# PolyGen: Fully Synthetic Vision-Language Training via Multi-Generator Ensembles

arXiv: https://arxiv.org/abs/2602.01370 · HTML: https://mlsanity.com/paper/2602.01370/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 27 of its 39 references that are in the map.
- First posted on arXiv 1 Feb 2026.

## Placed in

- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 27 of 27)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://mlsanity.com/paper/2102.05918/) (2021)
- [Describing Textures in the Wild](https://mlsanity.com/paper/1311.3618/) (2013)
- [Fine-Grained Visual Classification of Aircraft](https://mlsanity.com/paper/1306.5151/) (2013)
- [Adversarial Diffusion Distillation](https://mlsanity.com/paper/2311.17042/) (2023)
- [Contrastive Learning with Hard Negative Samples](https://mlsanity.com/paper/2010.04592/) (2020)
- [When and why vision-language models behave like bags-of-words, and what to do about it?](https://mlsanity.com/paper/2210.01936/) (2022)
- [On the detection of synthetic images generated by diffusion models](https://mlsanity.com/paper/2211.00680/) (2022)
- [Demystifying CLIP Data](https://mlsanity.com/paper/2309.16671/) (2023)
- [SugarCrepe: Fixing Hackable Benchmarks for Vision-Language Compositionality](https://mlsanity.com/paper/2306.14610/) (2023)
- [Improving CLIP Training with Language Rewrites](https://mlsanity.com/paper/2305.20088/) (2023)
- [StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners](https://mlsanity.com/paper/2306.00984/) (2023)
- [Scaling Laws of Synthetic Images for Model Training ... for Now](https://mlsanity.com/paper/2312.04567/) (2023)
- [Filtering, Distillation, and Hard Negatives for Vision-Language Pre-Training](https://mlsanity.com/paper/2301.02280/) (2023)
- [Learning Vision from Models Rivals Learning Vision from Data](https://mlsanity.com/paper/2312.17742/) (2023)
- [SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?](https://mlsanity.com/paper/2402.01832/) (2024)
- [TripletCLIP: Improving Compositional Reasoning of CLIP via Synthetic Vision-Language Negatives](https://mlsanity.com/paper/2411.02545/) (2024)
- [SUGARCREPE++ Dataset: Vision-Language Model Sensitivity to Semantic and Lexical Alterations](https://mlsanity.com/paper/2406.11171/) (2024)
- [Bridging the Gap: Enhancing the Utility of Synthetic Data via Post-Processing Techniques](https://mlsanity.com/paper/2305.10118/) (2023)
- [Stable Diffusion Dataset Generation for Downstream Classification Tasks](https://mlsanity.com/paper/2405.02698/) (2024)
- [Federated Knowledge Recycling: Privacy-Preserving Synthetic Data Sharing](https://mlsanity.com/paper/2407.20830/) (2024)
- [Synthetic Image Learning: Preserving Performance and Preventing Membership Inference Attacks](https://mlsanity.com/paper/2407.15526/) (2024)
- [Getting More Juice Out of Your Data: Hard Pair Refinement Enhances Visual-Language Models Without Extra Data](https://mlsanity.com/paper/2305.05208/) (2023)
- [Your Image Generator Is Your New Private Dataset](https://mlsanity.com/paper/2504.04582/) (2025)

## Built on by (most cited first, 1 of 1)

- [Synesthesia via Direct Latent Augmentation:Bypassing the Decode-Encode Loop for Cross-Modal Distillation](https://mlsanity.com/paper/2606.08336/) (2026)

## Abstract (arXiv)

Synthetic data offers a scalable solution for vision-language pre-training, yet current state-of-the-art methods typically rely on scaling up a single generative backbone, which introduces generator-specific spectral biases and limits feature diversity. In this work, we introduce PolyGen, a framework that redefines synthetic data construction by prioritizing manifold coverage and compositional rigor over simple dataset size. PolyGen employs a Polylithic approach to train on the intersection of architecturally distinct generators, effectively marginalizing out model-specific artifacts. Additionally, we introduce a Programmatic Hard Negative curriculum that enforces fine-grained syntactic understanding. By structurally reallocating the same data budget from unique captions to multi-source variations, PolyGen achieves a more robust feature space, outperforming the leading single-source baseline (SynthCLIP) by +19.0% on aggregate multi-task benchmarks and on the SugarCrepe++ compositionality benchmark (+9.1%). These results demonstrate that structural diversity is a more data-efficient scaling law than simply increasing the volume of single-source samples.