That time of the year again!

We've been covering AI as it happens all year.

Time to pause, reflect, and put it on a podium.

Starting with image generation.

It saw the most rapid improvements this year. We've been talking about it lately, so it made sense to close the loop.

January: AI couldn't spell words without turning them into gibberish. December: Production-quality outputs with perfect typography in seconds.

I tested 12 models this year and the winners are as follows.

We put together an AI Survival Hackbook for 2026: a practical guide on how AI changed work between 2019–2025, what will matter next year, and how professionals are quietly using AI for leverage (not hype).

It includes:

It’s free for the next two weeks.

This category has a clear specialist.

Ideogram 3.0 launched in March specifically to solve typography with native text rendering.

As of today, it’s a tiebreaker for models being able to replicate text in image accurately among various models, I chose Ideogram because it goes beyond just getting the text right.

For logos, posters, infographics or anything requiring readable text: Ideogram is killing it.

Caveat: Text rendering is all Ideogram is good at. Photorealism and artistic work? Read further.

Honorable mention: GPT Image 1.5. This could go either way between Nano Banana Pro and GPT Image 1.5, but OpenAI deserves credit here.

Look at the skin texture. The lighting. The way shadows fall.

Nano Banana Pro is really good at generating believable images.

The technical reason: Gemini 3 Pro generates "thought images" before the final output. Better compositional logic.

Runner-up: FLUX 2 Pro

I love that Midjourney keeps winning this.

They've continued to dominate this area.

Nothing else comes close to it for pure aesthetic quality. The color grading, the composition.

V7's Draft Mode is a great add-on. It is now 10× faster, and is half the cost.

The caveat (there is always one): Text rendering is still terrible.

Runner-up: Stable Diffusion

Fully open source. Runs locally. Zero cost per image once you have the hardware.

95,000+ models on Hugging Face, has a massive community and extensive tooling support (ComfyUI, A1111 WebUI).

Runner-up: FLUX.2 [pro]: Better quality, non-commercial license for the model (but outputs are commercially usable).

Google went from irrelevant to leader in 86 days.

August 12: Anonymous model "nano-banana" hits #1 on LMArena.
August 26: Public launch. 3D figurine trend explodes.
September: 23 million new users. 500 million images generated. Gemini app #1 on App Store.
November 20: Nano Banana Pro.

Runner-up: Black Forest Labs (FLUX 1.1 → FLUX 2, 12B → 32B parameters), but Google's consumer adoption was impressive.

DALL-E 3 just became irrelevant.

ELO ~984. 51% win rate (basically ties against average competition).

The model that pioneered mainstream AI image generation now loses to Flux, Midjourney, Nano Banana, Imagen, and even Midjourney v6.1 from 2024.

OpenAI effectively admitted this.

March: GPT Image 1.
December: GPT Image 1.5.
They rebranded DALL-E and tried to rejuvenate it.

The company's focus shifted to Sora (video) and GPT-5 (reasoning). Image generation got deprioritized until last week.

94.7% character consistency accuracy. Same face, same features, across dozens of images.

Released May 29. Three variants:

The technical breakthrough: Processes reference images alongside prompts. No separate fine-tuning needed. 8× faster than GPT-Image-1.

FLUX 2 extended this to 10 reference images at once with >95% consistency but didn't win for a reason.

Runner-up: Flux. 2

Yeah, GPT Image 1.5 hit #1 on LMArena last week. Doesn't matter.

I ran both through my standard tests. Nano Banana Pro won every single time.

Nano Banana Pro gets faces, and logos right. It gets the intent behind your prompt, not just the words.

Runner-up: GPT Image 1.5

2025 solved the image generation problem.

Text rendering, photorealism and character consistency are now problems of the past.

Use:

But see what we're still asking these models to do: Generate one image at a time. Manual iteration. No memory across sessions. Character consistency that still breaks the moment you switch to video.

2026's fight is going to be about:

Google’s rapid rise should tell you how fast this space moves. The model that wins 2025 might be irrelevant by March.

Next edition: AI for education and where the big dogs stand.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉Try Attio Pro for free

#AD 2

Google AdSense's Auto ads lets you designate ad-free zones, giving you full control over your site’s layout and ensuring a seamless experience for your visitors. You decide what matters to your users and maintain your site's aesthetic. Google AdSense helps you balance earning with user experience, making it the better way to earn.

Take control of your ads