Generate a dataset

Describe what you want the AI to learn to see. We create labeled images and package them for training — no photos, no hand-labeling.

Dataset name What should the images show? (prompt) Objects to detect (classes — comma separated)

Each phrase becomes one label, e.g. person, hard hat, safety vest

Free plan: you can use up to 6 classes per generation.

Class definitions IMPORTANT

Strongly recommended for accurate labels. Describe what each class looks like so the labeler can tell similar classes apart (e.g. cracked vs intact, defective vs good). Skipping this often causes mislabeled lookalike classes. ONE definition per line.

Number of images

Free plan: you can generate up to 15 images per generation.

Image size Environments (optional — ONE per line, rotated per image) Capture / realism variety (optional — ONE per line, rotated per image) Augmented copies per image

Extra training variants (flips, lighting, blur) so the model handles real photos. 0 = off.