<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Debug School: Jerry Simth</title>
    <description>The latest articles on Debug School by Jerry Simth (@jerrysimth).</description>
    <link>https://www.debug.school/jerrysimth</link>
    <image>
      <url>https://www.debug.school/images/TYf4xUz38_6uwbodTpLBELerRkRUh1nmhkWtzhFY6Nk/rs:fill:90:90/g:sm/mb:500000/ar:1/aHR0cHM6Ly93d3cu/ZGVidWcuc2Nob29s/L3VwbG9hZHMvdXNl/ci9wcm9maWxlX2lt/YWdlLzk5MC8xNWYz/MjgwZC00Nzc0LTQ2/NWItYTZlYy01NmI3/NWUwZDUxODQucG5n</url>
      <title>Debug School: Jerry Simth</title>
      <link>https://www.debug.school/jerrysimth</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://www.debug.school/feed/jerrysimth"/>
    <language>en</language>
    <item>
      <title>Same prompt, six AI image models: a no-cherry-picking shootout</title>
      <dc:creator>Jerry Simth</dc:creator>
      <pubDate>Mon, 09 Nov 2026 08:58:38 +0000</pubDate>
      <link>https://www.debug.school/jerrysimth/same-prompt-six-ai-image-models-a-no-cherry-picking-shootout-54jg</link>
      <guid>https://www.debug.school/jerrysimth/same-prompt-six-ai-image-models-a-no-cherry-picking-shootout-54jg</guid>
      <description>&lt;p&gt;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: &lt;strong&gt;the same two prompts, six models, one generation each — the first result is the one you see. No re-rolls.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;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.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prompt 1 — photorealism
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;A lighthouse on a rocky coast at golden hour, waves crashing against the rocks, dramatic orange clouds, photorealistic, shot on 35mm film&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Honestly? All six produced something publishable. The differences are taste and drama, not competence:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Nano Banana Pro&lt;/strong&gt; went full cinema — deep-red storm sky, the biggest wave, lamp already lit.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;FLUX 2 Pro&lt;/strong&gt; (5 credits) had the most natural light: backlit spray and layered rock that read like an actual photograph. The value standout.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GPT Image 2.6&lt;/strong&gt; nailed the "35mm film" brief — warm, classic, spray frozen mid-crash.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nano Banana 2.1&lt;/strong&gt; was the brightest and cleanest; &lt;strong&gt;Seedream 5 Pro&lt;/strong&gt; the calmest and most balanced; &lt;strong&gt;Z-Image Turbo&lt;/strong&gt; softer and simpler — a striped tower, a gentler sea, and completely usable for 1 credit.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://www.debug.school/uploads/articles/rrm39d60xap8cn8eb2nh.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://www.debug.school/uploads/articles/rrm39d60xap8cn8eb2nh.webp" alt="Nano Banana Pro lighthouse" width="960" height="640"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.debug.school/uploads/articles/coee1rnaltpjd99lburu.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://www.debug.school/uploads/articles/coee1rnaltpjd99lburu.webp" alt="FLUX 2 Pro lighthouse" width="960" height="644"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Prompt 2 — the spelling test
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;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&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;This is where it got interesting. &lt;strong&gt;Five of the six spelled MORNING BREW perfectly:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;GPT Image 2.6&lt;/strong&gt; went further and wrote its own tagline — "Good Coffee. Brighter Day." — spelled correctly, in script, unprompted.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nano Banana 2.1&lt;/strong&gt; produced the richest vintage design of the six (three extra slogans, every word correct) at 10 credits — out-designing models that cost twice as much.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nano Banana Pro&lt;/strong&gt; was the cleanest and most logo-like; &lt;strong&gt;Seedream 5 Pro&lt;/strong&gt; got even the small badge text right; &lt;strong&gt;FLUX 2 Pro&lt;/strong&gt; spelled it right but left the layout sparse.&lt;/li&gt;
&lt;/ul&gt;

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

&lt;p&gt;&lt;a href="https://www.debug.school/uploads/articles/4n3yxdipkbyy3gxk5xww.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://www.debug.school/uploads/articles/4n3yxdipkbyy3gxk5xww.webp" alt="GPT Image 2.6 poster" width="960" height="640"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.debug.school/uploads/articles/ubng7rs7dt9hretkv0b5.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://www.debug.school/uploads/articles/ubng7rs7dt9hretkv0b5.webp" alt="Nano Banana 2.1 poster" width="960" height="644"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.debug.school/uploads/articles/sxj7mfldzzes59iu2k9l.webp" class="article-body-image-wrapper"&gt;&lt;img src="https://www.debug.school/uploads/articles/sxj7mfldzzes59iu2k9l.webp" alt="Z-Image Turbo poster" width="960" height="640"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I actually learned
&lt;/h2&gt;

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

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

&lt;p&gt;&lt;em&gt;All images in this post are first, unedited outputs of the prompts above, generated November 9, 2026.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>machinelearning</category>
      <category>productivity</category>
      <category>webdev</category>
    </item>
  </channel>
</rss>
