GPT Image 2 dataset of 1,000 dreamcore images hits Hugging Face
Uncanny corridors and empty pools: a new open dataset for fine-tuning SD models.
Reddit user LukaDev13 leveraged OpenAI's GPT Image 2 at 2K medium resolution to generate approximately 1,000 images capturing the liminal space/dreamcore aesthetic—empty indoor pools, foggy parking lots at night, and unsettling corridors. Instead of letting the images sit idle, the user curated them into a dataset and uploaded it to Hugging Face, a leading platform for sharing machine learning resources. The dataset, titled "Liminal-Dreamcore-1K," is publicly available for download and use.
The primary use case for this dataset is fine-tuning generative models like Stable Diffusion to produce similar dreamlike, vacant scenes. It can also serve as a reference collection for artists and researchers interested in the liminal aesthetic. The creator explicitly invites feedback from anyone who uses the dataset for training, hoping to see how models adapt to this unique visual style. This release highlights a growing trend of community-driven dataset creation using cutting-edge image generation tools, democratizing access to specialized training data for niche aesthetics.
- Dataset contains 1,000 images generated with GPT Image 2 at 2K medium resolution.
- Images feature liminal/dreamcore subjects like empty pools, foggy parking lots, and eerie corridors.
- Freely available on Hugging Face, intended for fine-tuning Stable Diffusion or as a reference set.
Why It Matters
Opens up a niche aesthetic dataset for training AI models, democratizing specialized visual data for researchers and artists.