An AI image generator for non-English speakers

2026-03-17 · Source: ΑΙhub · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, short

Summary

Researchers at the University of Amsterdam Faculty of Science, in collaboration with Cohere labs, have developed NeoBabel, an open-source AI image generator designed to address digital inequality by supporting six languages: English, French, Dutch, Chinese, Hindi, and Persian. Unlike most English-centric models that rely on translation, NeoBabel directly generates images from multiple languages, preserving linguistic and cultural nuances and outperforming competing models in non-English prompts. The team made their entire research pipeline, including code and an expanded image-label dataset (from 40 million to 124 million pairs), publicly available to foster inclusive AI development. NeoBabel also features improved data labeling using multilingual language models, resulting in a smaller model with fewer parameters. Future applications include multilingual collaborative creative canvases and the long-term goal of expanding to video generation, though this requires more computational power and collaboration.

Key takeaway

NeoBabel is an open-source, multilingual AI image generator supporting six languages (English, French, Dutch, Chinese, Hindi, Persian) that directly generates images, avoiding translation-induced nuance loss. It matches English-centric model performance while outperforming them in other languages, leveraging an expanded 124M image-label dataset and a smaller model footprint. This breakthrough enables truly inclusive, collaborative visual content creation for non-English speakers and fosters open research into culturally specific AI, addressing digital inequality.

Topics

• Multilingual AI
• Text-to-Image Generation
• Open-Source AI
• Digital Inclusion
• Dataset Expansion

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, AI Researcher, AI Engineer, AI Student

Related on AIssential

• OpenAI Beefs Up ChatGPT’s Image Generation Model — OpenAI's ChatGPT Images 2.0 enhances image generation with multi-image and multilingual text capabilities.
• Multilingual and Multimodal LLMs in the Wild: Building for Low-Resource Languages — Multimodal LLMs must integrate text, speech, and vision for low-resource languages, moving beyond English-centric benchmarks.
• Annotating Nigerian English for AI: Practical Insights from a Native Lagos-Based Annotator — Nigerian English annotation requires native expertise to prevent AI model inaccuracies and cultural misrepresentation.
• Analysis of Machine Translators on Sentences Generated by Portuguese Image Captioning Models — Machine translation quality directly impacts image captioning model performance when adapting to new languages.
• AI’s English Problem—and Why We Should Care — English dominance in AI training data creates a global language gap, reinforcing digital divides.
• Now in Foundry: Qwen3-Coder-Next, Qwen3-ASR-1.7B, Z-Image — New Hugging Face models offer efficient, high-performance AI for coding, multilingual ASR, and advanced image generation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ΑΙhub.