DaVinci Magihuman brings native audio to open-source video
A new 15B open-source video generator called DaVinci Magihuman just dropped, capable of natively generating video with audio across multiple languages. In self-reported blind tests, it beat LTX 2.3 60% of the time with zero censorship, though running even its distilled version requires a hefty 61GB download.
DaVinci Magihuman is a new 15-billion-parameter open-source model designed to natively generate video alongside synchronized audio across multiple languages.
Performance and hardware demands
In self-reported benchmarks, it scored higher than LTX 2.3 in both visual quality and text alignment while maintaining a lower error rate. In human blind preference evaluations, DaVinci beat LTX 2.3 in 60% of comparisons. It also ships completely uncensored right out of the box.
The catch is the hardware requirement. Over on its Hugging Face repository, even the distilled version weighs in at a massive 61GB. That puts it well out of reach for standard consumer GPUs, meaning only those with high-end enterprise VRAM can run it locally right now. The code and setup instructions are available on GitHub and Hugging Face, though most users will need to wait for community quantizations.
Why it matters
Native audio-video generation is clearly becoming the next standard for open-source video models, cutting out the messy pipeline of generating silent video and splicing audio later. Claiming a 60% win rate against a strong open-source heavyweight like LTX 2.3 is no small feat.
That said, a 61GB footprint means it is strictly for researchers and tinkerers with serious hardware at the moment. If you are on consumer gear, we think it makes sense to wait until the open-source community releases smaller quantized variants.
Sources (1)
- youtube.com(opens in a new tab)Originalyoutube.com