Blog
Convert a Midjourney prompt for Nano Banana Pro
Drop the parameters, move the aspect ratio to the setting, and expand the noun phrase into four ordered passes. A worked conversion, and the shortcut if you still have the picture.
A Midjourney prompt does not run well on Nano Banana Pro, because the two models read differently. Midjourney takes a short noun phrase plus parameters; Nano Banana Pro takes ordered description and ignores parameters entirely. The conversion is mechanical: drop every double-dash parameter, turn --ar into the aspect setting, and expand the phrase into four passes — composition, light, colour, materials.
The conversion, in five rules
| In your Midjourney prompt | What to do |
|---|---|
--ar 3:2 and other ratios | Delete it. Set the aspect ratio in the generation panel instead. |
--sref, --sw, --p, moodboards | Delete. Upload the source picture as a reference with the Style role — that is the equivalent, and it is a file rather than a code. |
--cref (legacy) or the V8 Edit Model's references | Delete. Upload the character picture with the Character role. |
--stylize, --chaos, --weird, --no | Delete. There is no equivalent. Say what you want in words instead of tuning a dial. |
| The comma-separated noun phrase | Expand it into sentences, in order: composition, light, colour, materials, then the medium. |
The last row is the one that actually changes the picture. Midjourney fills gaps confidently from a short phrase; Nano Banana Pro takes what you give it more literally, so a five-word prompt gets you a five-word picture.
A worked conversion
A Midjourney-shaped prompt: brass oil lamp, temple steps, dawn, marigolds, moody, cinematic --ar 3:2.
The same idea written for Nano Banana Pro, with the ratio moved to the setting:
A brass kuthuvilakku oil lamp standing on a wet granite step at dawn, one lit wick with a small steady flame, a loose marigold garland coiled beside its base, cool blue morning light coming from the left, fine beads of water on the brass, a worn temple wall out of focus behind, photographic, shallow depth of field, natural film grain, no text, no people.
Notice what the expansion added: a direction for the light, a material behaviour (beads of water), a depth cue (the wall out of focus), and two exclusions. "Moody, cinematic" was dropped entirely — mood words are the least useful part of a Midjourney prompt when it crosses over, because on this model they are replaced by the light description that actually causes the mood.
The shortcut if you have the picture
If you still have the Midjourney image, do not convert the text at all. Decode the picture for 3 credits: you get a generation prompt already written in the four-pass shape, its palette with hex codes and a layered style DNA. That is faster and more accurate than rewriting by hand, because it describes what the picture became rather than what you asked for.
Then attach the Midjourney picture itself as a style reference alongside it. We did exactly this with a Midjourney portrait: palette, medium and subject crossed intact; the line character and the colour of the glow did not. The full comparison is here.
What does not convert
- Parameter tuning. There is no stylize or chaos dial. The equivalent lever is more specific words.
- Personalization. Midjourney can learn your taste; nothing here does. Consistency comes from references attached each time.
- Style codes. A numeric style code is a Midjourney object and means nothing to another model. Upload the picture instead.
- Negative prompts.
--nohas no direct equivalent and negatives are weak here. Describe the scene you want; an empty street beats "no people".
FAQ
How do I convert a Midjourney prompt for Nano Banana?
Delete every double-dash parameter, move the aspect ratio to the generation setting, and expand the noun phrase into ordered sentences — composition, light, colour, materials, then the medium. Mood words can usually go.
Do Midjourney parameters work on Nano Banana Pro?
No. Aspect ratio becomes a setting, style and character references become uploaded files with a role, and stylize, chaos, weird and no have no equivalent at all.
Is there a faster way than rewriting by hand?
Yes, if you still have the image. Decode it for 3 credits and you get a prompt already written in the shape this model reads best, plus the palette and style DNA — and it describes what the picture became rather than what you asked for.
Why does my converted prompt give a flatter picture?
Because it is too short. Midjourney fills gaps confidently from a few words; Nano Banana Pro takes what you give it more literally, so add the light direction, the material behaviour and the depth cues the short phrase left out.
What replaces a style reference code?
Uploading the picture itself and badging it as a style reference. That role tells the model to take the palette, mood and rendering and ignore the subject and composition.
The quickest test is to decode one Midjourney picture in the Studio and compare its prompt against what you originally typed. The prompt shape itself is part 7 of the generation series.