Adobe’s prompt documentation contains one hard instruction and four ways of thinking about everything else. The instruction is short enough to memorise; the interesting part is how Adobe teaches the rest — by ladder, writing the same scene three times and adding one kind of detail at each pass. That is a more useful structure than any prompt formula, and it is what the worked examples below copy.
The one rule Adobe publishes
It is a single sentence: “Use at least three words in your prompts, and avoid using words like generate or create. Stick to simple, direct language with a subject, descriptors, and keywords.”
Three parts, then. A floor of three words. No instruction verbs aimed at the tool. And a shape — subject, descriptors, keywords.
The ladder Adobe demonstrates
Adobe never states that ladder as a rule, but it demonstrates it three times. The cat version is the shortest.
| Pass | Prompt Adobe publishes |
|---|---|
| 1 | Image of a cat. |
| 2 | Image of a cat, digital painting. |
| 3 | Cat with bold stripes, digital painting, 2 colors, sitting by the window, shallow depth of field, cyberpunk fantasy. |
Pass two adds a medium. Pass three adds countable detail — a stripe pattern, two colours — plus placement, a camera term and a genre. Nothing is removed at any point.
The four headings Adobe sorts prompts under
Each of the four sections carries its own claim about what a prompt should contain. Read in order, they are the closest thing Adobe offers to a method.
Be specific is where the three-word rule lives, and its examples are ordinary scenes: “A fluffy cat sitting on a windowsill looking out at a cityscape.”
Be descriptive argues that “the more detailed your description, the more likely you are to get images that match your vision”. Its examples name materials and light: “Woman with long flowing hair and blue eyes, white dress, field of flowers, realistic picture.”
Be original asks you to “infuse prompts with your unique perspective by incorporating feeling, style, and lighting elements”. Its ladder ends at Studio shot of futuristic high heels, warm red blue gel lighting, shallow depth of field, mechanical materials as props — the lighting phrase is what the last pass added.
Be empathetic is the one people skip: “use words like love, gentle, and playful to generate a heartwarming image, or powerful, strong, and uplifting to create an inspiring one”. Mood as a prompt field, and Adobe’s own example carries it — “Tender moment of a child sharing a heartfelt conversation with their elderly grandparent, love, wisdom.”
The settings that carry the rest of the instruction
Everything above is text. The names below are the controls that text sits next to, copied from Adobe’s generate-images page and worth using exactly as written.
| What Adobe calls it | What Adobe lists |
|---|---|
Model | Firefly Image 3, Firefly Image 4, Firefly Image 4 Ultra, Firefly Image 5 |
Aspect ratio | Landscape (4:3), Portrait (3:4), Square (1:1), Widescreen (16:9) |
Content type | Art, Photo, or chosen from the prompt |
Visual intensity | a slider |
Composition | a reference image plus a Strength slider |
Styles | a reference image plus a Strength slider |
effects | optional style refinements |
Color and tone, Lighting, Camera angle | optional refinements |
Prompt features | Prompt suggestions, Prompt enhancement |
Content type
Three states: Art, Photo, or letting Firefly pick from the prompt. It overrides part of what the prompt asked for, which is why it is so often left wrong.
Aspect ratio
Four are offered: Landscape (4:3), Portrait (3:4), Square (1:1) and Widescreen (16:9). A fifth, Vertical (9:16), exists only for the Firefly Image 4 models, which is one of several differences set out in the Firefly Image 5 breakdown.
Styles, effects, color and tone, lighting, camera angle
Styles takes a reference image and a Strength slider controlling how closely the result follows it. effects, Color and tone, Lighting and Camera angle are leftovers from the prompt, promoted into switches.
Prompt features
Two toggles sit next to the Generate button: Prompt suggestions and Prompt enhancement. Adobe documents enhancement for the Firefly Image 4 models only, and adds that it “is available only for prompts written in English”. Afterwards a Show enhanced button appears, and Show original takes you back.
Example: the cat prompt, built in three passes
Take Adobe’s cat ladder and write the settings beside it. Pass one is deliberately bare.
Image of a cat.Model Firefly Image 3, Content type Art, Aspect ratio Square (1:1). Result: Adobe’s table says Firefly Image 3 returns “Four image variations”, so three words already produce four candidates.
Cat with bold stripes, digital painting, 2 colors, sitting by the window, shallow depth of field, cyberpunk fantasy.Same settings, one change: Model Firefly Image 4. Result: “Four image variations are generated”. The two passes differ only in the words after the opening phrase and one model choice, which makes this the cheapest way to see what descriptors buy.
Example: the clay art prompt, built in three passes
Adobe’s second ladder puts its descriptors in different places, which is why it is worth running alongside the first.
Clay art, a little girl wearing a raincoat standing in the rain.Model Firefly Image 5, Content type Art, Aspect ratio Portrait (3:4). Result: Firefly Image 5 generates “A single image”, so this model returns one candidate rather than four, at a published 10 credits per generation — the rate card sits in the Firefly pricing article.
Clay art, a little girl with a big smile wearing a raincoat standing in the rain with a backdrop of an iceberg.Same settings. Two things were added: a facial expression, and a background. Neither opened a new sentence — both were folded into the one already there, which is Adobe’s habit across every ladder it publishes. Moving between models inside the app is covered in the Adobe Firefly walkthrough.
What Adobe does not publish about prompts
A maximum length. The image prompt page sets a floor of three words and no ceiling. The only published number is on the video prompt page, which gives a 1800-word limit for video generation.
A formula. There is no required field order, no template and no list of mandatory slots. The subject-descriptors-keywords shape is described, not enforced.
A success rate. No hit rate, no score, no measurement of how often a prompt produces something usable.
Prompt enhancement for Firefly Image 5. Documented only for the Firefly Image 4 models, and only for English prompts.
Prompts also travel through the API, where the same text meets different limits and a different model identifier, documented in the Firefly API guide.
Frequently asked questions
How long should an Adobe Firefly prompt be? Adobe’s floor is three words and the image page sets no maximum. Longer is not the goal: the third pass of Adobe’s own ladders adds a handful of descriptors, not a paragraph.
Do I need to write “create” or “generate” in a Firefly prompt? No, and Adobe asks you not to — both words are named as ones to avoid.
Do the same prompts work on every Firefly model? No. Adobe notes that generation settings vary by model, and the models return different numbers of images: four variations on Firefly Image 3 and Firefly Image 4, a single image on Firefly Image 4 Ultra and Firefly Image 5.
Where do the settings live? In the left panel of the Generate image page, under General settings, plus the Prompt features toggles next to the Generate button.