Commentary on X
Highest rankedMetrics
11h agopublished9comments5
And finally. SeeDream 5 Pro. IDK, they boasted incredible 4k infographics in the release, I couldn't get it to generate text coherently! It breaks down on complex prompts really badly! https://x.com/altryne/status/2075001131185639521/photo/1
Nobody has run this test yet, so I did 🥊 The 4 frontier image models — Nano Banana Pro, GPT-Image-2, ByteDance's Seedream 5 Pro (out YESTERDAY) and @metaai brand new Muse: given the exact same 5,000-character design brief and the same single reference image. 8 real news stories each. Round 1: Tomorrow's GPT-5.6 launch. In order: Nano Banana Pro, GPT-Image-2, Seedream 5 Pro, Muse 🧵
First, GPT-image-2 is still the absolute king, I'll add the full prompt in alt here, but it still mogs everyone in comparison. Text on point, reference images represented in full detail. The thing it's really bad at is... mazes. It created unsolvable mazes 😂 https://x.com/altryne/status/2074885731978535352/photo/1
Nano Banana Pro is the most pleasant infographics maker, the data has room to breathe. But it's really really bad at input consistency, every Alex is different on every gen. I even have my eye color in the prompt and it's still not getting me correctly. Everything else tho... spot on.
OK Round 2: @OpenAI GPT-Live announcement (the full-duplex voice model cc @juberti and gang 👏) Same order every round — NBP, GPT-Image-2, Seedream, Muse. Watch how differently each one handles dense benchmark tables and fine print. https://x.com/altryne/status/2075000993394343963/photo/1
This is a meta post because I do these infographics to make @thursdai_pod interesting, so this thread is also an announcement of most of the major releases we're going to cover live on air tomorrow Join us 8:30am pacific here or on YT https://www.youtube.com/watch?v=YSfY61_h1cw
Liked this breakdown? Follow me @altryne and @thursdai_pod to stay up to date on everything AI related every week 🫡 https://twitter.com/5721582/status/2075000952403476533
Some users praised GPT-Image-2 as SOTA for consistency and input adherence in the model tests while others criticized SeeDream 5 Pro for incoherent text output and Nano Banana Pro for poor reference imagery.
Based on 7 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.