How AI turns a sentence into an image
Process map by Brian. 17 steps, 2 layers. Tags: ai, tech.
- You: Write a prompt
- Text encoder: Turn the words into a meaning vector
- Random noise: Start with a screen of pure random noise
- Denoising model: Predict which noise to remove, guided by the prompt
- Denoising model: Remove a little noise
- Denoising model: Is it clear enough?
- Safety filter: Check for disallowed content
- Safety filter: Is it allowed?
- You: You get an image
- You: Are the hands right?
- You: Done
- You: Rewrite the prompt and try again
- Safety filter: Block or blur it