Biitz retweeted
饾棯饾棽've 饾棶饾榾饾榾饾棽饾椇bled 饾椇饾椉re 饾榿饾椀an 饾煴饾煬0 饾棽饾棾饾棾饾棽cts, 饾榿饾棽饾棸饾椀饾椈iques, 饾棶饾椈d 饾椆饾椉饾椉ks 饾棾饾椏om 饾榿饾椀e 饾椀饾椂饾榾饾榿ory 饾椉f 饾椊饾椀饾椉饾榿饾椉饾棿raphy, 饾棾饾椂lm, 饾棿饾椏饾棶饾椊hic 饾棻饾棽饾榾ign, 饾棶饾椈d 饾棻饾椂饾棿饾椂tal 饾槂饾椂饾榾ual 饾棸饾槀饾椆饾榿ure.
饾棓饾椏饾椏饾棶nged 饾棸饾椀饾椏饾椉饾椈饾椉饾椆饾椉饾棿ically 饾棷y 饾棻饾棽饾棸ade 饾棶饾椈d 饾椏饾棽饾棶dy 饾榿o 饾棸饾椉py 饾棶s 饾椊饾椏饾椉饾椇pts.
饾棗饾椂饾榾饾棸over 饾榿饾椀em 饾棶t atlas.letz.ai 馃挜
Biitz retweeted
JEV for images, free to use on @inference_sh until the end of the month.
And deployed on EU servers 鉁岋笍馃槍
autotrust/jev-27b-vl is now live!
free to use until the end of October (limited with capacity)
it can make instant decisions on multiple questions in parallel and can evaluate images.
it's currently #2 on the @huggingface jev index for vision decision models.
Biitz retweeted
Nano Banana 2.1 is really good at maintaining likeness. Best model for that so far!
Now available on @letz_ai 鉁岋笍
Biitz retweeted
We've redesigned the LetzAI landing page to better explain Canvas, and I think it turned out really cool.
If you haven't tried Canvas yet, you definitely should!
Biitz retweeted
MIRA - Main Theme
Collab with Grammy-recognized sound engineer @serge_courtois
Biitz retweeted
This is our submission to the #FutureVisionXPRIZE, a global competition from the XPRIZE Foundation, Google and Range for optimistic science fiction about a future worth building toward.
Massive shoutout to Steven Quach for helping me bring this together in less than a week!
Biitz retweeted
AI movies no longer look like AI.
Introducing MIRA, a new original AI demo series.
Episode 1, made entirely with LetzAI and Seedance 2.5.
馃 Made with AI
Biitz retweeted
You can just use Seedance 2 with 1080p/4k to upscale your Seedance 2.5 720p scenes icydk
Biitz retweeted
A lot of comments on this about how these plans are not sustainable etc.
People don't seem to know how the AI Inference industry works.
There are different ways to make unlimited plans sustainable, including black-hat techniques.
The way many discounters operate is that they rely on bot farms that scrape the internet for open endpoints, free chat UIs, apps with free trial plans, and so on.
With today's AI and Computer vision capabilities, it's very easy to build agents that use available open interfaces across the internet to relay requests, wait for the reply, and scrape the result.
Because these indirect calls usually take longer to complete, many platforms hide unlimited plans behind "longer waiting times".
We've been attacked this way ourselves across different tools. It's the reason why we withdrew the free trial from Letz a few months ago, for example.
The white-hat alternative to offering unlimited is, of course, through partnerships with model providers, startup grants, and so on. VCs and platforms throw A LOT of money into securing distribution to acquire as many users as possible. It's a gold rush, because churn on AI platforms has changed from being super high to super low, and users are starting to stick to the platforms they've been using for the past year or so. It becomes harder and harder to acquire new users.
Because of all the above, running an AI platform has turned into a race to the bottom. Every new day, another platform comes along that wants to compete with free or impossibly low prices to secure users. There is a black market for AI inference that makes everything worse.
It's a very weird market to be in, and I think we don't speak enough about this.
