Keyword used in 70/71 images: “jp-drawn-style”. It was labeled and included in each image and tags were shuffled during training. Caveat: the sample images were all produced using Chat GPT DALL-E3. Some images used in the dataset were image to image using DALL-3 and photographs I took with a camera. Not all images produced were viable for training with Civit AI and I may upload those later in a different post. Training images are available for download. I used a consistent keyword in all but 1 image which was not tagged.

JP-drawn

JP-drawn_r1-000004.safetensors

1826057_training_data.zip

Files with SHA256: 2001db2bb6a7d2419041089bb2714b5eec183ded176c4c05491d40d4a0bc956e