AA Text-to-Video
- What does it measure?
- A ranking of which AI makes the better video from a written scene description. Natural motion, faithfulness to the description and image quality all feed into the viewer’s choice.
- Who evaluates it, and how?
- In the Video Arena of independent evaluator Artificial Analysis, two anonymous videos made from the same description are shown side by side, and the user’s choice is scored Elo-style.
- Reading the score
- This is a head-to-head record, not an exam score. There is no maximum; a 100-point gap means the leader is preferred about 64% of the time. When an AI has several quality settings, the setting with the most samples is shown, and AIs with few samples can swing widely. The scale differs from Arena Intelligence, so do not compare the numbers directly. Values in this table are currently about 1,200–1,300. Video length, generation time and cost are not reflected.
Rank
Top score1235
Models tested6
Scoring basisHigher is better
Last updated2026.11.20
Model release date
Which AI makes the images people prefer most?Arena T2I
GPT Image 2.5 Sunburst🥇
GPT Image 2.5 Sunburst🥇Arena T2I
GPT Image 2.5 Sunburst🥇Which AI edits images the way people prefer most?Arena Image Edit
GPT Image 2.5 Sunburst🥇
GPT Image 2.5 Sunburst🥇Arena Image Edit
GPT Image 2.5 Sunburst🥇Which AI do people prefer most for turning photos into videos?Arena Image-to-Video
Gemini Omni 1.2 Flash🥇
Gemini Omni 1.2 Flash🥇Arena Image-to-Video
Gemini Omni 1.2 Flash🥇
Which AI makes the videos people prefer most?
AA Text-to-Video
- 🥇SpaceXSIGrok Imagine Video1235
- 🥈AlibabaWan 2.71210
- 🥉MiniMaxH31123
- #4AlibabaHappyHorse 1.01032
- #5GoogleVeo 3.1961
- #6GoogleVeo 3.1 Lite942
Half of models ≤ 1078Measured by: User preference votes
4 of the 4 AIs released in the last 60 days have no score here yet · Avatar 6 Turbo, Seedance 2.5 Flash, Avatar 6 and 1 more
It’s a preference vote, not a check that the request was followed. A result that looks or sounds nicer can win even if it strays.
Which requests got made depends on who voted — they may differ from your use.
Scores labelled 'Self-reported' were published by the AI maker itself, not measured by a third party.