Jiaming Song

859 posts

Jiaming Song banner
Jiaming Song

Jiaming Song

@baaadas

Chief Scientist @LumaLabsAI. #LumaDreamMachine

Katılım Kasım 2014
1.1K Takip Edilen10K Takipçiler
Sabitlenmiş Tweet
Jiaming Song
Jiaming Song@baaadas·
Excited to introduce Uni-1, our new *unified* multimodal model that does both understanding and generation: lumalabs.ai/uni-1 TLDR: I think Uni-1 @LumaLabsAI is > GPT Image 1.5 in many cases, and toe-to-toe with Nano Banana Pro/2. (showcase below)
Jiaming Song tweet media
English
29
53
413
96.8K
Jiaming Song
Jiaming Song@baaadas·
#3 on Image Edit. #3 on Text-to-Image. @arena The compute we did it with would surprise you. Proud of this team @LumaLabsAI ! (A slightly more detailed report will come out soon)
Jiaming Song tweet media
Arena.ai@arena

Exciting news: UNI-1.1-Max and UNI-1.1 debuts making @LumaLabsAI the #3 lab in the Image Arena across both Text-to-Image and Image Edit! These are versions released without agentic search. Text-to-Image Arena - UNI-1.1-Max #6 overall (1193), +12 points over MAI-Image-2 - UNI-1.1 #7 overall (1190), +13 points over Reve-v1.5 Multi-Image Edit Arena - UNI-1.1-Max #7 overall (1315), +21 points over Seedream 4.5 - UNI-1.1 #8 overall, (1298) Single-Image Edit Arena - UNI-1.1-Max #7 overall (1337) - UNI-1.1 #11 overall, (1310) on par with Grok-Imagine-Image (20260207) Congratulations to @LumaLabsAI on this solid performance!

English
4
17
152
24.7K
Alex Nichol
Alex Nichol@unixpickle·
@JiaweiYang118 It's kind of insane that this actually works. I assumed it would immediately lead to a degenerate solution that produces adversarial examples
English
3
0
23
2.6K
Jiawei Yang
Jiawei Yang@JiaweiYang118·
Two months ago, I vaguely posted a number: 0.9 FID, one-step, pixel space. Now it is 0.75, and can be even lower. Many wonder how. I thought it might end as a small FID prank: simple and deliberate. It started with one question: can FID be optimized directly, and what does it reveal? Introducing FD-loss.
Jiawei Yang tweet media
English
55
155
919
210.1K
Shengjia Zhao
Shengjia Zhao@shengjia_zhao·
Excited to share what we’ve been building at Meta Superintelligence Labs! We just released Muse Spark, our first AI model. It's a natively multimodal reasoning model and the first step on our path to personal superintelligence. We've overhauled our entire stack to support scaling, and this is just the beginning. ai.meta.com/blog/introduci…
Shengjia Zhao tweet media
English
75
171
1.7K
231.8K
Nicholas Boffi
Nicholas Boffi@nmboffi·
all i wanna do is rip claude
English
2
0
13
2.6K
Jiaming Song
Jiaming Song@baaadas·
This is such an amazing idea that I should never have given up….
ok@okaris

personalization is back. most systems still fake it. we went back to our paper from two years ago, rebuilt it from the ground up, applied newer techniques, and replaced the base model with a flow-matching transformer. the result: our pipeline now beats all SOTA personalization approaches on FT-Arena. introducing Person2Person Diffusion 2. code and weights are out now (MIT, commercial use allowed). okaris.github.io/p2p-diffusion.…

English
3
1
56
10.6K
Tim Salimans
Tim Salimans@TimSalimans·
Touching down now in SF for onboarding at Anthropic! After 7 great years at Google, I'm excited to take on a new challenge and help make Claude even better. Grateful for everything I learned at Google DeepMind and Brain, looking forward to what's next.
English
64
18
1.2K
257.7K
Jiaming Song retweetledi
William Shen
William Shen@shenbokui·
Most image models are good at one thing. Uni-1 has been good at everything we've thrown at it. Our team generated thousands of images leading up to Uni-1 launch. We embedded them all into a single map where visual similarity determines proximity. The result speaks for itself.
English
23
44
326
48.2K
A.I.Warper
A.I.Warper@AIWarper·
For those wondering, UNI-1 by Luma can do a lot of unique styles as well. The retro anime in particular is really dope.. pretty unique vibes to it. Any suggestions on others to test? Comment below and I can run a few for some (provide an image if you want)
A.I.Warper tweet media
English
14
7
130
6.7K
Jiaming Song retweetledi
Luma
Luma@LumaLabsAI·
We are loving the energy around Uni-1! Quick note since we’re seeing questions: With Luma Agents, requests can route across models. If you want to make sure you’re using Uni-1, here’s how: - Select Create Image → Uni-1 - Or, explicitly ask the agent to use Uni-1 - Check the model label on outputs to confirm API access coming soon for more direct testing. Keep the feedback coming, and keep on creating → lumalabs.ai/uni-1.
English
19
45
255
19.5K
Jiaming Song
Jiaming Song@baaadas·
Here is the prompt and thinking in case you are wondering.
Jiaming Song tweet mediaJiaming Song tweet mediaJiaming Song tweet mediaJiaming Song tweet media
English
0
1
13
1.4K
Jiaming Song
Jiaming Song@baaadas·
A fixed text token length is a limitation of most existing text-to-image models. In Uni-1, we engineered around this, so that it can take at least 3500 tokens (even in markdown) if you want to promptmaxx. The differences is huge between Uni-1 (1st image) and Flux 2 (2nd image).
Jiaming Song tweet mediaJiaming Song tweet media
Atelier SG@_AtelierSG_

This has to be a prank.

English
1
5
57
7.3K

Keşfet

@oliver_wang2 @arena @LumaLabsAI @FEijkelboom @osclsd @daan_roos_ @mmbronstein @wellingmax