biggest thing separating a flat AI background from a Makoto-Shinkai-grade one. The haze between layers sells the depth.

  • @yunawilson485 · · 33 points

    OP - exactly, and you cannot reliably get multi-layer atmospheric depth from one gen, the model flattens it. Generating foreground/mid/background separately and hazing between them gives you real control over the depth falloff.

  • @priyabecker_fx191 · · 38 points

    no-characters in the prompt to stop it defaulting to a figure is such a simple fix for 'why does my background gen keep putting a person in it', genius