One Character, a Whole Summer: Reusing Anime Prompts Like Function Calls

Most "AI art prompt" lists are a pile of disconnected one-shots. I wanted something closer to how you reuse code: one base prompt, a swappable parameter, and a reproducible way to keep the same character across every run.

So I ran a small experiment over a weekend. One original character, four summer scenes, one variable changed at a time. Here is the setup, the prompts, and what moved the output.

I used PixAI, a browser-based, anime-first generator, but the method transfers to any model. If it is new to you, the beginner's guide covers the panel and settings.

The setup: treat the character as a variable

Think of a prompt as a function call. Most of it is fixed scaffolding (scene, lighting, shot type, style), and one part is the argument you vary: the character description. Keep that part isolated and you can reason about changes instead of guessing.

1girl, <CHARACTER DESCRIPTION>, <scene>, <shot type>, <lighting>, anime style, <palette>

Swap <CHARACTER DESCRIPTION> for your own OC and leave the rest alone. That one discipline is what turns a prompt list into a reusable set.

A model detail worth knowing

If you come from Stable Diffusion, one gotcha. Numeric tag weighting like (silver hair:1.3) is an SDXL / SD1.5 convention. On a DiT model such as PixAI's Tsubaki.2 it does nothing, so you steer with plain descriptions and word order instead of weights. It is the same failure mode as calling an API with a query param it silently ignores: confirm what the model honors before you build on it.

The four scenes

Same character, four settings. Copy a block, replace the character line, and run.

Beach:

1girl, honey-blonde hair in a ponytail, aqua-green eyes, light blue summer sundress, wide-brim straw hat, cowboy shot, sunny sandy beach, turquoise ocean, clear blue sky, soft summer sunlight, anime style, warm bright colors

Night festival:

1girl, honey-blonde hair in a side ponytail with a ribbon, floral summer yukata, holding a round paper fan, cowboy shot, lively festival street at night, food stalls and glowing paper lanterns, warm bokeh lights, anime style, festive summer evening

Fireworks:

1girl, honey-blonde hair in a ponytail, summer yukata, seen from behind looking back over her shoulder, medium shot, riverbank at night, large colorful fireworks in the dark sky, floating lanterns on the water, glowing reflections, anime style, deep blue and warm gold tones

Thumbnail:

1girl, short coral-pink bob, star hair clip, big energetic smile, close up, facing the viewer, bright summer background with palm leaves and blue sky, punchy saturated colors, empty space on one side for text, anime style, eye-catching thumbnail

What moved the output

Changing one variable at a time made cause and effect readable:

• Shot type is a real control. cowboy shot, close up, and wide shot change how much of the scene renders. The composition breakdown covers each one and when to use it.
• Give the hands a job. A paper fan or a candy apple in the hands cut down finger artifacts more than any negative wording did.
• Backlight the subject at night. Keeping the character slightly backlit kept the fireworks the brightest element and stopped the face from competing with the sky.
• Reference image equals reproducibility. To hold the same face across all four scenes, I fed one good result back in as a reference. That is your fixture. Without it, each run drifts into a slightly different person.

Takeaway

Prompt reuse is a discipline, not a library. Fix the scaffolding, isolate the character variable, pin consistency with a reference image, and a handful of base prompts covers a whole season.

Try the setup in PixAI and swap in your own character.