Skip to content
· 5 min read

AI Image Gen Showdown: Nano Banana 2 vs GPT Image 1.5 vs Midjourney V7

ai image-gen workflow tools

Six months of production use

I have been using Nano Banana 2, GPT Image 1.5, and Midjourney V7 across multiple projects since late 2025. Not for fun (though it is fun). For production work: client presentations, product mockups, campaign assets for Mostly Mortals, and creative direction for agency projects. After hundreds of generations, I have a clear picture of where each tool excels, where it falls short, and how they fit together in a real pipeline.

This is not a benchmark post. You can find those anywhere. This is about what matters when you are shipping creative work on a deadline.

Nano Banana 2: The precision tool

Google’s Nano Banana 2 does two things better than anything else on the market: text rendering and photorealism. If your output needs legible text in an image, NB2 is the only reliable choice. Midjourney still mangles text. GPT Image 1.5 gets it right about 70% of the time. NB2 nails it consistently.

The photorealism is genuinely impressive. Product shots, architectural renders, and lifestyle photography come out with the kind of lighting and material detail that used to require a professional photographer and a post-production team. The downside is creative range. NB2 produces technically perfect images that can feel clinical. When you want mood, atmosphere, or artistic interpretation, it is not the first tool I reach for.

Best for: product mockups, anything with text overlay, architectural visualization, photorealistic portraits.

GPT Image 1.5: The conversation partner

GPT Image 1.5 changed how I think about image generation. The conversational editing model means you can iterate on an image the way you would with a designer. “Make the background warmer. Move the subject left. Change the lighting to golden hour.” Each instruction builds on the previous result, and the model maintains context across the conversation.

This iterative workflow is significantly faster than re-prompting from scratch in other tools. For exploratory creative work, where you know the direction but not the destination, GPT Image 1.5 gets you there in fewer attempts.

// Iterative generation with GPT Image 1.5
const conversation = await openai.images.edit({
  model: "gpt-image-1.5",
  image: previousResult,
  prompt: "Shift the lighting to late afternoon golden hour. Keep the subject position and expression identical.",
  size: "1536x1024",
  quality: "high",
});

// Each edit maintains context from previous generations
// No need to re-describe the entire scene

The weakness is consistency. When generating multiple images for a series (like scene illustrations for a campaign), maintaining a consistent style across generations requires careful prompt engineering. Midjourney handles this better with style references.

Best for: iterative creative exploration, client revision rounds, one-off hero images, concept art with specific feedback.

Midjourney V7: The aesthetic leader

Midjourney V7 remains the tool I reach for when the image needs to feel like something. The aesthetic quality, particularly for photorealistic lifestyle photography and cinematic compositions, is unmatched. The Omni-Reference system lets you feed in style references, character references, and composition references simultaneously, which makes consistency across a project dramatically easier.

For the Mostly Mortals campaign, Midjourney generates the scene illustrations. The fantasy art quality is a tier above what the other tools produce, and the style reference system means every image feels like it belongs to the same world.

// Midjourney V7 prompt with Omni-Reference
/imagine a dimly lit tavern interior, adventurers gathered around
a wooden table covered in maps, candlelight casting warm shadows,
fantasy illustration style --sref [campaign-style-url]
--cref [character-sheet-urls] --ar 16:9 --v 7

The downsides: text rendering is still unreliable, the Discord-based workflow is clunky for high-volume production, and the lack of an API means automation requires workarounds. For batch processing or programmatic generation, it is not a practical choice.

Best for: hero images, lifestyle photography, cinematic scenes, fantasy art, brand photography, anything where mood matters more than precision.

The production pipeline

In practice, I rarely use just one tool. The real workflow chains them together based on what each stage needs.

# Production pipeline for a project like Mostly Mortals
# Stage 1: Generate base image (choose tool based on need)
#   - Midjourney V7 for scene illustrations (aesthetic priority)
#   - NB2 for UI mockups and product shots (precision priority)
#   - GPT Image 1.5 for exploratory concepts (iteration priority)

# Stage 2: Refine and composite
#   - Krea.ai for upscaling and enhancement
#   - Adobe Firefly for Photoshop compositing and inpainting

# Stage 3: Final output
#   - Topaz Gigapixel for print-resolution upscaling
#   - Recraft V4 for vector conversion (logos, icons, SVGs)
#   - Illustrator for final vector cleanup

The key insight: generation is only the first step. Refinement, compositing, and format conversion are where the professional output lives. Treating any single tool as the entire pipeline produces amateur results.

When to use which

After six months, my decision matrix is straightforward:

  • Need text in the image? Nano Banana 2. Nothing else is reliable.
  • Iterating with a client? GPT Image 1.5. The conversational model saves rounds of revision.
  • Hero image or campaign asset? Midjourney V7. The aesthetic quality justifies the manual workflow.
  • Product photography? Nano Banana 2. The photorealism is unmatched for commercial use.
  • Exploring a creative direction? GPT Image 1.5 first for speed, then Midjourney V7 for polish.
  • Batch processing or API integration? GPT Image 1.5 or NB2. Midjourney lacks a proper API.

The real answer

The best AI image generator is all three of them, used intentionally for what each does best. Picking one and ignoring the others means leaving quality on the table. The tools have different strengths because they were built with different priorities, and those priorities map cleanly to different stages of creative work.

Build a pipeline. Learn each tool’s strengths. Stop looking for the one tool that does everything. It does not exist, and chasing it wastes the time you could spend shipping better work.

← Back to journal