Veo, Google DeepMind's video generation model, can model liquids surprisingly well.
Mostly true True in substance, wrong in a detail.
Factual: Statements about the present or the past, checked against the record: did they hold up?
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Judge against evidence available as of July 2025 on how Veo (Veo 2 or Veo 3) renders liquids. True: official demos, independent reviews or physics-plausibility benchmarks show that Veo generally produces realistic liquid behaviour (pouring, splashing, flow), and that this is notable for a video model. Mostly true: liquids are generally rendered well, with some documented failures. Mixed: the evidence is split between realistic results and frequent clear physical errors in liquids. Mostly false: independent evaluations show Veo often fails at liquid physics, with good results limited to curated demos. False: evaluations show Veo models liquids poorly.
Set before the research began.
Evidence (4)
-
E1 Google Veo 2 is one of the best AI video models I've ever seen — here's 5 examples of what it can do Tom's Guide · 17 December 2024
“Weirdly, pouring liquid is something other models struggle with but Veo 2 did it perfectly.”
-
E2 raw AI Wiki (aiwiki.ai) · 22 October 2026
“It wrote that Veo 2 seemed "to have a strong grasp of refraction and tricky liquids, like maple syrup," but that it "can't quite clear the uncanny valley."”
-
E3 Veo Physics Collapse Reference — Fluid Gravity Newtonian Failure AI Video Auditor (AVA) · 1 May 2026
“Water visibly flows upward at 0:02 in the central column of the falls”
-
E4 VideoPhy: Evaluating Physical Commonsense for Video Generation UCLA (Kai-Wei Chang bibliography page), ICLR 2025 paper by Bansal et al. · 1 January 2025
“Specifically, the best performing model, CogVideoX-5B, generates videos that adhere to the caption and physical laws for 39.6% of the instances.”
Reasoning
A Tom's Guide reviewer said Veo 2 handled pouring liquid "perfectly" where other models struggle (E1). Another reported review said Veo 2 had "a strong grasp of refraction and tricky liquids, like maple syrup," while still showing uncanny-valley effects (E2). The only documented liquid failure is a 2026 page describing water flowing upward in a Veo clip (E3), and the quoted VideoPhy result does not name Veo (E4). Overall the evidence supports generally good, notable liquid rendering, with some reported failures.
Cite this claim
Demis Hassabis, “Veo, Google DeepMind's video generation model, can model liquids surprisingly well.” (said 23 July 2025). Verdict: mostly true. Reliability API, claim c_8cbfdd9395d9, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_8cbfdd9395d9
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Demis Hassabis
-
Mostly true
“Performs better than 3.1 Pro on coding & agentic tasks”
Gemini 3.5 Flash performs better than Gemini 3.1 Pro on coding and agentic tasks.
-
Mostly true
“It can take photos, video & audio and build entirely new scenes.”
Gemini Omni can take photos, video and audio and build entirely new scenes.
-
Mostly true
“I didn’t feel like that was proper AI because that system, let’s take Deep Blue, okay, it’s world champion level at chess, at chess, but it can’t do anything else.”
Deep Blue was world-champion level at chess but could not do anything else.
-
Resolved true
“And it uses AI to actually digitally watermark any generated image.”
-
Mostly true
“They still can't really yet go beyond asking a new novel question or a new novel conjecture or coming up with a new hypothesis that has not been thought of before.”
Verdicts are made automatically; how verdicts are made.