Back to Releases

Janus Pro 7B

Feature Release
2025-01-27

Major update focusing on AI integration and user experience improvements

Highlights

  • Multimodal understanding and visual generation results from Janus Pro.
  • Comparison of text-to-image generation between Janus Pro and its predecessor, Janus.
  • Architecture of Janus Pro.

Multimodal Understanding

  • Integrates and processes data from both visual and textual inputs.
  • Evaluates performance based on multimodal benchmarks like POPE, MME-Perception, GQA, and MMMU.
  • Scales MME-Perception scores to the range [0, 100] for better interpretability.
  • Provides a comprehensive evaluation across various metrics for improved understanding of complex multimodal data.

Visual Generation

  • Evaluates visual generation on two key benchmarks: GenEval and DPG-Bench.
  • Assesses the model's ability to follow instructions and generate relevant visual content.
  • Improves upon previous state-of-the-art models, including both unified multimodal models and task-specific models.
  • Targets high performance in instruction-based visual generation tasks.

Unified Multimodal Model Performance

  • Combines multiple models into a single, cohesive framework for enhanced multimodal understanding.
  • Outperforms previous state-of-the-art models across both multimodal understanding and visual generation.
  • Demonstrates superior capabilities in handling complex multimodal tasks with more accurate results.

User Experience Optimization

  • Optimizes display for better viewing experience, particularly for evaluation metrics and results.
  • Ensures that the performance benchmarks and evaluation results are best viewed on screen for clearer insights.
  • Focuses on delivering a visually engaging and user-friendly interface for interaction with model outputs.