arXiv:2611.06270v1 Announce Type: cross
Abstract: Adversarial training enables one- and few-step image and video generation, but conventional adversarial losses supervise critic values without directly specifying the gradients used to update the generator. We present NEXTGAN, which trains a scalar critic by regressing its input gradient onto displacements between independently paired reference and generated features at interpolated locations in frozen representation spaces. For a fixed generator, this regression is equivalent to an adversarial objective with a gradient-energy penalty, while the generator retains the standard potential-based GAN update. The same objective adapts pre-trained image and video generators with up to 14B parameters, without auxiliary losses, online teacher scores, or trajectories. On ImageNet $256 \times 256$, NEXTGAN achieves one-step FIDs of 2.87 with pixel-space JiT-B and 1.54 with latent-space DiT-XL/1; its adversarial form further reaches 1.36. For text-to-video generation, a single trained Wan2.1-1.3B model supports one-, two-, and four-step sampling, achieving the highest VBench total scores among the compared 1.3B methods at one and two steps. At two steps, NEXTGAN scores 83.13, exceeding rCM and AnyFlow by 1.25 and 0.47 points, respectively. Code is available at https://github.com/LINs-lab/NextGAN.git.
arxiv
NextGAN: Towards Simple and Scalable GAN Training for Image and Video Generation
NEXTGAN trains a scalar critic by regressing its input gradient onto displacements between independently paired reference and generated features at interpolated locations in frozen representation spaces. The authors state that, for a fixed generator, this regression is equivalent to an adversarial objective with a gradient-energy penalty, while the generator keeps the standard potential-based GAN update. The method adapts pre-trained image and video generators with up to 14B parameters without auxiliary losses, online teacher scores, or trajectories. On ImageNet 256x256 it reports one-step FIDs of 2.87 with pixel-space JiT-B and 1.54 with latent-space DiT-XL/1 (1.36 for its adversarial form), and a single Wan2.1-1.3B model supports one-, two- and four-step text-to-video sampling, scoring 83.13 VBench total at two steps.
It offers a simpler adversarial training recipe for few-step image and video generation, backed by one-step FID and VBench comparisons, which is useful for researchers and practitioners building fast generative models.
Comments & Questions (0)
No comments yet
Be the first to comment!