Nano Banana vs. Midjourney v7: 20 Copy-Paste Prompts, Both Models (2026)

Asking "which AI image model is best?" is the wrong question. The right question is "which model is best for this specific task?" Nano Banana (Google's Gemini 2.5 image model) and Midjourney v7 have genuinely different strengths — not different points on a single quality scale. These 20 head-to-head prompts show exactly where each model earns its keep and where it falls short.

How These Comparisons Work

Each prompt appears in two versions. Nano Banana receives a natural-language description written the way its interface expects — conversational, reference-image-aware, detailed about preservation and change. Midjourney v7 receives a parameter-structured prompt using the syntax it's optimized for: --ar, --s, --sref, --cref, and stylistic vocabulary that MJ's training responds to. The same creative intent drives both versions, but the format differs because forcing one prompt format onto the wrong model produces worse outputs.

"Winner" in each section is a qualitative judgment about how well each model followed the intent of the prompt and how usable the output was on the first attempt. It is not a benchmark score, a pixel-level analysis, or a claim about which model is technically superior. Prompt-following behavior, output reliability, and format flexibility are the observable properties we're describing.

Tradeoffs always exist and they go beyond image quality. Nano Banana operates inside the Gemini ecosystem — it integrates with Google Workspace, supports multi-image inputs, and processes prompts conversationally. Midjourney v7 runs on Discord and a web interface, costs separately, and requires learning its parameter vocabulary. Speed, cost per image, and access patterns differ across tiers. Those operational factors matter and should factor into your tooling decisions.

The practical finding across all 20 prompts: most teams doing serious image work should use both models, routed by task type. Nano Banana handles the work that requires precision — edits with preservation constraints, character consistency, multi-image composition, and accurate text rendering. Midjourney v7 handles the work that requires mood — cinematic atmosphere, aesthetic depth, painterly artistic styles, and the kind of visual richness that makes a still image stop a scroll. Picking one and forcing it on every job is how you produce mediocre output at twice the cost. For prior context on how a similar comparison plays out, see our Midjourney vs DALL-E 3 breakdown.

We don't fabricate benchmark numbers. All observations describe prompt-following behavior based on running these prompts through each model. If you run the same prompts, your results will vary — that variability is itself informative.

Precise Photo Edits (Prompts 1–3)

Precise edits test whether a model can change one thing while preserving everything else. This is where the difference between a generation model and an edit-aware model becomes obvious.

1. Background Swap

[Nano Banana]
Take this product photo of a ceramic coffee mug on a white studio
background. Replace the white background with a cozy morning kitchen
scene — warm light, soft bokeh, wooden counter surface barely visible.
Keep the mug exactly as it is: same angle, same size in frame, same
lighting on the mug itself. Do not alter the mug's color, shape, or
any reflections on its surface.

[Midjourney v7]
ceramic coffee mug, cozy kitchen morning background, warm golden light,
wooden counter, soft bokeh, product photography, studio lighting on
subject --ar 4:5 --s 200 --style raw

Nano Banana: The edit-aware architecture handles this well. Nano Banana respects the "keep the mug exactly as it is" constraint, replacing the background without repainting the subject. The lighting match between the preserved mug and the new background is the variable — it performs well when the original and target backgrounds are reasonably compatible in light direction.

Midjourney v7: Midjourney v7 generates a fresh image matching the description rather than editing an existing photo. The result looks good aesthetically — warm, cinematic, intentional — but it's a new mug, not your mug. For product photography where the subject is proprietary or specific, this matters.

Winner: Nano Banana — it edits; Midjourney generates. When preservation of the original subject is the job, generation is the wrong tool.

2. Object Removal

[Nano Banana]
Remove the power cord visible in the lower left corner of this desk
photo. Fill the area naturally using the floor texture and color already
present in the image. Do not change anything else in the frame —
position of items on desk, lighting, and background should all remain
identical.

[Midjourney v7]
clean minimal desk setup, no visible cables or cords, natural light,
light wooden surface, tidy modern workspace --ar 16:9 --s 150
--style raw

Nano Banana: Object removal with fill is a direct test of instruction-following under constraints. Nano Banana attempts to fill the removed area using the surrounding texture, and the "change nothing else" constraint is respected in most cases. Complex fills (removing objects against highly textured or patterned backgrounds) require iteration.

Midjourney v7: Again generates a clean desk image that matches the description — aesthetically refined, nothing out of place. It cannot edit a specific photo's specific cord. If your goal is reusable stock imagery rather than fixing an actual photo, MJ's output is immediately usable.

Winner: Nano Banana — object removal from an existing image requires editing capability, not generation.

3. Color Change on a Specific Element

[Nano Banana]
In this photo of a woman wearing a red wool coat, change the coat color
to deep forest green. Keep her skin tone, hair, background, and all
other clothing exactly the same. The coat's texture, folds, and shadow
detail should remain intact — only the hue changes.

[Midjourney v7]
woman in deep forest green wool coat, professional street portrait,
autumn city background, photorealistic, natural light --ar 2:3 --s 100
--style raw

Nano Banana: The instruction to change only the coat hue while preserving texture, folds, and all surrounding elements is exactly the kind of constrained edit Nano Banana is built for. The forest green coat emerges with the original shadow and fold structure visible underneath — it reads as a recolor rather than a repaint.

Midjourney v7: Generates a polished portrait with a forest green coat. The composition and lighting are notably strong. But the woman's face, background, and secondary clothing items are all new — there's no continuity with the original photo.

Winner: Nano Banana — isolated element recoloring with texture preservation is an editing task, not a generation task.

Character & Subject Consistency (Prompts 4–5)

Consistency across frames is a production need, not a nice-to-have. Comics, storyboards, product catalogs, and branded social content all require the same subject to look like the same subject.

4. Same Character, New Pose

[Nano Banana]
Using this reference image of a young man with short curly brown hair,
round wire-frame glasses, and a navy blue hoodie: generate a new image
of the same person sitting at a cafe table with a laptop, looking
focused. Keep his facial features, hair style, glasses, and general
build consistent with the reference. Background: warm cafe interior,
shallow depth of field.

[Midjourney v7]
young man, short curly brown hair, round wire-frame glasses, navy
hoodie, sitting at cafe with laptop, focused expression, warm cafe
bokeh background, natural light --cref [reference_url] --cw 80
--ar 4:5 --s 150

Nano Banana: Multi-image input is a native capability here — passing a reference image alongside the generation prompt is how the model is designed to work. Character attributes specified in the text (hair, glasses, hoodie) are applied to the reference, and the output maintains more continuity with the source than a pure text-to-image approach.

Midjourney v7: The --cref character reference parameter was introduced specifically for this problem, and it works. With a good reference URL and --cw 80 (character weight), Midjourney v7 maintains facial features across the new pose with solid reliability. The aesthetic quality of the output — lighting, composition, the cafe environment — is typically stronger.

Winner: Tie — both have dedicated mechanisms for this. Nano Banana's native multi-image input gives it a slight edge when the reference and text instruction conflict; MJ's --cref output has a higher aesthetic ceiling.

5. Same Product, New Angle

[Nano Banana]
Using this product photo of a matte black wireless speaker as reference,
generate a new product shot showing the speaker from a low 3/4 angle
looking up, on a dark slate surface, with dramatic side lighting that
emphasizes the speaker's grille texture. Keep the speaker design,
color, and proportions identical to the reference. No background
clutter — just the product and surface.

[Midjourney v7]
matte black wireless speaker, low 3/4 angle upward perspective, dark
slate surface, dramatic single-source side lighting, grille texture
prominent, clean product photography, no props --ar 3:4 --s 200
--sref [reference_url] --sw 60

Nano Banana: Product photography consistency is a strong use case. The reference image anchors the specific design — button placement, grille pattern, logo position — so the new angle generation doesn't invent design elements that weren't there. Lighting match between reference and output requires attention in the prompt.

Midjourney v7: Using --sref with moderate style weight maintains visual language from the reference while allowing the composition to change. The resulting product shot often has a polished, editorial quality. Design detail accuracy — specific button placement, exact grille pattern — is less reliable than Nano Banana when the product is complex.

Winner: Nano Banana — product detail consistency across angles is more reliable when you can pass the reference directly to an edit-aware model.

Multi-Image Composition (Prompts 6–7)

Combining elements from multiple source images is a workflow, not a feature. Both models handle it differently.

6. Subject into a New Environment

[Nano Banana]
Composite image: place the person from Image A (a woman in professional
attire, standing confidently) into the environment from Image B (a
modern glass-walled conference room, city view visible through windows).
Make the lighting on the person consistent with the conference room's
natural light direction. She should appear naturally present in the
space — not pasted. Maintain her appearance from Image A exactly.

[Midjourney v7]
confident businesswoman in professional attire standing in modern
glass-walled conference room, city skyline view through windows, natural
light, photorealistic, corporate portrait --ar 3:4 --s 100 --style raw

Nano Banana: Multi-image composition — passing two distinct images and instructing the model to combine specific elements from each — is a native workflow. The output quality depends heavily on lighting compatibility between the source images, but the instruction-following on "take this person from Image A, place them in Image B" is more literal than any text-only prompt can achieve.

Midjourney v7: Without a direct multi-image composition feature that references two uploaded photos simultaneously in this way, Midjourney generates a fresh image matching the text description. The result is a high-quality professional portrait, but it's not the specific person from your original image.

Winner: Nano Banana — combining specific elements from specific photos requires a model that accepts and respects multi-image inputs.

7. Blending Two Products

[Nano Banana]
Create a lifestyle flat-lay composition that includes both products:
Product A (a cream-colored candle tin from Image A) and Product B
(a dried lavender bundle from Image B). Arrange them on a white linen
surface with soft diffused light. Both products should appear exactly
as they do in their source images — same labels, same colors, same
proportions. Style: minimal, clean, premium lifestyle product photography.

[Midjourney v7]
flat lay of cream candle tin and dried lavender bundle on white linen,
soft diffused light, minimal premium product photography, top-down view,
clean negative space --ar 1:1 --s 200

Nano Banana: This is the direct test of multi-image fidelity under a composition brief. Both products need to appear as they actually appear — specific label, specific container shape — not as Nano Banana imagines they might look. Passing both source images gives the model the actual product references it needs.

Midjourney v7: Generates a beautiful flat-lay that looks exactly like premium lifestyle photography. The candle tin and lavender bundle will be beautifully lit and composed. They won't be your specific products. For original brand assets where visual identity is exact, this gap is not negotiable.

Winner: Nano Banana — brand-accurate multi-product compositions require the actual product images as inputs.

Text-in-Image (Prompts 8–9)

Text rendering is a historically weak area for AI image models. In 2026, the gap between models has narrowed but not closed.

8. Branded Poster with Headline

[Nano Banana]
Design a vertical event poster. The headline text at the top should
read exactly: "SOUND & VISION FESTIVAL". Below that, in smaller type:
"July 18–20 · Portland, OR". Background: dark navy with abstract
light-streak photography. Font style: bold geometric sans-serif for
the headline, clean regular weight for the date. All text must be
correctly spelled and legible. Maintain high contrast between text
and background.

[Midjourney v7]
event poster, dark navy background, abstract light streaks, bold
geometric sans-serif headline, festival typography design, high contrast,
professional event design --ar 2:3 --s 150 --style raw

Nano Banana: Accurate text rendering is a defined strength. Specifying the exact strings — "SOUND & VISION FESTIVAL" and "July 18–20 · Portland, OR" — and requesting correct spelling produces the specified text reliably. Character count, apostrophes, and special characters like em-dashes are handled with considerably more accuracy than older image models.

Midjourney v7: Midjourney v7 has improved text rendering over v6, but exact multi-word headlines are still unreliable on the first attempt. The poster's aesthetic — the light streaks, the dark atmosphere, the typography's visual weight — will look stronger. The text will likely require regeneration to get spelling right, and complex strings with punctuation still trip it up.

Winner: Nano Banana — accurate multi-word text with punctuation in a designed layout is more reliable here, consistently.

9. Packaging Label

[Nano Banana]
Create a product label design for a coffee bag. The label must include
this text exactly as written:
- Brand name: "ALTO ROAST"
- Subtitle: "Single Origin Ethiopia"
- Roast level: "Medium Dark"
- Net weight: "12 oz / 340g"
Typography style: artisanal, clean, slightly premium. Background color:
warm kraft paper texture. Layout: centered, stacked vertically. All
text correctly spelled.

[Midjourney v7]
artisan coffee bag label design, warm kraft paper texture, premium
typography, centered vertical layout, clean serif and sans-serif type
pairing, coffee brand packaging --ar 2:3 --s 250

Nano Banana: Packaging labels are a text-dense design challenge. Specifying four distinct text strings with exact content — brand name, subtitle, roast level, and weight — and receiving them rendered correctly on the first attempt is where Nano Banana's text accuracy provides real workflow value. Each label field appears in the right visual hierarchy.

Midjourney v7: The aesthetic quality of the packaging — the kraft texture, the typographic feel, the premium visual language — is often better. The specific text strings will be wrong, approximate, or partially correct. Using Midjourney for initial packaging direction and layout exploration, then finalizing in a design tool, is a practical workflow.

Winner: Nano Banana — any task requiring multiple exact text strings in a designed output needs a model that can render them accurately.

Photorealistic Generation (Prompts 10–12)

Pure generation without a reference image — the traditional strength test for AI image models.

10. Lifestyle Scene

[Nano Banana]
Generate a photorealistic lifestyle image: a woman in her late 30s
sits cross-legged on a light grey linen sofa, reading a hardcover book.
She has loose natural hair and is wearing an oversized cream sweater.
Morning light comes through large windows to camera right, casting warm
shadows across the sofa. The room feels minimal and lived-in.
Full frame, slightly elevated camera angle.

[Midjourney v7]
woman late 30s, natural hair, oversized cream sweater, cross-legged
on light grey linen sofa, reading hardcover book, morning window light
camera right, warm shadows, minimal lived-in living room, elevated
angle, photorealistic --ar 4:5 --s 150 --style raw

Nano Banana: Natural language lifestyle briefs translate reliably into well-composed images. The model follows the specific staging instructions — camera angle, light direction, sofa color — with good accuracy. Skin tone rendering and natural hair texture have been consistent strengths.

Midjourney v7: The compositional quality and lighting artistry here tend to be stronger. The "minimal lived-in" instruction is where Midjourney excels — it adds the kind of subtle background detail and tonal warmth that makes a lifestyle image feel like an editorial shoot rather than a render.

Winner: Midjourney v7 — for lifestyle imagery where there's no preservation constraint and aesthetic quality is the primary goal, MJ's compositional judgment produces more compelling results.

11. Food Photography

[Nano Banana]
Generate a food photography image: a whole roasted chicken on a dark
cast iron skillet, resting on a worn wooden kitchen table. The chicken
skin is deeply golden-brown and glistening. Garnish: fresh rosemary
sprigs and lemon halves around the skillet. Lighting: dramatic
single-source overhead spot with shadows falling to camera left.
High-resolution, commercially usable food photography style.

[Midjourney v7]
whole roasted chicken golden brown skin, dark cast iron skillet, worn
wooden table, fresh rosemary garnish, lemon halves, dramatic overhead
single-source lighting, deep shadows, commercial food photography
--ar 4:5 --s 200 --style raw

Nano Banana: Follows the scene description accurately — cast iron skillet, garnish placement, light direction. The output is usable, clean, and meets the brief. Skin texture and glossiness are handled well.

Midjourney v7: Food photography is a category where Midjourney's aesthetic processing visibly shows. The skin texture has depth. The shadows have gradation. The cast iron picks up subtle reflected light. The overall image has the quality of a styled food editorial, not just a technically correct render.

Winner: Midjourney v7 — when the brief is open (no reference, no exact elements to preserve), MJ produces more visually appetizing food photography.

12. Portrait

[Nano Banana]
Generate a portrait photograph: a man in his late 50s with silver hair
swept back, a close-cropped grey beard, and weathered but warm features.
He's wearing a dark olive linen shirt, collar open. Background: blurred
urban street, late afternoon. Shot on a short telephoto lens with
shallow depth of field. Expression: calm, direct gaze into camera.
Photorealistic, editorial portrait quality.

[Midjourney v7]
man late 50s, silver swept-back hair, close-cropped grey beard, weathered
warm face, dark olive linen shirt open collar, urban street background
bokeh, late afternoon light, short telephoto shallow DOF, direct gaze,
editorial portrait --ar 2:3 --s 150

Nano Banana: The portrait description is followed with good fidelity — age indicators (silver hair, grey beard, weathered features), clothing, and technical camera parameters all appear in the output. The expression instruction ("calm, direct gaze") is interpreted literally.

Midjourney v7: Portrait photography is a MJ stronghold. The skin rendering — fine texture, the slight roughness of weathered skin, the way late afternoon light catches the face — is where Midjourney's output tends to pull ahead. The result looks like a portrait a photographer actually took, not a model's interpretation of one.

Winner: Midjourney v7 — portrait realism and the tactile quality of skin, texture, and light lean toward MJ when there's no reference image to preserve.

Cinematic & Mood Stills (Prompts 13–15)

Atmosphere, tension, and visual poetry. This is Midjourney's home territory.

13. Neo-Noir City

[Nano Banana]
Generate a cinematic still: a lone figure in a long dark coat stands
at the end of a rain-soaked alley, facing away from camera. Neon signs
in Korean and English reflect in standing puddles. The light is deep
blue and magenta. Shallow depth of field — figure sharp, street receding
into blur. Mood: urban isolation, late night, classic noir aesthetic.
Photorealistic, wide-aspect ratio.

[Midjourney v7]
lone figure dark coat, rain-soaked alley, neon signs Korean English,
blue magenta reflections in puddles, deep noir atmosphere, late night
urban isolation, cinematic still, shallow depth of field, figure sharp
background blur --ar 21:9 --s 750 --style raw

Nano Banana: The scene elements are present and correctly placed — figure, alley, neon signs, puddle reflections. The color brief (blue and magenta) is followed. The output is technically solid and usable.

Midjourney v7: Cinematic mood is where the stylization score (--s 750) earns its keep. The light behaves like light in a noir film — there's scatter, haze, directional drama. The puddle reflections have depth rather than flatness. The figure's silhouette is composed, not just placed. This is a meaningful visual difference for anyone producing content where atmosphere is the product.

Winner: Midjourney v7 — neo-noir atmosphere requires the kind of aesthetic intelligence that Midjourney's training gives it over a text-to-image model optimized for instruction-following.

14. Period-Drama Interior

[Nano Banana]
Generate a cinematic still from a period drama set in 1920s England:
a wood-paneled study with floor-to-ceiling bookshelves, a leather
Chesterfield sofa, and a single lamp casting warm amber light in an
otherwise dark room. Thick shadows. A glass of whiskey sits on a small
side table. No people. Shot from a low angle, wide. The feel: heavy,
secretive, money and dread.

[Midjourney v7]
1920s English study interior, wood paneling, floor-to-ceiling bookshelves,
Chesterfield sofa, single amber lamp, thick shadows, whiskey glass side
table, no people, low wide angle, cinematic, period drama, heavy
atmospheric mood --ar 21:9 --s 600

Nano Banana: The room description is followed faithfully — bookshelves, Chesterfield, lamp, whiskey glass, no people. The color temperature (amber in darkness) and camera angle (low, wide) are applied correctly. The output covers the brief.

Midjourney v7: The specific prompt instruction "money and dread" isn't directly translatable, but Midjourney's period-drama vocabulary translates the mood without being told how. The shadow quality — the way it pools and edges around furniture — is what separates this output from a technically correct render. It feels like a production still.

Winner: Midjourney v7 — period drama atmosphere with the kind of tonal weight this brief calls for is exactly what high stylization scores produce.

15. Atmospheric Horror

[Nano Banana]
Generate a horror-adjacent atmospheric still: an empty Victorian
children's bedroom at night, low moonlight through tattered curtains,
a wooden rocking horse centered in frame facing camera. Dust motes
visible in the moonlight. Deep shadows everywhere. The mood should be
unsettling through emptiness and implication — no monsters, no blood,
just wrongness. Desaturated palette, slight cold color cast.

[Midjourney v7]
empty Victorian children's bedroom, night, moonlight through tattered
curtains, wooden rocking horse centered facing camera, dust motes in
light, deep shadows, desaturated cold palette, atmospheric dread,
wrongness through stillness and emptiness --ar 16:9 --s 800

Nano Banana: The compositional instructions are followed — rocking horse centered, tattered curtains, moonlight, dust motes. The desaturated palette with cold cast is applied. The image is eerie and usable for most production contexts.

Midjourney v7: Horror atmosphere built through composition and light rather than explicit monsters is precisely the brief MJ responds to strongly. The high stylization score allows it to lean into the tonal language of horror filmmaking — the particular quality of moonlight in this context, the way shadows fall around the rocking horse, the texture of the tattered curtains. The output tends to feel composed rather than generated.

Winner: Midjourney v7 — atmosphere-as-subject with implicit dread is one of MJ's clearest wins.

Illustration & Artistic Style (Prompts 16–17)

Painterly traditions and editorial illustration test both models' understanding of art history and craft technique.

16. Editorial Illustration

[Nano Banana]
Create an editorial illustration in a flat vector-adjacent style for
an article about urban loneliness: a single figure sits on a bench
in a large empty plaza, surrounded by towering glass office buildings.
Warm muted palette — dusty rose, slate grey, pale yellow. Clean lines,
minimal texture, limited color palette. Style should feel contemporary
editorial — the kind used in publications like The Atlantic or Wired.
No text in the image.

[Midjourney v7]
editorial illustration, lone figure on bench, vast empty plaza, glass
office towers, urban loneliness concept, flat contemporary illustration
style, dusty rose slate grey pale yellow palette, clean lines, minimal
texture, magazine editorial quality --ar 4:3 --s 300

Nano Banana: The style brief — flat, vector-adjacent, named-publication-level quality — is interpreted with reasonable accuracy. Palette constraints (dusty rose, slate grey, pale yellow) are followed. The conceptual framing (urban loneliness, scale contrast between figure and buildings) is captured in the composition.

Midjourney v7: Editorial illustration is a MJ strength. The model understands the visual language of contemporary magazine illustration — the specific way flatness is deployed, the tonal range within a limited palette, the compositional grammar that makes the image read quickly as an editorial piece. The output is typically more confident stylistically.

Winner: Midjourney v7 — the aesthetic intelligence built into MJ's training makes editorial illustration feel authorial rather than assembled.

17. Watercolor Scene

[Nano Banana]
Generate a watercolor painting of a morning market in a small coastal
town — fishing boats visible in the background harbor, market stalls
selling produce in the foreground, a few people browsing. Painting
style: loose, gestural watercolor with visible paper texture, wet-on-wet
blooms in the sky, slightly muted coastal palette (sea green, warm
ochre, soft blue-grey). The technique should feel like the painting
was done on location, not digitally generated.

[Midjourney v7]
coastal morning market, fishing harbor background, produce stalls,
loose gestural watercolor, visible paper texture, wet-on-wet sky blooms,
sea green ochre blue-grey palette, plein air quality, painterly,
expressive brushwork --ar 4:3 --s 500

Nano Banana: The watercolor brief is followed in terms of subject matter and stated technique. Wet-on-wet blooms and paper texture appear in the output. The scene is accurate to the description. The painterly quality has improved meaningfully in recent Gemini image model iterations.

Midjourney v7: Painting technique at the medium level — not just "watercolor style" as a filter but the specific properties of wet paper behavior, granulation, and edge quality — is where MJ's training depth shows. The difference between a watercolor effect applied to a digital image and an image that reads like a painting lies in the model's understanding of how the medium actually behaves. MJ's outputs in this category are closer to the latter.

Winner: Midjourney v7 — painterly traditions understood at the medium level, not just as visual filters, is where MJ has a genuine depth advantage.

Logo & Brand Mark (Prompts 18–20)

Brand identity work requires both aesthetic judgment and reliable text output — a split that maps directly onto each model's strengths.

18. Wordmark

[Nano Banana]
Design a wordmark logo for a brand named "VERDANT". The name should
be rendered in a clean, modern sans-serif with slightly custom letterforms
— specifically, the crossbar of the A should be dropped and the terminals
should be geometric. Color: deep forest green (#2D5016 equivalent).
White background. The word "VERDANT" must be spelled correctly and
appear as the only element in the design. Minimal, premium brand feel.

[Midjourney v7]
VERDANT wordmark logo, modern geometric sans-serif, forest green,
clean minimal premium brand identity, custom letterforms, white
background, logotype design --ar 3:1 --s 200

Nano Banana: Wordmarks require exact text rendering above all else. "VERDANT" spelled correctly, in the right weight, with the specific letterform details described, is achievable with Nano Banana's text accuracy. The model applies the typographic customizations described (dropped crossbar, geometric terminals) with variable success — complex glyph customizations sometimes require iteration.

Midjourney v7: Wordmarks are where MJ's text limitations directly constrain output quality. The visual aesthetic — the weight, the feel of the letterforms, the color application — is often excellent. "VERDANT" is the problem: it may render as "VERDANT", "VERDENT", "UERDANT", or other approximations. Getting exact spelling across multiple letters reliably requires regeneration.

Winner: Nano Banana — wordmark logos must spell the brand name correctly, and that basic requirement is met more reliably here.

19. Abstract Brand Mark

[Nano Banana]
Create an abstract geometric brand mark (no text) for a fintech brand:
an interlocking set of three shapes that suggest both connectivity and
forward motion. Colors: deep navy and electric blue. The mark should
work at small sizes — clean, uncluttered lines. No gradients. The
overall feel should communicate trust and technology without cliché
(no circuits, no dollar signs, no obvious arrows).

[Midjourney v7]
abstract geometric brand mark, three interlocking shapes, connectivity
forward motion concept, deep navy electric blue, fintech brand identity,
works at small size, no gradients, clean minimal logomark, trust and
technology --ar 1:1 --s 300

Nano Banana: Abstract logomarks without text play to the layout and composition strengths of both models. Nano Banana follows the constraint list (no gradients, no clichés) reasonably well. The three-shape interlocking instruction is followed with good accuracy.

Midjourney v7: Abstract brand marks with a mood brief ("trust and technology") and aesthetic constraints are well inside MJ's competency. Without the text accuracy burden, the stylization engine produces marks with the kind of intentional visual balance that a designer would have constructed. The electric blue and deep navy palette is handled with visual sophistication.

Winner: Midjourney v7 — abstract logomarks are pure composition and aesthetic judgment, no text requirements, which is MJ's clean-lane advantage.

20. Full Identity Suite

[Nano Banana]
Create a brand identity exploration for a brand called "PLUMA" — a
premium journaling and stationery brand. Generate a flat-lay composition
showing: a notebook with "PLUMA" on the cover, a pen, and a small wax
seal stamp. The brand name must be spelled correctly on the notebook.
Palette: warm off-white, terracotta, and deep charcoal. Clean, minimal,
premium lifestyle feel. The layout should suggest a cohesive brand world.

[Midjourney v7]
PLUMA premium stationery brand identity flat lay, notebook pen wax seal,
warm off-white terracotta charcoal palette, minimal premium lifestyle,
coherent brand world, editorial product photography --ar 4:5 --s 400

Nano Banana: Brand suite compositions that include text on physical objects (notebook cover) sit squarely in Nano Banana's combined strengths: multi-element composition, exact text rendering, and layout following. "PLUMA" appearing correctly spelled on the notebook cover is the functional requirement that makes this execution usable.

Midjourney v7: The aesthetic quality of the lifestyle flat-lay — how the terracotta, off-white, and charcoal interact, the light quality, the way the objects are composed — is typically stronger. The notebook cover will say something close to "PLUMA" but may not be exact. For early-stage mood and identity exploration before a designer finalizes the work, this output is valuable.

Winner: Nano Banana — when text on objects (labels, product names, cover text) must be legible and correctly spelled, the text accuracy advantage decides the outcome.

Model Selection Rubric

Stop choosing a model. Start routing by task.

What to Use Right Now

Both models are worth having in your workflow. The question is which one you reach for first on a given job.

Build better prompts for either model with the AI prompt generator — it structures your intent into the format each model responds to. For Nano Banana-specific prompts and use cases, the best Nano Banana prompts guide covers 30 more templates across editing, generation, and composition workflows. For Midjourney v7 parameter usage and style prompts, start with the best Midjourney v7 prompts guide. And for a complete framework on how to write image prompts that work across models, the AI image prompting complete guide covers the underlying principles that translate across any tool you use.

Route by task. Both models are tools, not allegiances.