Leo Antunes (@lenxism) header
Leo Antunes (@lenxism) profile picture

design engineer

Joined December 2019
Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)Photo by Leo Antunes (@lenxism)
got completely obsessed with this animation and spent a few hours recreating it just to understand how it works, ended up open-sourcing it so anyone can play with it github.com/lenxism/dither
Made this interaction of Linear’s dithered logo for our latest release. It’s using canvas with a lot of tiny dots. There’s an invisible circle around the cursor. Any dot inside it gets pushed outward with a cubic falloff so it’s gentle on the edges, and strong at the center.
15
22
3
704
89,649
Hard pill to swallow, ouch
There’s 100% a kind of mania you can get into with coding agents, where it feels like you’re getting a lot done and working all the time but nothing is really happening. I think it's often a form of avoidance of something else in life.
72
FIESP Headquarters in São Paulo, Brazil, 1979. Brutalist architecture that feels straight out of Blade Runner.
52
591
59
7,081
108,979
we need more founders like Ray with actual taste
satisfying :)
1
2
96
we've achieved RSI. recursive slop inbreeding
2
106
hot take: you're not dogfooding enough
1
3
143
Using Opus 5.5 is a hard thing to come back from. It’s incredibly smart, has great design taste, actually respects your instructions, and works really well with agent workflows like pstack. I have 3 banked resets on Codex, but I just can’t get into the current results from Sol 6.1 or Astra, the design taste isn’t there either.
July 2026: Anthropic has the best code models. Gap isn’t very big though. They’re slow, expensive, and the “claudeisms” are at an all time high. And your $200 sub is so limited that you can kill it in a day. Preferring OpenAI models made a lot of sense. You lose a slight edge on capability, but they’re faster, cheaper and more pleasant to use. Your $200 sub feels unlimited, and you get resets from Tibo every 3 days on average September 2026: Anthropic has the best code models. The gap is way bigger now. Opus 5.5 is fast, surprisingly cheap, and the prose is pleasant to read. Your $200 sub suddenly feels nearly limitless. Preferring OpenAI models for code makes almost no sense. You lose a lot of capability and reliability. They’re slow as hell without fast mode, which makes the price far too high. You can kill your $200 plan in a few hours, and the $500 plan goes even faster with UltraFast. I know OpenAI can come back from here, but it’s looking rough right now.
132
hear me out: Opus 5.5 feels like the same paradigm shift for design that devs had a while ago, across graphic design, motion, even 3D sculpting. It still needs a lot of hand-holding, but damn, it’s incredibly good. I honestly thought this would take longer.
126
1) what?
Introducing FLUX 3 Image. Control every pixel. Make precise multi-turn edits without changing any other pixel. Lay out the image exactly how you want using bounding boxes. Generate in up to 4K to preserve details. Use up to 10 references to compose an image. Commercial Weights available for companies running image generation at scale. Open Weights version of FLUX 3 Image is launching in the coming weeks.
137
I don’t know how to spend my Codex credits anymore. anything creative turns into slop. spent ~$800 on Cursor API usage just to use Opus 5.5. back to Claude Code. had a good run.
5
4
344
David does incredible work, and I’m pretty sure a lot of the major AI video platforms use Mediabunny. You should consider sponsoring him, he clearly puts a ton of love into what he builds 🩶🤝
I've been a little sad the last couple of days. Mediabunny's first, longest and most loyal sponsor, Gling, stopped their gold sponsorship of Mediabunny this week. Their CTO @yonatanbd left the company, and they're now restructuring their efforts and cutting costs where they can. An understandable move from their part, and it's got nothing to do with me. I'm deeply grateful for the years of Yonatan's support; he alone was the basically the reason Mediabunny even got started and he supported me through its development for half a year before I released the version to the public. Thanks to his encouragement and financial support (as well as the support of @konstipaulus), I had the courage the quit my job back then and to focus on building cool media stuff for the web full time! On the financial side, I'd be lying if I said losing a gold sponsor doesn't hurt. It does, a little. I'm still alright though! GitHub Sponsorships are my main way of making a living at the moment. If you or your company have gotten considerable value from Mediabunny, consider a donation! I'm sure your logo would look beautiful on the Mediabunny home page ;) I'm working on super cool upcoming stuff that I haven't shared yet, and I'd love to keep working on it. People still haven't experienced the true power of web-based media processing!
170
One thing I like about the Grok bot is how ambitious the agents feel. You can give it a huge, tedious task and it’ll spawn 10 agents to split the work and grind on it 24/7 until it gets somewhere. I’ve had around 10 agents running basically nonstop and I still haven’t used even 20% of my quota. That said, I still think the agents are kinda dumb tbh.
1
1
141
Astra Ultra in Codex feels much faster now, but also way dumber and less creative. One bad idea after another
148
With @recraftai new Flash model, you can now get a cinematic look in seconds and turn it into a video right away
2
4
1,005
Explore dozens of ideas in seconds
1
39
Choose your favorite look and turn it into a video
16
So proud of the @recraftai team for this launch. Getting high-quality images this fast is now a reality, and you can still push the results even further with the more advanced model. What a time!
The fastest image model on the market is here. Zero lag in your workflow. Recraft V4.1 Flash produces images in 1.3 seconds. Pick Flash from the model picker, generate, then hit Refine to clean up the details. Check it out and give it a try ↓
2
189
90% of the demos using Jev here didn’t even spend five minutes understanding what it’s capable of. It’s over, we really are in the slop generation era.
145
The Audeum (Audio Museum) by Kengo Kuma  Seoul, South Korea 🇰🇷
1
37
1
375
12,805
Arrow 2 is insane, you all need to try this. It generated this SVG file github.com/lenxism/arrow-2-s…. Congrats @joanrod_ai and team!
Introducing Arrow 2 Our latest and most advanced models for generating precise, editable vector graphics. Higher quality. Faster outputs. Available now in App and API.
1
1
6
2,379
Maison Hermès Ginza. Rebuilt in Blender.
4
1
1
3
554