@VisualInference

postmodern pygmalion // techno optimist // avid generalist

 NYC
felixbrener.com
Joined February 2026
  • Tweets906
  • Following622
  • Followers160
52Media
I spent two days pushing FLUX 3 Klein to its limits, trying to defeat its infamous wax faces. It ended up changing how I think about image models in general. Would you rather have one model that does everything decently, or a lean base you can shape?

FLUX 3 Klein Isn't Broken. One-Size-Fits-All Is.

Article · @VisualInference
8
1
6
2,104
Replying to @LucasSchiavini
I like the landing, very cool
1
12
Replying to @distributedkv
Genius
1
18
Replying to @ErwannMillon
The Krea models are great at natural textures. A lot of other image models produce these weirdly repetitive, almost fractal-ish patterns in things like foliage and clouds. I've noticed this tends to show up when I distill models into turbo variants. Is this something you've dealt with? and if so, how do you avoid it?
64
Replying to @Rafa_Schwinger
the researchers seem pretty cool. im confident they're open to the idea or already working on it
1
1
12
Replying to @VisualInference @somewheresy
Nothing wrong with a bit of help editing, but I hate reading stuff that sounds like “LinkedIn: the comment”
20
Replying to @somewheresy
The most telling part is that the comments were copied and pasted from an LLM So, the posters probably don’t understand the topic well enough to describe it on their own
1
1
23
getting crunk off that whistlepig maple syrup latte
27
Replying to @juliegirltweets
real
1
18
ouuaagghh aahh ahhgg I'm getting poaached. labs! help! they're poaching meee
26
Lately there has been something ominous in the city I struggle to put into words, since I don't want to pick apart mundane things or exaggerate. Yet...
𝓚
@StasinspiceOct 18
Replying to @geetz__romo
Everytime I visit nyc from La it’s like the stereotype of people of La infected nyc
1
112
Hey Apple, what if instead of one Mac Mini I had two Mac Minis
Trevor Wood
@trevorwood22221h
Apple should fund independent AI guys to improve inference on MLX. The hardware is there, just needs the software @johnternus
2
119
Felix Brener retweeted
Shannon Sands
@max_paperclips20h
Really getting tired of the "AI bad because NFTs" argument. From now until the end of time everything new will be NFTs all over again. If you don't understand the problem with making that parallel, your opinion is honestly just to biased to listen to
Kenney@KenneyNLOct 18
Replying to @TimSweeneyEpic
I understand, we're both trying to offer guidance to the next generation. The thing is, speaking as an artist, we've been force fed NFT's, metaverse, blockchain and now AI. Let these things naturally take its course, they'll find their place (or not).
14
17
1
514
8,927
best one yet. they put all of Pinterest in this thing
Ting Chen Liang
@ting_Oct 18
MOSS Image mix texture with picture, paper and glitters
1
72
Replying to @ivanfioravanti @Open_MOSS
The colors look so rich. What a great release!
1
43
Replying to @MosiAI_Official @huggingface @Open_MOSS
Congrats on the MOSS-Image release! I love using it locally
1
81
These pics look noticeably better than GPT-Image 2.5 comeback time?
Lumina
@LuminaBenchOct 18
This is probably GPT Image 3 then aurora-red-alpha is a faster model but worse quality aurora-green-alpha is slower and better quality Haven't managed to test blue out yet
2
52
It's time to move to Metal on Apple Silicon @ComfyUI You are leaving so much performance on the ground by using MPS. I bet you can migrate easily leveraging AI 😉
10
3
82
7,500
MOSS-Image by @Open_MOSS now runs smoothly on a 16GB Mac. Go experiment with it! huggingface.co/VisualInferen…

VisualInference/MOSS-Image-SFT-MLX-compact · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co
Felix Brener
@VisualInference22h
MOSS-Image is the beautiful open image model that writes long English and Chinese text Introducing the best port to MLX. Mine vs mlx-vlm’s: – 4x lower peak memory (9.3 vs 38.7 GB) – half the download – zero text loss at 8-bit – same seed makes same pic as OpenMOSS samples 🧵
92
Round of applause for MOSS-Image everyone
Ting Chen Liang
@ting_Oct 18
MOSS-Image Base vs SFT both Seed 42, 50 steps had so much fun with this theme 🤣 this one prefer the base version prompts below
1
133
MOSS-Image itself is gorgeous. Go try it out and give @Open_MOSS a follow! Demo: huggingface.co/spaces/OpenMO… Site: openmoss.ai/MOSS-Image/ Code: github.com/OpenMOSS/MOSS-Ima… ImageWeights: huggingface.co/OpenMOSS-Team… On a Mac: pip install moss-image-mlx MLX weights: huggingface.co/collections/V… My code: github.com/felix-ab/moss-ima…
75
Load more