AI content supply and demand are strong; NVIDIA is focusing on local AI creation

21st Century Business Herald Reporter Peng Xin

Not talking about gaming graphics cards, NVIDIA’s focus at this year’s GDC (Game Developers Conference) is to promote more users to use AI for content creation, further lowering the development threshold.

During the game developer conference held this week in San Francisco, NVIDIA announced the launch of a simplified “Application View” for the AI drawing tool ComfyUI, introduced RTX Video super-resolution, and provided NVFP4 and FP8 variant support for models like LTX-2.3, FLUX.2 Klein, improving the local AI video generation process.

NVIDIA aims to shift the originally niche generative AI workflow towards a broader audience of beginners and designers. According to NVIDIA, the new “Application View” in ComfyUI is mainly aimed at designers unfamiliar with node workflows. Users only need to input prompts, adjust parameters, and click to generate AI content, lowering the barrier to AI content production.

(Image source: NVIDIA)

In terms of performance optimization, NVIDIA states that since September last year, ComfyUI’s experience performance on RTX GPUs has improved by 40%; on GeForce RTX 50 series GPUs, performance can be boosted up to 2.5 times when using NVFP4 data format, with a 60% reduction in VRAM usage; using FP8 data format, performance can be increased by up to 1.7 times, with a 40% reduction in VRAM consumption.

NVIDIA is also trying to improve the efficiency of 4K video production. It explains that previously, creators would generate low-resolution previews first to ensure efficiency, then upscale the quality. Currently, processing 10 seconds of 4K video usually takes several minutes. In response, NVIDIA states that with RTX Video super-resolution technology, users can now quickly upscale videos to 4K. This technology consumes very little VRAM and is 30 times faster than other mainstream local resolution enhancement tools.

The background for NVIDIA’s vigorous promotion of local AI video production workflows is the significant growth of AI-generated content. In early 2024, OpenAI’s Sora brought “AI-generated videos” from concept to reality with astonishing visual realism, gaining market acceptance. ByteDance and Kuaishou quickly followed up, launching Ji Meng and Ke Ling, rapidly attracting a group of professional creators to produce low-cost AI short videos. By 2025, “AI-generated videos” exploded rapidly, entering a stage of large-scale growth, and have now secured a place on content platforms.

Taking the currently popular AI manga dramas as an example, AI manga dramas refer to a new content form that uses generative artificial intelligence technology to directly convert online short story IPs into dynamic images through multimodal large models, achieving automated production of scene generation, character modeling, and voice synthesis. Market forecasts show that by 2026, China’s manga drama market size is expected to surpass 24 billion yuan, with over 300 million users. “Enabled by AI toolchains, the traditional production cycle of several months has been compressed to weeks or even shorter, with costs reduced by over 60%. Several leading platforms are sequentially deploying AI manga drama businesses, promoting the formation of an independent content track,” said Zhou Yuan, founder and CEO of Zhihu.

With strong demand for AI-generated content, NVIDIA is trying to lower the barriers to local AI creation tools through software, models, and toolchain collaboration, binding its GPUs and AI production workflows to cater to the AI production wave. The reporter learned that the mainstream production of AI short dramas has now formed a mature process: first, an intelligent agent generates prompts, then ComfyUI workflow generates images and converts them into videos, followed by storyboard adjustments and detailed refinement, and finally lip-syncing and sound effects are added. Behind this complex workflow, almost all rely on NVIDIA GPU computing power.