Justine Moore: Veo 3 lacks image-to-video with audio, reverting to Veo 2
Justine Moore · AI Video Is Eating The World — Olivia and Justine Moore, a16z · Jul 9, 2025 · at 9:58
a16z partner Justine Moore explains technical and interface limitations in Google's Veo generative video model.
0:00 / 0:21exact quote · 21.1s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“Cause the thing to know about VO three that I think a lot of people don't know until you use it is you actually can't do image to video with audio. Google hasn't released that yet. I would assume for trust and safety reasons in the interface, you can start with a frame, but once you do that, if you're doing it on the VO three text to video model, it will switch you back to VO two.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Justine Moore
Justine Moore: Video creators rely on recognizable archetypes for character consistency
“Which means that it makes it super hard for, like, character consistency, right, if you're not able to start with an image of the same character over and over again, which is part of why you're seeing so many of these viral trends using things like a stormtroo…”
Justine Moore: AI avatars let creators bypass social media beauty standards
“Before, honestly, to be, like, an Instagram or YouTube influencer, most of them are hot people. And now it's, like, anyone can be a popular influencer, and you don't have to be a hot person. Like, how many friends do you have that have, like, are, like, really…”
Justine Moore: Creators choose Fal and Replicate over Google's $125 Veo plan
“What we see is, like, a ton of creators are just going to the model enablement layer, whether it's a consumer-facing interface, like CREA, or more of a developer-facing interface, like a FALL or a Replicate, where you can generate Like, videos on a one-off bas…”
Justine Moore: Foundation models are eating away ComfyUI use cases
“I would agree that more of the Comfy UI use cases are starting to be eaten away by kind of the core foundation model companies.”
Justine Moore: Viral AI video content now originates on TikTok and Instagram
“Now, especially after VO three and, like,
Like Minimax too, and like with the animals diving and that sort of thing, we've actually seen like a flip where most of the viral content is now originating on the true consumer, like everyday consumer platforms, whic…”