PaperFlume
⌕
Accounts · soon

Library · Diffusion models · Multimodal LLMs

0:00

BLIP3-o: A Family of Fully Open Unified Multimodal Models—Architecture, Training and Dataset

Jiuhai Chen; Zhiyang Xu; Xichen Pan; Yushi Hu; Can Qin; Tom Goldstein; Lifu Huang; Tianyi Zhou; Saining Xie; Silvio Savarese; Le Xue; Caiming Xiong; Ran Xu

arXiv:2505.09568 4:04 6 chapters CC0 1.0 · adaptation, not endorsed by the authors Source checked · visual review incomplete

Download the deck .pptx Transcript .txt · cited Captions .vtt

Editable text and shapes, not pictures of slides.

Feedback

Was this worth your five minutes?

We are in beta. Tell us whether it helped — and if something is wrong, tell us that too.

Not rated
Goes to a person, not a form queue.
Next

Papers near this one.

Walkthroughs

4:46

JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation

2411.07975
shares Diffusion models
4:15

BLIP3o-NEXT: Next Frontier of Native Image Generation

2510.15857
shares Diffusion models
4:05

Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling

2501.17811
shares Diffusion models

Short cuts

0:47

JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation

2411.07975
shares Diffusion models
0:47

BLIP3o-NEXT: Next Frontier of Native Image Generation

2510.15857
shares Diffusion models
0:47

Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling

2501.17811
shares Diffusion models
Authors

Want one of these for your paper?

Send the PDF and we make the walkthrough, the deck and the reel. It stays private unless you publish it yourself.

Add your paper How it works
01

Something people finishFive minutes of method beats a PDF nobody opens past the abstract.

02

Slides for your next talkAn editable deck built from your own figures, with the narration in the notes.

03

A cut you can postFifty seconds, vertical, for the places a paper link gets ignored.

04

A page that can be foundA cited transcript sits on the page, so the words in your method are searchable.

Based on “BLIP3-o: A Family of Fully Open Unified Multimodal Models—Architecture, Training and Dataset” by Jiuhai Chen; Zhiyang Xu; Xichen Pan; Yushi Hu; Can Qin; Tom Goldstein; Lifu Huang; Tianyi Zhou; Saining Xie; Silvio Savarese; Le Xue; Caiming Xiong; Ran Xu · arXiv:2505.09568 · CC0 1.0 · licence checked 2026-10-27 · Changed from the original: rewritten as a narrated walkthrough with slides, using the paper's own figures. An adaptation, not endorsed by its authors.

Request takedown — we aim to answer within 72 hours.