AWhat changed
Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods.
Scientific scope
Generating an image in one network pass is attractive, but simply making a GAN wider can destabilize training. GigaGAN combines a bank of filters selected for each example, attention between image regions and words, and training signals at several image scales. A base generator makes a small image, and a separate text-conditioned upsampler adds resolution. Kang and colleagues demonstrate billion-parameter-scale adversarial generation and editable latent representations. The contribution is this specified architecture and its reported experiments, building on GANs, StyleGAN, CLIP and earlier attention and dynamic-filter methods. Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.
Evidence
The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. Original timings concern specified A100 configurations and include different generation or upsampling stages. Training is substantial, and one upsampler uses internal stock images. Comparisons differ in data, training, encoders and output resolution. A lower distributional image score does not imply better realism or correct prompt composition: the authors explicitly show weaker visual results than production DALL·E 2. Individual ablations have mixed metrics. Upsampling can invent detail rather than recover the original scene; better perceptual scores do not mean better pixel fidelity.
Contributors
- The named authors in the research record · SCIENCE: Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.
Provisional WRS assessment, edition 2026-10-23.9. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction