Midjourney and Nano Banana Prompts: Same Idea, Two Workflows

Use this page when you already know the picture you want and need to express it in Midjourney or Google Nano Banana without copying the wrong control language into the wrong product.

It covers versions and surfaces, parameters and reference roles, parallel prompt shapes for the same brief, editing patterns, troubleshooting, and a pre-flight checklist. It does not declare which product is better for realism, text, consistency, or creativity.

Universal visual language lives in the AI Image Prompt Guide. Gemini product-map depth beyond this cheat sheet lives in How to prompt Google Gemini.

Write one sentence that locks subject, action, setting, and mood before you touch product syntax. Then adapt that brief twice: once for Midjourney, once for Nano Banana.

This guide uses VISOR: Vision, Interface, Specs outside the prose, Objects & refs, and Refine. VISOR is a practical checklist created for GetBetterPrompts for this guide. It is not an industry standard.

• V (Vision): one sentence: subject + action + setting + mood.
• I (Interface): name the surface (Midjourney web or Discord; Gemini app, AI Studio, API, or Vertex).
• S (Specs outside the prose): aspect, size or HD, model version, stylize or raw, API image_size. Set these in controls or trailing parameters, not as fake flags inside the wrong product.
• O (Objects & refs): decide image prompt vs style vs omni or character refs; label each ref's job; respect model caps.
• R (Refine): change one variable per iteration; write preserve vs change lists for edits.

Hard rule: Midjourney parameters are not Nano Banana prompt text. Nano Banana size and model controls are not Midjourney parameters.

As of July 2026, Midjourney's documented default model is V8.1 (default since June 10, 2026; released April 30, 2026). Pin --v 8.1 when a project needs a stable version. Older versions remain available for different aesthetics.

Web and Discord both work; settings sync. Web offers Imagine-bar slots, Editor, and Style Creator. Discord uses /imagine with URL parameters. Official Prompt Basics prefer short, clear phrases and describe what you want along axes such as subject, medium, environment, lighting, color, mood, and composition. Put parameters at the end.

Midjourney has described SD jobs as appearing in about four seconds in a product update; treat that as an ops description, not a latency SLA. Prefer iterating in SD and finalizing in HD when using V8.1 (editorial recommendation, not a platform rule).

Daily V8.1 controls (ranges from Midjourney docs; re-check Version and Parameter List on ship day):

• --ar: aspect ratio (default 1:1). V8.1 HD max aspect is tighter than SD (docs: up to 4:1 HD).
• --hd / --sd: resolution modes; HD is roughly 2048px-class at 1:1; SD is roughly 1024px-class.
• --stylize / --s: 0–1000, default 100.
• --chaos / --c: 0–100, default 0.
• --raw: reduces default auto-styling for stricter control.
• --seed: reproducibility aid; not a character or style lock.
• --no: exclusions; prefer describing what you want first.
• --q: quality is documented clearly for V7 (1, 2, 4); V8.1 quality matrix may be incomplete in Version docs, so verify before relying on it.

Image Prompt: inspiration, not an exact copy. Web slot or Discord URL at the start; --iw on V8.1/V7 defaults to 1 (range 0–3).

Style Reference (--sref): look and feel (color, medium, texture, lighting), not objects or people. --sw 0–1000, default 100. Avoid conflicting style adjectives when a style ref is attached.

Omni Reference (--oref): V7 only: Omni Reference is compatible with V7 only. Adding Omni automatically runs the prompt in V7 even from the V8.1 UI. One Omni image; --ow 1–1000, default 100. Intricate details may not match. Do not treat --oref as a normal daily V8.1 parameter. Legacy --cref applies to Midjourney/Niji v6; on V7 use Omni.

Nano Banana is a family of Gemini image models, not one singular model. As of the Gemini API docs checked for this guide:

• Nano Banana 2 Lite: gemini-3.1-flash-lite-image: fastest/cheapest for high volume; 1K only; not optimized for multi-reference or multi-turn sequential editing.
• Nano Banana 2 (Flash): gemini-3.1-flash-image: default workhorse; refs up to 10 objects + 4 characters; sizes 0.5K–4K.
• Nano Banana Pro: gemini-3-pro-image: complex / premium control; refs up to 6 objects + 5 characters + 3 style; sizes 1K–4K.
• Nano Banana (legacy): gemini-2.5-flash-image: docs recommend migrating; works best with about 3 input images.

Do not apply one family-wide reference ceiling or one family-wide "1K to 4K" range without splitting by model. Marketing ceilings and surface UIs may expose fewer inputs than the API role table.

Surfaces differ: Gemini app (conversational; routing may be automatic), Google AI Studio (try models), Gemini API (exact IDs, image_size, multi-turn), and Vertex / Gemini Enterprise (enterprise controls; 4K may remain Preview). See also How to prompt Google Gemini for the wider product map.

Google's prompting tips structure scenes around subject, composition, action, location, and style, plus camera and lighting when useful. Quote exact on-image text. Define each reference image's role. Iterate with specific edit instructions. Generated images may include SynthID watermarks per Google docs.

Set aspect and size in the UI or API (response_format / image_size), not by pasting Midjourney flags into the prose. Label refs: "Image 1 = product geometry; Image 2 = style only." Prefer Flash Image as the general API workhorse per Google's current model selection guidance; use Lite for efficiency; Pro for complex production; migrate off legacy 2.5.

Multi-turn editing works best when you say what changes and what must stay. Prefer Flash or Pro for sequential edits; Lite is not optimized for that pattern. Limitations remain: text, factual accuracy, and complex edits can fail. Verify lettering and details visually. Do not promise clean typography or perfect identity lock.

Shared brief for the examples below: a matte black ceramic pour-over dripper on a walnut counter, morning side light, shallow depth of field, quiet product-photo mood. Optional label text: ORIGIN BLEND. These are examples, not proven winners, and not quality rankings.

A. Text-to-image

Midjourney (V8.1):

matte black ceramic pour-over dripper on walnut counter, morning side light, shallow depth of field, quiet product photography --ar 4:5 --raw --stylize 100 --v 8.1

Nano Banana (Flash Image prose):

Photorealistic product photo of a matte black ceramic pour-over dripper on a walnut counter. Soft morning light from the left, gentle shadow to the right, shallow depth of field, quiet commercial still life. Square-ish portrait crop feel. No props other than the dripper. No watermark.

Set aspect / 1K–2K in the UI or API, not as Midjourney flags. Midjourney prefers short phrases + trailing parameters; Nano Banana prefers fuller natural-language constraints.

B. Image references

Midjourney: Image Prompt slot (web) or image URL(s) at the start (Discord) + short text of what must remain; optional --iw 1.5 within 0–3.

Nano Banana: Use Image 1 only for the dripper's shape and material. Do not copy the background from Image 1. Place it on a walnut counter as described.

C. Style references

Midjourney: Style Reference slot / --sref <URL> + content-focused text; optional --sw 100.

Nano Banana: Prefer a style-reference image on Pro (style-ref role exists in the API table) or describe style in prose on Flash/Lite: Match the muted film still palette and soft grain of Image 2. Keep the product geometry from the text brief.

D. Character or object continuity

Midjourney: Omni --oref <URL> --ow 100 with a clear text scene. This job will use V7 while Omni Reference is V7-only.

Nano Banana: Upload within the model’s role caps; Keep the same ceramic dripper identity as Image 1 (object). New scene: … Neither tool guarantees perfect logos or intricate details.

E. Editing

Midjourney: Remix or web Editor; change only the lighting phrase; or Vary Subtle before large rewrites.

Nano Banana: Keep the dripper, counter, and camera angle. Change only the lighting to overcast window light. Do not add new objects.

F. Typography

Midjourney: … small label reading "ORIGIN BLEND" in clean sans-serif … --raw --v 8.1 (verify; not perfect).

Nano Banana: Add a small label on the dripper that reads exactly "ORIGIN BLEND" in bold white sans-serif. No other text. Then verify glyphs.

G. Multi-turn refinement

Midjourney: Generate → pick → Remix one change → optional HD finalize.

Nano Banana: Turn 1 generate → Turn 2 Darken the background only via chat follow-up or API previous_interaction_id. Prefer Flash/Pro for sequential edits.

Midjourney editing is grid and Editor oriented: Vary, Remix, Pan/Zoom/Vary Region, Upscale. HD edits can downscale to SD in some flows; check Modifying and Remix docs. Change one phrase at a time when possible.

Nano Banana editing is conversational: preserve/change lists, one change per turn (guide recommendation). Prefer Flash or Pro for multi-turn; Lite is not optimized for sequential editing. Re-supply references when identity drifts.

Neither workflow guarantees zero drift. Compare each render to the VISOR vision sentence and stop when success criteria are met.

• Midjourney ignores half the prompt: shorten; try --raw; move specs to parameters.
• Grid looks the same: raise --c for exploration, then lock direction.
• Character drift on Midjourney: do not use --sref for identity; use Omni knowingly on V7; reinforce traits in text.
• Job suddenly on V7: Omni Reference attached; expected. Remove Omni to stay on V8.1.
• Gemini changes too much: preserve/change lists; one change per turn.
• Gemini drops a reference: over cap, unclear roles, or wrong model; label Image A/B; switch Flash/Pro.
• Soft or wrong resolution: Midjourney --hd/--sd; Gemini UI or image_size (Lite = 1K only).
• Text illegible: fewer words; higher contrast; Raw / simpler style; regenerate and verify.
• Same seed did not match: seeds are not identity locks.

1. One-sentence vision written (VISOR V).
2. Product + surface named (Midjourney web/Discord or Gemini app/Studio/API/Vertex).
3. Model/version pinned if the project needs it (--v 8.1 or gemini-3.1-flash-image, etc.).
4. Aspect ratio / size set in the correct control channel.
5. Reference images labeled by role; counts within the specific model's caps.
6. If using Midjourney Omni, accept V7 run + Omni limitations.
7. On-image text written exactly, with a plan to visually verify.
8. Exclusions short (--no or a few semantic negatives), not a 50-item dump.
9. First pass aimed at exploration (SD / 1K), not final hero size.
10. Success criteria stated (what must be true to stop iterating).

Tighten a brief in the free image prompt tool. Rebuild portable craft with the AI Image Prompt Guide. For Gemini surfaces beyond Nano Banana image controls, see How to prompt Google Gemini.

Keep the vision fixed. Swap only the product syntax.