Telegram AI Image Dataset — Cleaned for VLM LoRA Training
A cleaned dataset of 1,050 AI-generated images with their generation prompts, collected from a Chinese Telegram channel focused on GPT-Image-2 prompt engineering.
Each image is paired with a structured generation prompt in Chinese/English. Prompts have been cleaned of channel boilerplate — no bot instructions, hashtags, source credits, model name prefixes, emoji title lines, or channel footer ads.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/GCStream/telegram-channel-dataset.