Kevin Li

@curiouskid423

multimodal research @adobe | co-lead of firefly image 5 preview | ex vision research @berkeley_ai | 🇹🇼 | opinions are my own

San Jose, CA

thekevinli.github.io

Joined January 2017

734 Following

271 Followers

Replies Reposts Media

Pinned
Kevin Li
@curiouskid423
Oct 16, 2025
(1/n) 🚀 Your VLM can be a great multimodal encoder for image editing and generation if you use the middle layers wisely (yes, plural 😉). We are thrilled to present UniFusion - the first architecture uses only VLM as input-condition encoder without auxiliary signals from VAE
2
Kevin Li
@curiouskid423
Jun 26
Excited to see that @krea_ai shipped UniFusion design with Krea 2! More frontier models now recognize that multimodal semantics are never encoded in just a single layer. In the paper we showed that not only is Layerwise Attention Pooling better than all other conditioning
Manuel Brack
Krea
@krea_ai
Jun 23
00:24
today, we release the open weights of Krea 2. welcome Krea 2 Raw and Krea 2 Turbo, an undistilled model from mid-training meant to be fine-tuned, and a fast distilled version with a wide aesthetic diversity. read the details below 👇
2
Kevin Li
@curiouskid423
May 4
applies to image/video models too... fixing data may be boring but it's *much* more effective than any fancy architecture improvement!
Eric
@ericmitchellai
May 4
I am begging you to look at your data. Please look at the data evals worse than expected? look at the data evals better than expected? *definitely* look at the data evals about what you expected? believe it or not ....
Kevin Li
@curiouskid423
Apr 25
Tomorrow (4/26) at @iclr_conf - I will be presenting UniFusion at the Multimodal Intelligence workshop!! 🇧🇷 📍 Room 204C, 14:30–16:00 Two questions we kept hearing: are VLM semantic embeddings even sufficient to reconstruct or edit images? And if it can — what do you gain over
Manuel Brack
Kevin Li
@curiouskid423
Mar 18
Excited to share that UniFusion is accepted to ICLR Multimodal Intelligence workshop🎉 Cross-modality semantic embedding space is the future. See you all in Brazil 🌴 Camera-ready: openreview.net/pdf?id=V9l4tDF…
1
Kevin Li
@curiouskid423
Apr 20
Heading to #ICLR2026 🇧🇷 — would love to chat about image/video generative modeling and semantic representations! Btw our team at Adobe is hiring. DM if you want to work on foundation models that *actually ship* into products used by millions!
Kevin Li
@curiouskid423
Mar 18
Excited to share that UniFusion is accepted to ICLR Multimodal Intelligence workshop🎉 Cross-modality semantic embedding space is the future. See you all in Brazil 🌴 Camera-ready: openreview.net/pdf?id=V9l4tDF…

Kevin Li

Kevin Li’s posts