Same AI, same random seed, same prompt: an orange cat in an astronaut helmet, floating in space. So why are these five pictures so different? One hidden knob, called guidance. The model cleans up the picture twice, once reading your words and once with no words at all (with no words, it drew a house and a man), then pushes away from the no-words answer toward your prompt. The knob sets how hard.
At 1: a mess. At 3: just the helmet. At 7.5: our cat. At 15: the colours start to burn. Most apps hide this dial and pick the number for you.
AI, Under the Hood #9. Full story: Diffusion Models, Explained in 10 Minutes (pin the link).
Images: our own Stable Diffusion 1.5 runs (seed 1234; CreativeML OpenRAIL-M). Method: classifier-free guidance (J. Ho, T. Salimans, 2021).
#shorts #ai #stablediffusion #latenttheory #generativeai #prompt #aiart #imagegenerator #tech #machinelearning
At 1: a mess. At 3: just the helmet. At 7.5: our cat. At 15: the colours start to burn. Most apps hide this dial and pick the number for you.
AI, Under the Hood #9. Full story: Diffusion Models, Explained in 10 Minutes (pin the link).
Images: our own Stable Diffusion 1.5 runs (seed 1234; CreativeML OpenRAIL-M). Method: classifier-free guidance (J. Ho, T. Salimans, 2021).
#shorts #ai #stablediffusion #latenttheory #generativeai #prompt #aiart #imagegenerator #tech #machinelearning
- Категория
- AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.
