Postmodern Pygmalion Applied AI // Design Nerd // Avid Generalist
NYC
- Tweets 779
- Following 559
- Followers 67
- Likes 4,122
Pinned Tweet
Most image benchmarks test whether models can place objects and attributes.
I wanted to know which ones offer photographic control.
So I built the Filmmaker’s Benchmark to evaluate how film stock, lighting, and scan-treatment prompts change color, contrast, and texture.
1
4
1,285
Have been evaluating a bunch of image editing tools, and guys... Ideogram 4.5 is SOTA
I've been testing it on increasingly complex images, and it just never fails. I really hope @ideogram_ai open sources it soon
22
This is what I'm talking about! beautiful
We built the most physically accurate film emulation available. We’re releasing it for free.
126 stocks modeled on real film. Runs directly in your browser, @osmo_studio, DaVinci Resolve, and Premiere. Supports ARRI, Sony, Canon, RED, and more.
Here’s what it looks like:
34
Something needs to be done about the Arena leaderboards. I can't believe they rank the image models in that order, and then people share the graphs like it's gospel
11
Felix Brener retweeted
Introducing Ideogram 4.5, the most precise edit model.
With each edit, leading models add artifacts, pixel shifts, and color changes. Ideogram 4.5 eliminates artifact buildup, making multi-turn editing possible.
Live in Ideogram, the API, and launch partners. Open weights soon.
251
453
5,431
1,000,777
Felix Brener retweeted
asked opus 5.5 to make a music video with some of my old material. thought this was pretty pretty good
99
100
1,162
72,556
The next generation of Alibaba's Qwen Image 2.1 model is finally open-weight and out today. I had a chance to try it out before its release as one of fifty early access invitees on ModelScope and it certainly delivered. More to come soon!
1
152
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
5,703
20,135
120,478
75,036,536
2026 is the best time in human history for people who are smart in some domains but struggle with execution. Set up an agent to keep track of it. Delegate. You can’t blame your struggles on ‘spiky intelligence’ anymore
There is a certain type who will claim they have IQ 130+, but also claim that they can't independently figure out *extremely* mundane processes, like how to apply to colleges, how to run simple businesses, etc.
"The Birth of a Poster."
1
57
Give Codex or Claude access to Grok and tell it to point research subagents at X
For AI and systems work, X has way more alpha than the broader web
44
With Sol, I didn't notice a huge difference in usage between normal and fast mode, but Astra on fast is a whole different beast. Something to keep in mind...
26
Portrait I drew at 16 versus portrait you drew in your 30s
99% of “AI artists” will tap out if you hand them a piece of paper with a pencil and ask them to draw a person standing in front of them. They won’t do it. They can’t.
See, when you invest decades practicing a skill, you develop something called pride. (Not the bad kind.) We don’t want other hands touching our work, our signature, our legacy. When you can race a motorcycle at 200 mph, you don’t need or want training wheels anymore. It would be embarrassing to use them. Your peers would laugh.
Fake artists want to be peers, but they didn’t earn it. They will get no respect. Every professional knows how to spot a fake.
1
36
Went on vacation, and came back to see that the 6-bit MLX Qwen I made is at nearly 2k downloads on @Huggingface!
I’ll be pushing an update shortly to improve some long-session behavior
1
69
Felix Brener retweeted
You can just RL a coding model to paint with javascript btw
251
783
8,495
1,720,634
Felix Brener retweeted
The new image carousel is the biggest UX downgrade in this site's history. There was an entire art form around 4x image polyptych, especially among Japanese photographers, who I learned a lot from stylistically.
With 4 seen all together you create a mood, its very impressionist
74
284
5,201
220,349
Maybe a weird take, but I’m currently more interested in image gen speedups than better image quality. Give me Z-Image Turbo’s realism but at 2–3 seconds on consumer hardware.
This portrait quality is already more than enough for me
38
