Ir para o conteúdo
Atlas de IA

Imagem para imagem

Regerar uma imagem semelhante a partir de outra

Geração e edição de imagemIniciante #19
entradaImagemImagem

O texto completo é apresentado em inglês; o título e o resumo estão traduzidos.

O QUE ESTA CAPACIDADE SIGNIFICA

Takes an image and outputs a new one that resembles it in content but is rewritten in style or detail. The knob controlling similarity is usually noise strength: less noise stays close to the original, more noise approaches fresh creation. Unlike image editing it does not require a text instruction naming what to change; it restyles or reinterprets the whole image.

Como é feita tecnicamente

The method encodes the input into latent space, adds noise for a chosen number of steps, and denoises back from that point, so the output keeps the structure while taking on a new style. Finer control comes from stacking conditioning branches: edges, depth, pose or a reference style each enter as extra conditions, which is especially common in tasks such as completion and line-art colourisation.

Produtos representativos

6

Organizações relacionadas

Usos típicos

  • Style transfer and photo stylisation
  • Sketch and line-art colouring and completion
  • Iterating from rough layouts to renders
  • Series of variants on one subject

Como avaliar se funciona bem

FID
Distance between results and the target distribution
LPIPS perceptual distance
Perceptual difference from the input image
Structural consistency
How well contours and subject placement follow the input

Limites e dificuldades

  • Balancing structure retention against rewriting is hard; the same settings behave differently across images
  • At higher strength the content drifts, and faces or text are the first details lost
  • Wholesale redrawing wrecks layout, so posters and UI screenshots cannot be preserved as-is

Conceitos por trás