Midjourney V6 Llava: 100K+ AI Images with LLaVA-1.6 Captions

Containing between 100,000 and 1,000,000 Midjourney V6 images re-captioned using the LLaVA-1.6 vision-language model. Released by brivangl in October 2024, the data serves as an augmented version of the CortexLM/midjourney-v6 repository for multimodal research.

Use Cases

• Fine-tuning text-to-image diffusion models using LLaVA-1.6 captions for better prompt alignment
• Training vision-language models to recognize and describe Midjourney-specific aesthetic styles
• Benchmarking synthetic captioning quality against original Midjourney prompts

Strengths

• Scale of 100,000 to 1,000,000 records
• Captions generated using the LLaVA-1.6 vision-language model
• MIT licensed

Limitations

• Captions are synthetic and may contain model hallucinations
• Legal ambiguity regarding the copyright of the underlying Midjourney V6 images
• Lack of human-verified ground truth labels

Provenance

Source

CortexLM/midjourney-v6

Collection Method

synthetic annotation via LLaVA-1.6 model

Freshness

Last updated October 2024.