Creative & Visual Media
fofrAI demonstrated Nano Banana 2’s advanced generative image capabilities through tests recreating exact book pages, including layout, text, illustrations, and handwritten notes, such as the “jellyfish” page from a book.
One of my tests with Nano Banana 2:
> Show the jellyfish page from this book https://t.co/SHvAjbjyka pic.twitter.com/7mR1MAcMX9
— fofr (@fofrAI) February 26, 2026
https://t.co/cOKhqkasr8 pic.twitter.com/98fIj11eRx
— fofr (@fofrAI) February 26, 2026
The model also showed strong performance in prompts like wildlife portraits, anime designs, and text-in-image compositions, often compared favorably or unfavorably to the prior “Pro” version.
Prompt: A photo of a monstera leaf against a simple backdrop, cleverly and organically appearing within the natural gaps of the leaf are the words Nano Banana
Left: pro
Right: nb 2 pic.twitter.com/77fOuYi8J6— fofr (@fofrAI) February 26, 2026
Prompt: A wildlife portrait photo of a snow leopard, viewed from above, a steep drop behind the animal. The leopard stands out from its surroundings, appearing lighter against the dark and shaded mountains behind. pic.twitter.com/iYLkgxCorb
— fofr (@fofrAI) February 26, 2026
Prompt: Anime character design, full color, concept sketch against white. Just the characters, no other sketches or words. A young adult man and woman on their phones, sitting cross legged, back to back. Pick interesting fashion choices, hair style and unusual footwear.
left:… pic.twitter.com/lBHWqlOudr
— fofr (@fofrAI) February 26, 2026
NB2 introduces a cheaper, faster 0.5K resolution option at ~$0.045, undercutting Pro pricing for higher resolutions.
By the way, NB2 comes with a smaller 0.5k resolution, for cheaper and faster generation:
– 512px, 747 tokens, ~$0.045
– 1K (1024x1024px), 1120 tokens, ~$0.067
– 2K (2048x2048px), 1680 tokens, ~$0.101
– 4K (4096x4096px), 2520 tokens, ~$0.151Pro prices for comparison:
– 1K and…
— fofr (@fofrAI) February 26, 2026
Simon Willison commented on Google’s launch of Nano Banana 2, described as their best image generation and editing model, accessible through AI Studio and the Gemini API as Gemini 3.1 Flash Image, noting surprise that it arrived before the base Gemini 3.1 Flash model.
I didn't think we would get Gemini 3.1 Flash Image before we got Gemini 3.1 Flash! https://t.co/MZ80LtDXTu
— Simon Willison (@simonw) February 26, 2026
It's all over IG
— @levelsio (@levelsio) February 26, 2026
Fal released Nano Banana 2, praised for excellent text rendering, realism, advanced image editing, 4x faster generation, and lower cost.
an incredible model in terms of visual presentation https://t.co/vkdZGwRHzZ pic.twitter.com/NRvcVYEqHx
— ilker (@ailker) February 26, 2026
Nano banana 2, first few attempts at complicated toasting
Much faster, and not perfect but real improvements in text and ability to handle complexity – even getting detailed labels right at a level we haven't seen before (though still issues sometimes) https://t.co/52MNKQWVYc pic.twitter.com/t1FxN5KJ6i
— Ethan Mollick (@emollick) February 26, 2026
I had some early access to Nano banana 2. It isn't perfect but it is the first model to handle really complex images and diagrams with some consistency.
"show me a where's waldo set in ancient Venice, but instead of waldo it is an otter wearing a blue striped pilots outfit." pic.twitter.com/MPeB3nTE1w
— Ethan Mollick (@emollick) February 26, 2026
Ethan Mollick tested it on complex prompts like overcomplicated toasting flowcharts and a “Where’s Waldo” scene in ancient Venice featuring an otter in a pilot outfit, noting consistent handling of details, labels, and complexity, with improvements over prior versions though not perfect.
Nano banana 2, first few attempts at complicated toasting
Much faster, and not perfect but real improvements in text and ability to handle complexity – even getting detailed labels right at a level we haven't seen before (though still issues sometimes) https://t.co/52MNKQWVYc pic.twitter.com/t1FxN5KJ6i
— Ethan Mollick (@emollick) February 26, 2026
I had some early access to Nano banana 2. It isn't perfect but it is the first model to handle really complex images and diagrams with some consistency.
"show me a where's waldo set in ancient Venice, but instead of waldo it is an otter wearing a blue striped pilots outfit." pic.twitter.com/MPeB3nTE1w
— Ethan Mollick (@emollick) February 26, 2026
Software Development
Guillermo Rauch highlighted Cloudflare’s vinext claiming 94% Next.js API coverage but criticized it as a superficial checklist, with an agent used for testing failing on complex features like parallel routes and client-side behaviors; real Next.js suite shows only 10-20% compliance.
Well put. Thank you for your @nextjs leadership Jimmy.
I also appreciate all the work you and the team have put into Adapters in collaboration with OpenNext, Google, Netlify, et al.
The world now has good data (test compliance, 9+ vulnerabilities on day one, etc) to make…
— Guillermo Rauch (@rauchg) February 26, 2026
He praised Vercel team’s adapters work with partners like OpenNext for better infra decisions.
Well put. Thank you for your @nextjs leadership Jimmy.
I also appreciate all the work you and the team have put into Adapters in collaboration with OpenNext, Google, Netlify, et al.
The world now has good data (test compliance, 9+ vulnerabilities on day one, etc) to make…
— Guillermo Rauch (@rauchg) February 26, 2026
Alex Volkov noted Anthropic rejecting a “–dangerously-skip-permissions” flag for “WarClaude Code” amid DoW tensions.
Just in: @AnthropicAI refuses to add –dangerously-skip-permissions to WarClaude Code
— Alex Volkov (Thursd/AI) (@altryne) February 26, 2026
Peter O’Mallet shared updates on Desloppify 0.7.2, which now supports 28 languages and biases scoring toward subjective analysis, but noted the upcoming version could run autonomously for weeks if not for unreliable LLM providers.
The version I'm going to ship tomorrow would be able to operate autonomously for weeks if it weren't for unreliable LLM providers https://t.co/OnBsyE1Yr4
— POM (@peteromallet) February 26, 2026
Automation & Orchestration
goose_oss showcased their open-source AI agent Goose integrating with the Neighborhood MCP server to automate ordering lunch, linking to a detailed blog post on implementation.
What if your AI agent could order your lunch. Well it can now with the Neighborhood MCP server. Check out our post showing you how it works and make sure you give it a try yourself.https://t.co/O4KpPB571g
— goose (@goose_oss) February 26, 2026
Hah, yeah "hoard" does kind of hint at Smaug… it's genuinely how I think about it though, I just hoard things in public
— Simon Willison (@simonw) February 26, 2026
@levelsio suggested OpenClaw serves as an entry point for mainstream users to adopt advanced agentic tools like Claude Code or Chat for handling large-scale content synthesis and refinement.
Or Claude Chat* actually in this example
— @levelsio (@levelsio) February 26, 2026
Bro normies don't even know what OpenClaw is.
— Mike Assad (@mikeassad77) February 26, 2026
Machina discussed monetizing OpenClaw for B2C apps.
how to make money with openclaw and B2C apps: https://t.co/fRb7GdIQFj
— Machina (@EXM7777) February 26, 2026
Riley Brown noted Perplexity Computer as a lighter version of Manus, though Manus lags behind OpenClaw in usefulness.
Perplexity computer is basically lite manus.
— Riley Brown (@rileybrown) February 26, 2026
Strategy & Ecosystem
Guillermo Rauch argued that AI has commoditized shipping speed, making “taste, quality, and restraint” the new competitive edge, as seen in recent infrastructure debates.
If you thought your company's edge was "how fast you ship", you're in for a rude awakening.
Everyone can ship fast now. Obviously, not everyone can ship tastefully, with quality and restraint in mind. That's the new edge.
— Guillermo Rauch (@rauchg) February 26, 2026
Cristóbal Valenzuela reiterated his prediction that over 50% of major public software companies will fail in the next five years due to AI shifting focus from tool mastery to intention-driven outcomes.
We are trending pretty well on this prediction. Tbh probably will be more than 50% https://t.co/7z0hOKyDd6
— Cristóbal Valenzuela (@c_valenzuelab) February 26, 2026
Alex Volkov praised Anthropic CEO Dario Amodei’s firm stance against DoW demands for unrestricted Claude access, emphasizing safety over military concessions.
Good on you Dario! 👏
— Alex Volkov (Thursd/AI) (@altryne) February 26, 2026
Dario will not budge! 👏 https://t.co/MhleE3wfln pic.twitter.com/j8TSonElr0
— Alex Volkov (Thursd/AI) (@altryne) February 26, 2026
Emad Mostaque (@EMostaque) shared a chart from Citadel Securities showing software engineer job postings rising amid falling overall postings, sarcastically attributing it to hiring “software claudesters” to automate other roles, predicting temporary gains.
You hire software claudesters to automate away all of the other humans in the company so of course this chart will go up
For a while pic.twitter.com/WXUqELEBKW
— Emad (@EMostaque) February 26, 2026
I was surprised by the word "hoard", because it didn't sound like you, but of course your advice is what I'd expect. Perhaps "collect", "gather", or more prosaic "document"? Hoard makes it sound like a defensive manoeuvre, secreting things away, which has never been your M.O.! 😊
— Jeff Waugh (@jdub) February 26, 2026
He reacted “Yeesh” to Jack Dorsey’s announcement of halving staff at their company due to AI tools enabling smaller teams, amid stock gains.
Yeesh https://t.co/DxUC32M1LR pic.twitter.com/jgAAG5Jas4
— Emad (@EMostaque) February 26, 2026
It's all over IG
— @levelsio (@levelsio) February 26, 2026
Ethan Mollick expressed skepticism about Blocks’ 50% layoffs attributed to AI-driven efficiency, arguing little experience exists in organizing work around agents and that visionary CEOs should use AI for expansion.
Two things:
1) Given that effective AI tools are very new, and we have little sense of how to organize work around them, it is hard to imagine a firm-wide sudden 50% efficiency gain
2) CEOs with vision who hired well should also use AI for expansion & augmentation, not decimation https://t.co/knZxOVTVWn— Ethan Mollick (@emollick) February 26, 2026
Machina urged maxing out AI subscriptions like Claude and ChatGPT to experiment and build amid rapid progress, and recommended parallel deep research with Perplexity and GPT-5.2 Pro (merged via Opus 4.6) for comprehensive coverage.
if you're doing serious research, run perplexity deep research AND gpt5.2-pro side by side
different methodology, different blind spots… together they cover ground nothing else touches (use opus 4.6 to merge both outputs)
gpt takes about an hour to come up with a report tho
— Machina (@EXM7777) February 26, 2026
Riley Brown stated that in 2026, being an influencer is safer than a traditional job.
In 2019 it was somewhat rare and cool to forgo a safe traditional job to be an influencer.
In 2026 it is safer to be an influencer than to get a traditional job.
— Riley Brown (@rileybrown) February 26, 2026