A model’s speciality is defined by what inputs it was trained and configured on:
- Text-to-Image
- Generate images from written prompts.
- See post [Link].
- Image-to-Image
- Transform an existing image into a new version.
- See post [Link].
- Inpainting / Outpainting
- Edit a specific area of an existing image.
- See post [Link].
- Image Upscaling / Enhancing
- Improve the resolution and detail of low-quality images.
- See post [Link].
- Other specialities
- Style Transfer
- Control-Based Generation
TRAINING AND INFERENCE
Diffusion is a technique for systematically destroying data and then reconstructing it.
It begins with a Forward Process, where Gaussian noise is added to a clean image in successive steps until it becomes pure static (left to right).
Inference (aka “denoising”) is the Reverse Process, where a model, guided by a text prompt, tries to predict which noise was added at each step, gradually reconstructing a clean image.
Generated images can be quite impressive. Check out my AI Gallery at [Link].
READ MORE