Multimodal Understanding

Multimodal Understanding

Fine-Grained Image–Text Retrieval

Fine-Grained

A next-generation vision-language model for fine-grained cross-modal retrieval. It addresses the bag-of-words effect and imprecise local alignment found in CLIP-derived models, surpasses SigLIP 2 and MetaCLIP 2, and supports both English and Chinese.

High-Precision Multimodal RAG

Multimodal RAG

A multimodal embedding model for high-precision retrieval across text, images, video, and mixed-modality inputs. It helps enterprise document understanding, retrieval, and question answering move beyond OCR plus plain text to represent multimodal information directly.

Multimodal VLA

Agent-Ready

Built for web and computer automation, this capability helps foundation models generalize to enterprise applications. It enables model-driven interaction with internal web systems without relying on brittle DOM identifiers and improves as operational experience accumulates.

Multimodal Generation

Multimodal Generation

High-Consistency Makeup Transfer

Controllable Image

High-fidelity makeup transfer for virtual try-on, ecommerce, and avatar workflows while preserving subject identity, with reusable template-based experiences.

Controllable Image Generation

Layer Editing

Precise layout control for complex, multi-subject scenes and layer-aware generation for professional creative workflows.

Controllable Image Editing

Video Creation

Decompose existing images into editable layers or extract a selected object as its own layer, giving professional creators precise control instead of unpredictable trial and error.

Controllable Video Generation and Editing

Controllable Image

Control the placement of multiple subjects within a shot and decompose existing video into editable element layers for professional post-production.

Tech Blog

Tech Blog

View all
FLUX-Makeup: One-Click Makeup Transfer for Creative Face Painting

FLUX-Makeup: One-Click Makeup Transfer for Creative Face Painting

FLUX-Makeup transfers makeup from a reference image to a source portrait in one click. No physical makeup or professional retouching is needed—just a face photo and a reference look.

360 AI Research · 2026-10-08
Stop Rerolling: Edit AI Images Layer by Layer with Reveal-Layer

Stop Rerolling: Edit AI Images Layer by Layer with Reveal-Layer

Reveal-Layer separates an AI-generated image into independent, transparent, editable layers. Like layers in Photoshop, each element can be refined individually, making iterative editing part of AI image creation.

360 AI Research · 2026-09-11
How VisFlow Operates Web Pages Without Dedicated APIs

How VisFlow Operates Web Pages Without Dedicated APIs

VisFlow observes the page people actually see, interprets its current state, and chooses the next action without requiring a dedicated business API.

360 AI Research · 2026-08-27
VisFlow Enters Public Beta: Giving General-Purpose Agents Real Browser Execution

VisFlow Enters Public Beta: Giving General-Purpose Agents Real Browser Execution

The VisFlow public beta connects coding and general-purpose agents to reliable browser execution through a remote MCP service.

360 AI Research · 2026-08-03
Meet VisFlow: A Browser Agent That Learns from Your Demonstrations

Meet VisFlow: A Browser Agent That Learns from Your Demonstrations

VisFlow combines natural-language task understanding, visual browser operation, and demonstration learning for controllable web workflows.

360 AI Research · 2026-06-30
Beyond One-Off Generation: Reveal-Layer Brings Photoshop-Grade Editability to AI Images

Beyond One-Off Generation: Reveal-Layer Brings Photoshop-Grade Editability to AI Images

Reveal-Layer decomposes a single image into independent RGBA layers so creators can edit, rearrange, and reuse visual elements.

360 AI Research · 2026-06-16
FG-CLIP Wins Silver in the 2026 CSIG Innovation Technology List

FG-CLIP Wins Silver in the 2026 CSIG Innovation Technology List

FG-CLIP received a silver award for fine-grained cross-modal understanding and its transition from research to deployed services.

360 AI Research · 2026-06-02
Two 360 AI Research Projects Accepted to ICML 2026

Two 360 AI Research Projects Accepted to ICML 2026

RevealLayer and FG-CLIP 2 were accepted to ICML 2026, advancing editable image decomposition and fine-grained bilingual vision–language understanding.

360 AI Research · 2026-05-06
Two Multimodal Generation Projects Accepted to CVPR 2026

Two Multimodal Generation Projects Accepted to CVPR 2026

RefTON and NAMI improve virtual try-on and high-resolution image generation through more practical conditioning and faster inference.

360 AI Research · 2026-04-07
FLUX-Makeup: High-Consistency Makeup Transfer Without Extra Face-Control Modules

FLUX-Makeup: High-Consistency Makeup Transfer Without Extra Face-Control Modules

FLUX-Makeup transfers makeup from a reference image while preserving identity, facial structure, and background consistency.

360 AI Research · 2026-01-27
From Multimodal Research to Real-World Impact: 360 AI Research in 2025

From Multimodal Research to Real-World Impact: 360 AI Research in 2025

A year in review spanning fine-grained vision–language alignment, enterprise retrieval, controllable generation, detection, and edge inference.

360 AI Research · 2025-12-30
FG-CLIP 2 in a Smart Elder-Care Monitoring Course Project

FG-CLIP 2 in a Smart Elder-Care Monitoring Course Project

A student project shows how fine-grained vision–language understanding can support practical, multilingual monitoring scenarios.

360 AI Research · 2025-12-10
Research Highlights

Selected Publications

FG-CLIP: Fine-Grained Visual and Textual Alignment ICML 2025

FG-CLIP: Fine-Grained Visual and Textual Alignment

A next-generation cross-modal VLM built for fine-grained understanding

RefVTON: Person-to-Person Try-on with Additional Unpaired Visual Reference CVPR 2026

RefVTON: Person-to-Person Try-on with Additional Unpaired Visual Reference

End-to-end virtual try-on without complex human-detection preprocessing

LMM-Det: Make Large Multimodal Models Excel in Object Detection ICCV 2025

LMM-Det: Make Large Multimodal Models Excel in Object Detection

Native object detection for LMMs without an external detector

PlanGen: Towards Unified Layout Planning and Image Generation in Auto-Regressive Vision Language Models ICCV 2025

PlanGen: Towards Unified Layout Planning and Image Generation in Auto-Regressive Vision Language Models

A unified model that plans object layouts and generates images

IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities AAAI 2025

IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities

A ControlNet-inspired adaptor architecture for multimodal LLMs

Bridge Diffusion Model: Bridge Chinese Text-to-Image Diffusion Model with English Communities AAAI 2025

Bridge Diffusion Model: Bridge Chinese Text-to-Image Diffusion Model with English Communities

Native Chinese text-to-image generation compatible with the Stable Diffusion ecosystem

Qihoo-T2X: An Efficient Proxy-Tokenized Diffusion Transformer for Text-to-Any-Task ICLR 2025

Qihoo-T2X: An Efficient Proxy-Tokenized Diffusion Transformer for Text-to-Any-Task

An inference-efficient architecture for Text2Any tasks

HiCo: Hierarchical Controllable Diffusion Model for Layout-to-Image Generation NeurIPS 2024

HiCo: Hierarchical Controllable Diffusion Model for Layout-to-Image Generation

Layout-controllable image generation for professional creative workflows