Text-to-image generation

Here’s the prompt we used for our comparison:

“A young woman with shoulder-length red hair wearing a white shirt and denim jacket, standing on a cobblestone street at sunset. She holds a vintage camera and smiles slightly, looking directly at the viewer. The light is warm and cinematic, with soft shadows and realistic skin texture. Ultra-realistic photograph, 85mm lens, shallow depth of field.”

Why this prompt: We chose it because it’s a solid stress test for realism, detail, and character consistency.

The specific description and natural background make it easy to compare outputs across models and its neutral photographic style keeps the focus squarely on generation quality.

Also, having a consistent character gives us a clear benchmark for testing and editing.

Results: In round one, it all came down to realism and attention to detail.

Midjourney immediately pulled ahead of the pack. Its image looks like an actual photograph, nailing the lighting and realistic skin texture.

Nano Banana secured a strong second place, offering great color reproduction and a natural-looking pose, even if it was slightly soft on finer details like fabric textures and sharp facial features.

As for GPT and Grok? They were decent, but their images had that classic “AI look”. A bit too polished and artificial, which killed the natural vibe.

What to look for when writing text-to-image prompts? When it comes to text-to-image prompts, precision is everything. The more detailed your description, the better the result.

Always define the core elements: the character, the setting, the lighting, and the mood. Avoid generic terms like “beautiful photo.” Instead, get specific with details like “soft golden light at sunset.”

Use natural but structured language. Start with the main subject, then layer in the details. And if you’re testing different models, keep your prompts consistent.

Most importantly, write like a director, not an observer: describe what you want to see, not just what you’re thinking about.

The time-saver: 82% of creative professionals report that generative AI speeds up their workflow.

Editing existing images

Generating from scratch is great. But what’s just as crucial, especially in marketing, is the ability to quickly edit or expand on what you already have.

Features like inpainting (filling in or replacing parts of an image) and outpainting (extending an image beyond its original borders) let you adapt visuals for different formats or channels without needing a whole new photoshoot.

The Playbook? Here’s the prompt we used:

“Keep the original scene exactly the same. Change only two details: the woman’s hair is now blonde instead of red, and her denim jacket is now a brown suede jacket. Do not alter the lighting, pose, background, or camera style.”

Key insight:In this editing task, the models split into two camps.

• Midjourney and Nano Banana nailed the key challenge of keeping the character’s identity consistent, but they struggled to follow the full prompt, only changing the jacket.
• On the flip side, while GPT and Grok made both changes precisely (hair and jacket), they did it at the cost of character consistency, essentially creating entirely new people.

For marketing use, the Midjourney and Nano Banana approach is the much safer bet.

Rules for editing prompts: When writing prompts for editing, think about precision. Limit your changes to only what you really need to fix.

Always start with a phrase like, “keep everything the same except…” This gives the model clear guardrails. Describe the modifications precisely: hair color, type of clothing, lighting… but don’t try to change the whole scene.

Finally, avoid stacking too many commands at once. One prompt, one specific change. Remember, AI works best when you tell it what not to touch, not just what to change.

The Priority: Over 90% of managers state that AI will be key to reducing costs over the next 18 months.

Generating and maintaining consistent characters (brand consistency)

Imagine if Coca-Cola changed its red every season?

Visual consistency is the holy grail of every brand. But maintaining it is often a logistical and budgetary nightmare.

Want to feature the same model in your winter and summer campaigns? Normally, that means two separate photoshoots. With AI? A few clicks, zero extra cost. Wow.

Here’s the prompt we used:

Create the same woman as in the attached photo, but in a completely new scene and outfit. Scenery: a cozy winter city at night snowflakes gently falling under the glow of street lamps, soft golden light shining from shop windows, light snow on the ground. Outfit: a cream-colored coat, woolen scarf, matching hat and gloves, holding a warm coffee cup in her hands. Mood: warm, cinematic, and romantic like a scene from a Christmas love movie. Ultra-realistic photography, shallow depth of field, 85mm lens, soft lighting, detailed skin texture, natural expression.

Our opinion: Midjourney passed the test of maintaining character consistency in a new setting with flying colors. The generated image is not only consistent with the original but also impressive in its artistic quality and attention to detail.

Nano Banana proved to be a solid tool for this type of task, faithfully recreating the character and scenery.

GPT had noticeable issues with creating a natural facial expression, while Grok failed to maintain character consistency and didn’t deliver on all the prompt’s requirements.

Key rules for brand consistency:

• Build a visual brand library. Start with one key character or element—like a mascot in your brand colors, with a specific pose and studio lighting. Establish a consistent style: soft light, sharp outlines, and a recurring color palette.
• Replicate that theme across different scenarios: in the office, at a coffee shop, against a city backdrop, or during everyday activities. This way, every image will be instantly associated with your brand, regardless of the context.
• Batch edit your images. After generating a series, do a quick retouch to align the contrast, tone, and overall vibe to keep everything tight.

What you get: A visual system ready for campaigns, emails, and ads that builds brand recognition without the design chaos.

The Expansion: The global AI image generator market is booming. According to a Grand View Research report, it’s projected to grow from $349.6 million in 2023 to a whopping $1.08 billion by 2030.