---
title: Write better prompts
display: 'Write better *prompts*'
nav_title: Prompt images
summary:
  - "Whatever your prompt leaves out, the model invents, so write image prompts that leave nothing important to chance."
  - "Write six parts in order: subject, composition, style, light, text, constraints, with exclusions last."
  - "About 100 words is usually enough; to change a result, edit one part and keep the rest word for word."
description: Write image prompts in six parts (subject, composition, style, light, text, constraints) and adapt them to the model you use.
section: methods
order: 20
type: guide
level: Start
roles: [creator, artist, art-director, marketer, team-lead, developer]
time: { read: 8, do: 15 }
last_verified: 2026-11-11
models_checked: ["GPT Image 2.5 Sunburst", "Nano Banana 2.1", "FLUX.3 Image", "Seedream 5.0 Flash", "Kling O3 Text to Image"]
hero: methods/prompt-images/hero-liseron-thin-vs-structured.webp
related: [text-and-typography-in-images, generate-images]
video: tutorials/prompt-images.mp4
video_caption: Write better prompts
about: "The film writes one prompt in six parts on GPT Image 2.5 Sunburst, a game prop with a line of stenciled text, and generates it. Then it changes only the light and generates again, and ends with Prompt Spark turning a few words into a first draft to edit. The models and settings in this lesson are as of November 2026."
chapters:
  - { at: 0, title: "Six parts", image: "/media/images/methods/prompt-images/ch-01-six-parts.webp" }
  - { at: 65.1, title: "One part at a time", image: "/media/images/methods/prompt-images/ch-02-one-part-at-a-time.webp" }
  - { at: 80.2, title: "Prompt Spark", image: "/media/images/methods/prompt-images/ch-03-prompt-spark.webp" }
thumb: methods/prompt-images/thumb.webp
steps:
  - title: "Write the subject"
    text: "Create › Images › GPT Image 2.5 Sunburst. Start the Prompt box with what's in the picture: material, color, how many. Labels such as Subject: are optional but keep a long prompt easy to edit."
  - title: "Write the composition"
    text: "The camera angle and where things sit; ask for empty space here if a title goes on later."
  - title: "Name one style"
    text: "One style only: two styles fight and the model averages them."
  - title: "Describe the light"
    text: "Its source, direction and color, and what the shadows do."
  - title: "Add the text, in quotes"
    text: "Only if the image needs it: the exact words in quotes, where they go and how they look."
  - title: "Say what must not appear"
    text: "End with a plain sentence of exclusions, since most models have no negative prompt field. The whole prompt is about a hundred words."
    prompt: |-
      Subject: a VOLTARIS supply drop crate, a hexagonal armored crate in matte gunmetal with burnt-orange side panels, its lid open on three glowing cyan power cells, a chunky handle on each side, on a scuffed steel landing pad.
      Composition: three-quarter view from slightly above, the crate in the left half of the frame, open sky and floating islands on the right.
      Style: stylized 3D game art, like the best free-to-play hero shooters.
      Light: late-afternoon sun from the right, a cool cyan glow from the power cells, long soft shadows across the pad.
      Text: stenciled in white on the front panel, the words "SUPPLY DROP 07".
      Constraints: no other text or logos, no characters, no weapons.
  - title: "Generate and check"
    text: "Settings: Image Count 1, Dimensions 2048 × 1152. The cost shows on Generate. Tick every part of the prompt in the image, and check the text against the quotes."
  - title: "Change one part at a time"
    text: "Select the Light line, replace it and keep every other part word for word, then generate again."
    prompt: |-
      Light: cold blue moonlight from above, a strong cyan glow from the power cells, deep shadows across the pad.
  - title: "Get a head start with Prompt Spark"
    text: "Type a few words and click Prompt Spark, the sparkle button under the box. Edit its draft to match your brief before you generate."
agent_steps:
  - title: "Connect the Scenario MCP"
    text: "Connect your agent and sign in once at [mcp.scenario.com](https://mcp.scenario.com). Then tell it which Scenario team and project to use, so every file lands there."
  - title: "Give the agent the files"
    text: "Nothing to upload: the prop is made from the prompt alone, with no reference."
  - title: "Brief the agent"
    text: "The six-part prompt, with its labeled lines kept as they are."
    prompt: |-
      Use GPT Image 2.5 Sunburst (model_openai-gpt-image-2-5-sunburst): width 2048, height 1152, numOutputs 1. No reference. Quote the cost with a dry run first and wait for my go.

      Prompt, word for word:
      Subject: a VOLTARIS supply drop crate, a hexagonal armored crate in matte gunmetal with burnt-orange side panels, its lid open on three glowing cyan power cells, a chunky handle on each side, on a scuffed steel landing pad.
      Composition: three-quarter view from slightly above, the crate in the left half of the frame, open sky and floating islands on the right.
      Style: stylized 3D game art, like the best free-to-play hero shooters.
      Light: late-afternoon sun from the right, a cool cyan glow from the power cells, long soft shadows across the pad.
      Text: stenciled in white on the front panel, the words "SUPPLY DROP 07".
      Constraints: no other text or logos, no characters, no weapons.
  - title: "Change one part"
    text: "Only the Light line changes, so you know what made the difference."
    prompt: |-
      Run the same prompt again with the same settings, but replace the Light line with this one, word for word, and keep every other line exactly as it was:
      Light: cold blue moonlight from above, a strong cyan glow from the power cells, deep shadows across the pad.

      Quote the cost with a dry run first and wait for my go. Then show me the two results side by side.
  - title: "Check and redo"
    text: "Accept when every part of the prompt is in the image and the text reads exactly \"SUPPLY DROP 07\". Reject a missing or changed part, misspelled or extra text, or a mixed style, and rewrite only the part that didn't land."
    prompt: |-
      For each result, go through the six parts and tell me which landed and which didn't, and whether the front panel reads exactly SUPPLY DROP 07 with no other text or logos. Don't rerun anything without a new quote.
  - title: "File the results and report"
    prompt: |-
      Create a collection called "VOLTARIS supply crate" and add both results to it. Report the model, the settings, every asset ID and the total cost spent.
updated: 2026-11-11
canonical: https://learn.scenario.com/methods/prompt-images
---

Last updated: 2026-11-11

# Write better prompts

> Write image prompts in six parts (subject, composition, style, light, text, constraints) and adapt them to the model you use.

In short:

- Whatever your prompt leaves out, the model invents, so write image prompts that leave nothing important to chance.
- Write six parts in order: subject, composition, style, light, text, constraints, with exclusions last.
- About 100 words is usually enough; to change a result, edit one part and keep the rest word for word.


Whatever a prompt leaves out, the model invents. This lesson writes one image prompt in six parts on GPT Image 2.5 Sunburst, for a game prop with a line of stenciled text, then changes a single part to see what it does, and ends with Prompt Spark turning a few words into a first draft.

:::note As of November 2026
The models, settings and screens in this article are the ones Scenario offered in November 2026. New models and updates arrive often, so check the app for newer options before you choose.
:::

:::steps Quick version
1. Write the prompt in six parts, in this order: subject, composition, style, light, text, constraints.
2. Put any words the image must show in quotes, and end with what must not appear.
3. Keep it to about 100 words, then click **Generate**.
4. To change the result, edit one part and keep the rest word for word.
5. For a head start, type a few words and click **Prompt Spark**, then edit its draft to match your brief.
:::

## Before you start

- **Where it lives:** **Create** › **Images** opens the image panel. The **Prompt** box sits under the model, on the left.
- **Have a brief.** Know what the image is for, what must be in it, and what must not. The prompt turns the brief into words.
- **The model matters.** This lesson uses GPT Image 2.5 Sunburst, built for detailed, precise images. To pick one, see [Image models guide](https://learn.scenario.com/images/image-models-guide.md).

## Why a short prompt isn't enough

The film starts from a four-word prompt:

```
a VOLTARIS supply crate
```

The model filled every gap with its own guesses: a closed rectangular crate in a dark warehouse, a big logo, and lines of text nobody asked for.

![The result of the short prompt: a closed gunmetal crate in a warehouse, covered in invented logos and slogans](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/bab4b4a7356e719419dfdf7db6f817534b50a8571b145689a559c5c962090a6c.webp)

The brief was something else: an open hexagonal crate with glowing power cells, on a landing pad in the sky, in the game's style, with one line of text. Every one of those decisions has to be in the prompt.

## Step 1: Write the subject

Start with what is in the picture, with the details that matter: material, color, how many. Name the thing, then describe it as if the reader had never seen it.

![The Prompt box with the subject being typed on GPT Image 2.5 Sunburst](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/1e8a4e9e738131b69e56d146bd5dbcf1cf5cbf267c56a209a2e75268560f8973.webp)

The film labels each part (**Subject:**, **Composition:** and so on). The labels are optional, but they keep a long prompt easy to read and easy to edit later.

## Step 2: Write the composition

Then the composition: the camera angle, and where things sit in the frame. If you need empty space, for a title or a logo added later, ask for it here: "open sky on the right" keeps that side free.

## Step 3: Name one style

Then the style. Name only one, because two styles fight. "Stylized 3D game art" gives a clear target; "stylized 3D game art, watercolor, photorealistic" makes the model average them.

## Step 4: Describe the light

Then the light: where it comes from, and how it falls. It's the quickest way to set the mood. Name the source, its direction, its color, and what the shadows do.

## Step 5: Add the text, in quotes

Add text only if the image needs it. Put the words in quotes, so the model writes exactly those, and say where they go and how they look (stenciled, painted, engraved).

## Step 6: Say what must not appear

Last, what must not appear. Most models have no negative prompt field, so write it as a plain sentence at the end: "no other text or logos, no characters, no weapons."

![The full prompt in six labeled parts, ending with the constraints](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/eddb3aa844ed13ee4f0200f0e43443475269347ba67f6f45ad739300950e74cb.webp)

The whole prompt the film writes:

```
Subject: a VOLTARIS supply drop crate, a hexagonal armored crate in matte gunmetal with burnt-orange side panels, its lid open on three glowing cyan power cells, a chunky handle on each side, on a scuffed steel landing pad.
Composition: three-quarter view from slightly above, the crate in the left half of the frame, open sky and floating islands on the right.
Style: stylized 3D game art, like the best free-to-play hero shooters.
Light: late-afternoon sun from the right, a cool cyan glow from the power cells, long soft shadows across the pad.
Text: stenciled in white on the front panel, the words "SUPPLY DROP 07".
Constraints: no other text or logos, no characters, no weapons.
```

About a hundred words is usually enough. Past that, a prompt gets harder to control.

## Step 7: Generate and check

Under **Settings**, the film makes one image (**Image Count** 1) at **Dimensions** 2048 × 1152. Click **Generate**; the cost shows on the button.

![The prompt and the Settings: Image Count 1, Dimensions 2048 x 1152](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/091aa5bf9c7c1097483292571a0be69f55fe2a8c25e5848f4574b77e6dbf6d40.webp)

Everything the prompt names is in the picture, and the text matches the quotes.

![The result: the open hexagonal crate with three cyan power cells and SUPPLY DROP 07 on the front, floating islands on the right](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/a1723eceaf5312262316bb37c8cf96ca9b8dcffb9efe4465bc1c458c5417b9b0.webp)

:::check
**Accept when:** you can tick every part of the prompt in the image: the subject and its details, the framing, one style, the light, the exact text, and nothing from the constraints.
**Reject when:** a part is missing or changed, the text is misspelled or extra text appeared, or the style is mixed. Find the part of the prompt that didn't land and rewrite only that part.
:::

## Step 8: Change one part at a time

To change something, edit one part and keep the rest word for word. Here, just the light. Select the **Light** line.

![The Light line selected in the Prompt box](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/2c27def666e99fa29037366f5fbde52583e5774280daed78c89e3a7327ec77dc.webp)

Replace it with the new light:

```
Light: cold blue moonlight from above, a strong cyan glow from the power cells, deep shadows across the pad.
```

![The prompt with the new moonlight Light line and every other part unchanged](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/0b16b89d2089c91e9db2c829b17eaa7c4c218b011afb4e96dad55c5e484c6fae.webp)

Generate again. Since nothing else changed, you know what made the difference: the same crate, the same framing, the same text, now at night.

![The two results side by side in the Session: the crate at night under the moon, and the crate in the late-afternoon sun](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/5d8fdd90b8e10a0dc651dbc5df451393dd80d5c4081d9f6607bd9591554db8b0.webp)

## Step 9: Get a head start with Prompt Spark

When the box is empty, it reads: "Describe your scene or subject, and let prompt tools generate, complete, or translate it." Type a few words, then click **Prompt Spark**, the sparkle button under the box.

![A few words in the Prompt box, with the cursor on the Prompt Spark button](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/a4f142e66e9ff3b3f758debd627a610a496eb600daeabdbcba236b02e297e174.webp)

Prompt Spark writes a full prompt for you. It fills the gaps with its own guesses. From "a VOLTARIS supply crate", it chose a photorealistic, rectangular, matte-green military crate, none of which is in the brief.

![The draft Prompt Spark wrote: a photorealistic industrial supply crate with matte-green panels](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/94387cca4d6ed7e1a406daa31f01512f6db7aa39c0fac9246671e8f94bf781b7.webp)

So edit the draft to match your brief before you generate: fix the subject, the style and the light, add the text in quotes, and end with the constraints.

## The six parts at a glance

| Part | What to write | Example from the film |
|---|---|---|
| Subject | What is in the picture: material, color, how many | A hexagonal armored crate in matte gunmetal, three glowing cyan power cells |
| Composition | Camera angle, where things sit, empty space | Three-quarter view from slightly above, crate in the left half |
| Style | One style only | Stylized 3D game art |
| Light | Source, direction, color, shadows | Late-afternoon sun from the right, long soft shadows |
| Text | The exact words in quotes, where and how | Stenciled in white on the front panel, "SUPPLY DROP 07" |
| Constraints | What must not appear, as a plain sentence | No other text or logos, no characters, no weapons |

## Another example: one part at a time

The same method on a different brief, a leather bag in a workshop. A thin prompt gives a busy workshop with no clear subject; the six-part prompt gives the bag, framed and lit as the brief asks.

![Left, the prompt "a leather workshop": a busy workshop with no clear subject. Right, the six-part prompt: the burgundy bag on a workbench](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/4573b3f1a42ca8ccb02b6ceb64daf2498d4b6cef7d20a72498a68e22597a29b0.webp)

Built up one part at a time, each image adds one part to the one before. With the subject alone, the model adds a cluttered background and invents labels on the boxes. The constraints at the end clear them away.

:::gallery The bag prompt, one part at a time
![The burgundy bag on a cluttered workbench, with invented labels on boxes behind it](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/646a4b294b74b762f888a9930cf1452ac37a798fb79558ab7d53b783a5a4f15b.webp "Subject")
![The bag from a new angle against a plaster wall](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/22ca47e090d797994f2c71a22d2dbd71e157b8e9b0a298861c1614a074e07fcc.webp "+ Composition")
![The bag with the style part added](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/490363538c8a88daf5d817e0a82d11877bb73fa6948c86e942b698d6dd874c08.webp "+ Style")
![The bag in soft window light](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/d73529b77c218354e9799ed4af8d93389b34c31cfd8d0301b3742341fe3ade0f.webp "+ Light")
![The bag with a brass plaque on the wall](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/0d643ceabe1c98d55f27a8686d049fdb21b1749596ed19a4f6650d240e98a672.webp "+ Text")
![The bag with the plaque and nothing else written in the scene](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/b2b908e59b371091e4a8d1b4379ba316c6cb390a91813e770151c5b4bd9e59db.webp "+ Constraints")
:::

Quotes matter for text. Without them, the model treats the words as a description and may add its own punctuation:

![Two brass plaques: ATELIER LISERON. with a stray period without quotes, and ATELIER LISERON exactly with quotes](https://ufgc0c9mhl4ueq76.public.blob.vercel-storage.com/academy/15847c86054ad736dd86724719f3c9c2ff06745ee2e022063d782ce93627a8be.webp)

## Tips

- **Put the most important thing first.** Many models give more weight to what comes early in a prompt.
- **Be concrete.** "Three glowing cyan power cells" beats "some glowing parts".
- **One idea per sentence.** Short sentences are easier for the model and for you to edit.
- **Reuse a style paragraph.** Keep the same style and light lines for every image of a project, so they look like one set.
- **Translate when you need to.** The button next to **Prompt Spark** translates your prompt into English.
- **Turn an image into a prompt.** Upload an image to the prompt box and Prompt Spark writes a prompt that describes it, ready to edit.

## Common problems

| What you see | What to do |
|---|---|
| The model invented things you didn't ask for | Describe them yourself, or rule them out in the constraints |
| Two styles blended into a mess | Keep one style and delete the others |
| The text is misspelled or has extra punctuation | Put the exact words in quotes, keep them short, and say where they go |
| Extra text, logos or watermarks appeared | End the prompt with "no other text or logos" |
| You changed the prompt and can't tell what helped | Go back and change one part at a time, keeping the rest word for word |
| A long prompt is hard to control | Cut it back to about 100 words and move what matters to the front |
| Prompt Spark changed your idea | Treat its draft as a start: fix the subject, style and text to match your brief |

## What's next

You can write a prompt that leaves nothing important to chance. Next, change part of an image without starting over in [Edit images](https://learn.scenario.com/images/edit-images.md). For posters, labels and signs, see [Text and typography in images](https://learn.scenario.com/images/text-and-typography-in-images.md).

Source: https://learn.scenario.com/methods/prompt-images