Scene Storytelling — GPT Image 2 prompt template

When to use it

Use for storyboards, worldbuilding, live scenes, and emotional narrative frames. It sits in the Scenes & Storytelling group of the catalog, which covers storyboards, narrative scenes, livestream frames, and worldbuilding. If your brief is closer to one of the neighbouring genres, the related templates at the bottom of this page are the faster route — picking the right template matters more than perfecting the wrong one.

The template

Copy the block, replace everything inside brackets with your own specifics, and delete any line that does not apply to your brief. The template text is published in Chinese, as it appears in the upstream catalog; the structure is what carries the value, so translating the labels into your own language changes nothing about how it works.

生成[故事主题]场景图，发生在[时间+地点]。
主事件：[事件描述]，主角：[角色]，冲突点：[冲突]。
镜头语言：[广角建立镜头/中景叙事/特写]。
氛围：[紧张/温暖/悬疑]，色调：[冷/暖/高反差]。
输出：具备叙事张力的场景概念图。

JSON version (agent-friendly)

Every Scenes & Storytelling template shares one JSON block, which expresses the same kind of decisions as key–value fields rather than prose. Use it when a script, an agent or a form fills in the values, or when you want to change a single field between runs and see exactly what changed; swap the example values for your own.

{
  "type": "Narrative Scene",
  "story_context": "The exact moment an ancient seal breaks",
  "environment": "Crumbling stone temple overgrown with glowing blue vines",
  "action": "A young explorer dropping their torch as a massive beam of light shoots into the sky",
  "atmosphere": {
    "mood": "Awe-inspiring, terrifying",
    "lighting": "Blinding central light casting long dramatic shadows"
  },
  "camera": "Low angle shot, emphasizing the scale of the light beam"
}

How to fill it in

The block above asks you to make 8 explicit decisions before you send anything. That is the point of a template: the decisions you do not make are made for you, and in image generation the defaults are what produce work that looks generic.

Answer the five questions the block asks — who, where, when, conflict, emotion — in that order, since the conflict is what makes the frame a scene instead of an illustration of a place.

Then choose the framing deliberately. A wide shot sells the world, a close one sells the feeling, and the template will do either once you say which.

• Define who, where, when, conflict, emotion, and camera framing.
• Use scene details to support narrative rather than decoration.

A working order for the whole thing: fill the structural slots first (subject, layout, format), then the visual ones (palette, light, material), then the text that has to appear verbatim, and only then the mood words. Run it once, change one field, run it again — a template you iterate one variable at a time tells you which variable did the work, which a rewritten prompt never does.

Common mistakes

Generic fantasy backdrops are what a prompt returns when the scene has a place but no event: atmospheric, interchangeable and forgettable.

The other failure is burying the narrative cue outside the frame. If the story depends on a detail, that detail has to be visible in the composition you described.

• Avoid generic fantasy backgrounds.
• Keep narrative cues visible in the frame.

Example results

These are the public posts this template was distilled from: the creator shared the image and the prompt behind it, and the recurring structure across cases like these became the block above. Each card links back to its source, and a public post is not a licence to reuse the image.

• Moonlit Livestream Scene

  生成一张直播间的图片，直播间氛围是月下美女跳舞的画面，直播间有很多人评论

Related templates

• History & Classical Themes — Dynasty clothing, scroll narrative, poetry, and traditional motifs.
• Document & Publishing — White papers, manuals, encyclopedic plates, and page systems.
• Concept Product Breakdown — R&D boards, exploded views, mixed tasks, and special outputs.
• UI Screenshot System — High-fidelity app, web, dashboard, and social interface prompts.
• Infographic Engine — Structured diagrams, timelines, knowledge maps, and explainers.

Frequently asked questions

Can I use this template for a storyboard sequence?

Run it once per beat, keeping the who and where fixed and changing the conflict and framing. Generating several panels in a single image tends to average the beats together.

How much detail should a scene prompt carry?

Enough that every element serves the story. The template asks for scene details that support the narrative, which is a useful filter: if a detail does not carry meaning, it is competing with the ones that do.

Is the Scene Storytelling template free to use?

Yes. The template text comes from the open-source awesome-gpt-image-2 catalog by freestylefly, published under the MIT licence, and PicGens reproduces it with credit. Copy it, fill in the brackets and paste it into whichever tool you run GPT Image 2 in.

What is the JSON version of the template for?

It carries the same decisions as key–value fields instead of prose, which is easier for an agent or a script to assemble programmatically and easier to diff when you iterate. Use the plain-text block when you are writing by hand and the JSON block when something else is filling in the values.

Where do the example results come from?

They are public posts by creators who shared both the image and the prompt behind it, collected in the same open catalog. Each one links back to its source, and a public post is not itself a licence to reuse the image.