30초 요약
Matthew Berman reacts to the January 2026 LTX-2 release and NVIDIA’s demonstration. He is enthusiastic about synchronized audio and video, local use and creative controls; the stated speed, duration and openness require context about benchmark conditions, hardware and the model license.
- 1An audio-video modelBerman highlights a model that can generate video, sound and dialogue together.
- 2Local customizationHe discusses local use and fine-tuning, subject to hardware and license conditions.
- 3A speed comparisonThe 18× figure is a vendor-reported per-step comparison against Wan 2.2 on H100 under specific settings.
- 4Creative controlsThe video describes camera, pose and lip-sync workflows; actual control depends on the tools and configuration.
핵심 정리
- LTX-2 generates synchronized audio and video and can be used in local workflows on suitable hardware.
- The published community license includes conditions and is not an unrestricted grant.
- The approximately 18× figure is a vendor comparison with Wan 2.2 under specific H100 settings, not a universal speedup.
- The video’s 60-second claim should not be assumed for every LTX-2 configuration.
챕터별로 다듬은 전문 읽기
The LTX-2 demonstration
Berman reacts to an NVIDIA and ComfyUI demonstration of synchronized generation. The model and its local use have hardware and license conditions.
I am excited about LTX-2, a model that can generate video and synchronized audio, including dialogue. NVIDIA demonstrated a workflow in ComfyUI, and I see it as a step toward giving individuals more creative capability on their own computers. Comparing it to a billion-dollar studio is an expression of enthusiasm, not a measured production equivalence.
The model weights can be downloaded for local workflows and adapted, including through fine-tuning or LoRA techniques, on suitable hardware. I call it fully open source, but the LTX-2 community license contains conditions and restrictions, including a paid license requirement for certain larger commercial entities. Availability of weights does not mean every use is unrestricted.
Speed and creative control
He discusses duration, a speed comparison and editing controls. The stated figures need the relevant benchmark and configuration context.
I say clips can reach 60 seconds and that LTX-2 is 18 times faster than Wan. The 18× figure is Lightricks’ comparison of step throughput against Wan 2.2 14B on an H100 under specified generation settings; it does not guarantee an 18× end-to-end speedup on every machine. The initial LTX-2 release announcement described up to 10 seconds of synchronized audio and video, while NVIDIA described up to 20 seconds in its demonstration; the video’s 60-second figure should not be generalized to every configuration.
I highlight first-to-last-frame animation, local RTX optimization, camera movement, actor poses and lip-synced dialogue. These are capabilities demonstrated or discussed in particular workflows, not a guarantee of precise control or realistic physics in every generated clip. I think such tools could eventually let more people make shows and films, but that is a prediction about the future.
같은 채널의 다른 글
34:21EN
Matthew Berman· 2026년 10월 27일Matthew Berman Examines Anthropic’s IPO Filing, AI Growth and GovernanceMatthew Berman, Sam Altman · Forward FutureThe presenter reviews the S-1’s financial metrics, compute strategy and voting structure while questioning the tension between rapid AI development and safety.
0:26ENMatthew Berman· 2026년 10월 25일Sam Altman on Human Values and Problem-Solving in an AI WorldSam Altman · OpenAIIn a short interview excerpt, Altman considers the enjoyment of solving hard problems and the human role in interpreting results.
0:36ENMatthew Berman· 2026년 10월 25일Kimi K3.1 and a Rubik’s Cube Solver DemonstrationMatthew Berman · Forward FutureMatthew Berman shows a scrambled virtual cube being solved and gives the Kimi K3.1 test a favorable assessment.
0:21ENMatthew Berman· 2026년 10월 24일An AI-Generated Kimi Benchmark Chart Used Invented ScoresMatthew Berman · Forward FutureMatthew Berman asks for a graph of Kimi K3.1 Pro benchmarks, then discovers that the displayed figures are illustrative placeholders.
25:50ENMatthew Berman· 2026년 10월 24일Kimi K3.1 Pro Tested: Strong 3D Demos, Uneven Results and Competitive PricingMatthew Berman · Forward FutureMatthew Berman tests the open-weight model on coding, 3D rendering and visual explanations, then compares benchmark results and prices.
32:20ENMatthew Berman· 2026년 10월 23일Who Is Local AI For? The Costs, Trade-offs and a Hybrid FutureMatthew Berman · Forward FutureMatthew Berman weighs data control and independence against hardware cost, setup complexity and inference speed.