aiqbench analysis
NEXTGAN introduces a novel training objective that regresses critic gradients directly onto feature displacements, enabling efficient one- and few-step generation for large-scale image and video models. By eliminating auxiliary losses and online teacher scores, this method successfully adapts pre-trained generators with up to 14B parameters while achieving state-of-the-art FID scores on ImageNet and leading VBench results in text-to-video tasks. The approach represents a significant step toward simplifying GAN training pipelines without sacrificing generation quality or scalability.
NextGAN: Towards Simple and Scalable GAN Training for Image and Video Generation
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.
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 this https URL.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2611.06270 [cs.CV]
(or arXiv:2611.06270v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.06270 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 281 words
Related coverage
Math2Visual-X: A Modular Framework for Pedagogically Aligned Lower-Primary Math Visuals Generation
arXiv · Sep 22, 2026
SynthRCT: Scalable Conditional Deformation Synthesis for Synthetic Repeat CT Generation
arXiv · Sep 9, 2026
SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation
arXiv · Sep 30, 2026
EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
arXiv · Oct 7, 2026