Justin Maier

Autostep uncovers repetitive tasks ready for AI. Then builds or finds the agents to solve them. Your team repeats the same work hundreds of times a week. No one notices. Autostep does. Congrats on the launch, @aidan__pratt! ycombinator.com/launches/PYq-a…

📹 Panel Project Odyssey with Matty Shimura, King Willonius, James Gerde, Nik Kleverov and Justin Maier at #UpscaleConf 2025 - The new AI Film Creative Stack

[Project Name: Origin, Meaning & Humor] The name CivitAI comes from the Latin word Civitas, meaning community. It reflects the platform's goal of fostering a collaborative space for AI creativity. | “What if we have this thing be an AI community?” When the team introduced the logo, the AI part was highlighted in blue, leading users to read the name as Civit AI. | “Honestly, I love it too, because ‘Civit’ is easier to say than ‘Civitai’. I honestly don’t care which way they say it.” Some users associated the name with the civet animal — a Southeast Asian mammal known for producing expensive coffee via digestion. |“The civet animal… they poop stuff out and they use the seeds to make special coffee.” |“Poop equals gold.” @JustMaier expanded the metaphor: |“The whole way a model works is it ingests… and outputs very special content. Like the civet animals create special coffee beans.”

@JustMaier is scaling @HelloCivitai into the world’s leading platform for generative AI creators. He’s sharing the journey and the vision at Upscale Conf SF. 📍 May 20–21 | SFJAZZZ

This week, @JustMaier joins @WhatTheLJW @devrelius to talk about the community of open-source AI artists he’s building at @HelloCivitai 🌐 Catch it tomorrow: Apr 24, 9:00 am PST •⁠ 12:00 pm EST

The cofounders of @HelloCivitai, @JustMaier and @MfHulker, noticed a problem. Creators were making custom models and posting them disparately on forums, but "they'd just kind of vanish after that," Maier says. That's when the idea to create a place for them to share their models and tools was born. The cofounders share their journey from a niche concept to a thriving platform with millions of users👇

🚀Flux is officially available in our onsite generator🚀 To celebrate, we’ve curated a little visual masterpiece featuring some of the most popular Flux images on the website so far! 👀 Don’t have the GPU power to run Flux locally!? Have no fear, get your Flux on here 👇🏽 civitai.com/models/618692/… Animated using @LumaLabsAI DreamMachine 👾 Sound on 🔊 😉

At @a16z, we're big believers in the future of AI art. We couldn't be more excited to share a recap of our first artist retreat hosted last month in NYC. So grateful to all of the creatives and companies who spent time with us. Hear from them 👇

@jboogx_creative Are you gonna make a how to post on Civit?

Ahhhh yes. The future. And just in time for the election cycle 🤣 Live Portrait + Runway Gen 3

@cloneofsimo What does that say about the other models on the market?

I made huge blunder last run. I was actually not following important scaling law regarding batch size and was SIGNIFICANTLY undertraining everything... DAMN IT!!! Literally 5th fundamental mistake I made during training auraflow. htf is this thing SoTA in GenEval?

@Machinedelusion Wow. Finally the nodes we all needed 😂

been playing with JS this weekend, i built tetris and bullet hell into comfyUI nodes #ComfyUI

@OverPowere13959 @cloneofsimo Does it make sense to collect at the highest res possible and offer on-demand downsampling? I’m trying to focus on the long term value of the dataset.

@JustMaier @cloneofsimo Although 1024x1024 is enough, a perfect resolution maybe 1536x1536. Perfect size for both quality and performance. For those who want lower resolution can scale it down.

AuraFlow got a lot more attention than I thought! First time is the hardest, we'll only get better from now. My plan for next version. * Much better Encoder-Decoder, with higher spatial compression ratio, such as stable cascade-level compression (im looking at this btw arxiv.org/abs/2407.02158) * Put a lot more compute * Adopt lot more of recent literature on scaling, like adam-atan2 and paying closer attention to weight decay of adamW, have to dig through this rabbit-hole * Bit smaller version of the model? Like 4B? * Do a lot more data works. I learnt a lot from community review of the AuraFlow And a lot more on fine-tuning stage * A lot more datawork on fine-tuning * Quantization-aware training * DPO maybe?

@cloneofsimo How high resolution? And how many would you need at a minimum?

@JustMaier High quality, complex scene, high resolution dataset is tough to get, and having a large number of those would be awesome

@InnerRefle11312 Civit Article when?!

I am so beautiful! ha ha ha.

@drewisgooden I just watched your last video about AI ruining the internet. Any plans to cover the opposite end of the spectrum where people are using it to make awesome stuff and take memes to the next level? BBL Drizzy is a pretty good example.

Excited to see what this new process can do for the indie film scene.

@shl Very helpful insights. Thanks for taking the time to publish it here on X.

@FAL Awesome. Are you able to do custom nodes yet?

all-in-one comfyUI experience create workflows and turn them into scalable APIs, side by side

@cloneofsimo I love what you're doing and I wonder if there is a way to have this effort be community funded. We've been planning a KickStarter style funding thing for Civitai and it'd be cool to help get you money from the community.

"bUt iT woN't Be aS goOd wiTH yoUR teeNy coMpUte" nah i dont care im not raising cash bro, gaining this experience of handling 100M-scale dataset, pretraining billion-scale vision model from scratch, post-hoc analysis... *all as a hobby in my free time*, is what matters 😎