Yuanbao, Qian Qiao's cat

Qian Qiao

Independent Researcher
Shanghai, China

Researching video generation, multimodal models, and efficient learning.

👋 I am Qian Qiao (乔谦), an independent researcher. I received my M.Sc. and B.Eng. from Soochow University, advised by Professor Fanzhang Li. My previous research focused on multimodal understanding, image generation, text spotting, and few-shot learning.

My current interests are real-time video generation, joint video–audio generation, efficient fine-tuning, and multimodal models. I also have four years of experience in blockchain technology and investment. I am actively seeking collaborators in these fields.

Experience

Soochow University logo M.Sc. & B.Eng. Soochow University
Soul AI Lab logo Researcher Soul AILab Real-time Interactive Video Generation
SAC
DeFi Researcher SAC 2017 – 2021
🎓Google Scholar|Selected papers and citations

Featured Open-Source Projects

yyyyyxie / TextFlux

OCR-free diffusion-transformer synthesis for high-fidelity multilingual scene text.

PythonTextFlux GitHub stars

MindLab-Research / LongStraw

Long-context reinforcement-learning infrastructure beyond 2M tokens under a fixed GPU budget, using resident-prefix response-only GRPO.

PythonLongStraw GitHub stars

MindLab Research / UI4A

A component-native Generative UI harness, with Macaron Artifacts providing local runtimes and streamed previews for Claude Code and Codex.

TypeScriptMacaron Artifacts GitHub stars

My research develops practical generative and multimodal systems, with an emphasis on high-quality real-time generation, efficient model adaptation, and visual-language understanding.

Agent Models, Long-Context RL & Generative UI
LongStrawarXiv · 2026→ UI4AMind Lab · 2026→ Macaron-V1Research Release · 2026
Real-Time Video & Audio Generation
SoulX-FlashTalkTechnical Report · 2026→ SoulX-FlashHeadTechnical Report · 2026→ Uni-ViGUTechnical Report · 2026
Multimodal Understanding & Generation
AIMAAAI · 2025→ TextFluxEurographics · 2026→ TransDiffTechnical Report · 2025
Efficient Learning & Model Compression
QPrunerNAACL · 2025→ LLM CompressionICLR · 2026

Selected publications

LongStraw and Macaron-V1 project visual
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU BudgetarXiv, 2026 · #1 Hugging Face Paper of the Day[Paper] [GitHub] [NVIDIA NeMo]
UI4A benchmark visual
UI4A: A Component-Native Harness for Generative UIMind Lab, 2026 · Core Contributor[Article] [GitHub]
Macaron-V1 project visual
Introducing Macaron-V1Mind Lab, 2026 · Research Release[Article]
Uni-ViGU paper visual
Uni-ViGU: Towards Unified Video Generation and Understanding via a Diffusion-Based Video GeneratorTechnical Report, 2026[Project] [arXiv]
SoulX-FlashHead paper visual
SoulX-FlashHead: Oracle-Guided Real-Time Streaming Talking HeadSoul AILab, 2026[arXiv] [GitHub]
SoulX-FlashTalk paper visual
SoulX-FlashTalk: Audio-Driven Real-Time Streaming Digital HumanSoul AILab, 2026[arXiv] [GitHub]
TextFlux paper visual
TextFlux: An OCR-Free DiT Model for High-Fidelity Multilingual Scene Text SynthesisEurographics, 2026[GitHub]
Large language model compression paper visual
Large Language Model Compression with Global Rank and Sparsity OptimizationICLR, 2026
TransDiff paper visual
Marrying Autoregressive Transformer and Diffusion with Multi-Reference AutoregressionTechnical Report, 2025
AIM paper visual
AIM: Let Any Multimodal Large Language Models Embrace Efficient In-Context LearningAAAI, 2025
TALDS-Net paper visual
TALDS-Net: Task-Aware Adaptive Local Descriptors Selection for Few-Shot Image ClassificationICASSP, 2024
Lie group Laplacian support vector machine paper visual
A Lie Group Laplacian Support Vector Machine for Semi-Supervised LearningNeurocomputing, 2025
DeepTTS paper visual
DeepTTS: Enhanced Transformer-Based Text Spotter via Deep Interaction Between Detection and Recognition TasksPRICAI, 2024
QPruner paper visual
QPruner: Probabilistic Decision Quantization for Structured Pruning in Large Language ModelsNAACL, 2025[arXiv]
DNTextSpotter paper visual
DNTextSpotter: Arbitrary-Shaped Scene Text Spotting via Improved Denoising TrainingACM MM, 2024[arXiv] [Project]

Other Publications

  • ACL 2026 — Deputy: Accelerating Large Language Model Inference with Dynamic Low-Rank Substitution
  • ACL 2026 — CoRE: A Fine-Grained Code Reasoning Benchmark Beyond Output Prediction
  • ACL 2026 — Balancing Fidelity and Plasticity: Aligning Mixed-Precision Fine-Tuning with Linguistic Hierarchies
  • BVRCC: Bootstrapping Video Retrieval via Cross-Matching Correction
  • Temporal-Conditional Referring Video Object Segmentation with Noise-Free Text-to-Video Diffusion Model
  • STRA: A Simple Token Replacement Strategy Alleviating Exposure Bias in Text Generation
  • Efficient Fine-Tuning of Quantized Models via Adaptive Rank and Bitwidth

Academic Services

  • Conference Area Chair: ICME 2026, PRCV 2025
  • Program Committee Member: ICLR, ICML, NeurIPS, ACL ARR, AAAI, IJCAI, ACM MM, CVPR, ECCV, ICME, ICASSP, and more.
  • Journal Reviewer: IEEE TIP, IEEE TMM, IEEE TCSVT, Pattern Recognition, Neurocomputing.

Collaboration

I welcome discussion and collaboration on real-time video generation, audio-driven avatars, multimodal models, and efficient fine-tuning.

✉️ Contact me by email.

Honors & Awards

  • 2025 — Outstanding Graduate Award (Top 5%)
  • 2024 — National Scholarship, M.Sc. student
  • 2023 — Special Grade Scholarship, M.Sc. student

Academic Competitions

ICADR 2024 — Text Map Challenge (Click to view results ↗)

With my collaborator Yu Xie, we won three first places and one second place, and were invited to present a technical report.