Same prompt, six AI image models: a no-cherry-picking shootout

Jerry Simth — Mon, 09 Nov 2026 08:09:05 +0000

I run a small studio site that hosts 13 AI image models side by side, and the question I get most is "which one should I use?" Tier lists are cheap. So I ran an experiment instead: the same two prompts, six models, one generation each — the first result is the one you see. No re-rolls.

The six: GPT Image 2.6 (Medium, 9 credits), Nano Banana Pro (2K, 24 credits), Nano Banana 2.1 (2K, 10 credits), FLUX 2 Pro (5 credits), Seedream 5 Pro (2K, 15 credits) and Z-Image Turbo (1 credit). Prices are what each image costs in the studio; aspect ratio 3:2 throughout.

Prompt 1 — photorealism

A lighthouse on a rocky coast at golden hour, waves crashing against the rocks, dramatic orange clouds, photorealistic, shot on 35mm film

Honestly? All six produced something publishable. The differences are taste and drama, not competence: Nano Banana Pro went full cinema — deep-red storm sky, the biggest wave, lamp already lit. FLUX 2 Pro (5 credits) had the most natural light and is the value standout. GPT Image 2.6 nailed the "35mm film" brief. Nano Banana 2.1 was the brightest and cleanest; Seedream 5 Pro the calmest; Z-Image Turbo softer and simpler — completely usable for 1 credit.

Prompt 2 — the spelling test

A vintage coffee shop poster, wide landscape layout, with the headline text 'MORNING BREW' in bold retro letters, a steaming cup of coffee illustration, warm cream and brown colors, flat design

Five of the six spelled MORNING BREW perfectly. GPT Image 2.6 even wrote its own tagline — "Good Coffee. Brighter Day." — correctly, unprompted. Nano Banana 2.1 produced the richest vintage design at 10 credits. Seedream 5 Pro got even the small badge text right; FLUX 2 Pro spelled it right but left the layout sparse.

And the one real miss is the instructive one: Z-Image Turbo replaced the "RE" in BREW with the coffee cup, so the headline reads "MORNING B–W". At 1 credit it's a brilliant drafting tool — just proofread its lettering before anything ships.

What I actually learned

On photos, price buys drama and polish, not basic competence — a 5-credit model held its own against a 24-credit one.
In-image text has quietly become table stakes: five of six models now nail a two-word headline plus small print.
Every model has a failure mode; the cheap one's just happens to be visible in the headline.

The full 13-model comparison — including editing, GIF generation and bilingual typography, with per-image prices — lives in the model guide on Pizyn (https://pizyn.com/blog/which-ai-image-model/), and the studio itself is at https://pizyn.com if you want to run your own prompts (20 free credits at signup, drafts from 1 credit).

All images in this post are first, unedited outputs of the prompts above, generated November 9, 2026.

DEV Community: Jerry Simth

Same prompt, six AI image models: a no-cherry-picking shootout

Prompt 1 — photorealism

Prompt 2 — the spelling test

What I actually learned