Google makes video generator "Gemini Omni 1.2 Flash" generally available, extending clips to 15 seconds and cutting 720p to $0.07 a second

Clips extended to 15 seconds, prices cut by about 30%

Google on Nov. 5 (U.S. time) announced "Gemini Omni 1.2 Flash," a new version of its video generation AI, and made it generally available in the Gemini API for developers and in Google Cloud. The video that can be created in a single generation was extended from 10 seconds in the previous 1.1 to 15 seconds. The price for 720p video was cut from about $0.10 to $0.07 per second. Google cites speed, low cost and video length as features of the new version.

Detection bot posts model name; context and maximum output unconfirmed

@gptbotio, an X account that detects and posts about new models, posted on Nov. 5 (1 a.m. Nov. 6 Japan time) that "gemini-omni-1.2-flash" had appeared in the development environment "Google AI Studio." According to the post, the context (the amount of information that can be read in at once) is 131,072 tokens and the maximum output is 65,536 tokens.

The model name matched Google's official materials. The 2 figures, however, do not appear in Google's announcement or its official model list and remain only the bot's observations. ORIENS has not been able to confirm them. Whether the post came before or after the announcement is also unknown, as Google's blog does not give a publication time.

Up to 6 reference images, 30 seconds in total with extensions

According to Google, 1.2 lets users specify length in 1-second increments between 3 and 15 seconds. It can use up to 6 reference images and has improved ability to keep the appearance of characters and products consistent. It can also take motion video or audio of up to 3 seconds as a cue, and conversational editing to fix backgrounds or lighting can be applied to an entire 15-second video.

The extension feature, which adds a continuation to a created video, adds 3 to 10 seconds at a time, up to 30 seconds in total. For 1.1, the total was 40 seconds, according to a DataNorth report. Google says it is continuing to improve fast sports movements, sudden script changes and the depiction of complex physics.

Video output price cut from $17.50 to $12.00

Per 1 million tokens, the price for video output was cut from $17.50 to $12.00. Input stays at $1.50 and text output at $9.00. Per-second reference prices are $0.023 for 360p, $0.07 for 720p, $0.104 for 1080p and $0.209 for 4K, making a 15-second 720p video about $1.04.

It cannot be used on the API's free tier. On paid tiers, limits such as the number of requests per minute (rate limits) were doubled. From Dec. 3, API requests to 1.1 will be automatically routed to 1.2 and billed at 1.2's prices.

720p list price below other companies' video models

Comparing 720p list prices per second, 1.2's $0.07 is below the $0.30 of China's ByteDance's "Seedance 2.5," the $0.17 of Germany's Black Forest Labs' "FLUX 3 Max" and the $0.157 of Chinese company Kuaishou's "Kling 4.0." However, FLUX 3 Max has cut 720p to $0.0425 through Nov. 7. At 1080p, 1.2's $0.104 is higher than the roughly $0.096 for Vidu Q4 that AIReiter confirmed in October.

Some models, such as Seedance 2.5, also charge for the length of video used as a reference.

No third-party performance evaluation yet; 1.1 ranks 10th

Google's announcement gives no benchmark figures, and the comparison with 1.1 is limited to examples of footage. As of Nov. 6, 1.2 was also not on the leaderboard of third-party evaluator Artificial Analysis. In its text-to-video evaluation, the previous 1.1 ranked 10th with an Elo (a score calculated from win rates in head-to-head matchups) of 1,129. The leader is China's Alibaba's "Wan 3.0" at 1,172.

Gemini app rolling out starting with Ultra subscribers

In apps, 15-second videos can be created in Google Vids at no extra charge. In the Gemini app, it is rolling out gradually to Google AI Ultra subscribers. It can also be used in "Flow," for video production, and "Google Ads," for advertising. Created videos carry "SynthID," an invisible watermark indicating AI generation.

15 seconds and price cut match earlier information; "1.5" still unannounced

Fragments of this version had circulated on X before the announcement. The X account @lyraxana posted on Oct. 18 (Oct. 19 Japan time) that the next version, "Gemini Omni 1.5," would extend clips to 15 seconds. The 15 seconds proved correct, but the version number was 1.2. Google has not announced 1.5, and whether it is being developed as a separate version is unknown.

On Nov. 5, the same day as the announcement, @mrfanduu posted an image of a Google Cloud screen and a claim that it was about 30% cheaper than 1.1. OrcaRouter's blog reported this. At the time OrcaRouter checked, there was no mention of 1.2 on Google's pricing page or elsewhere, it said. When ORIENS checked on Nov. 6, it was listed on the pricing page and in the model list.

Unconfirmed information about Google's models continues to surface. For the next-generation lightweight model, there are reports that "Gemini 4 Flash" is being tested in AI Studio and elsewhere, and rumors that it will launch in a form that routes each question to a model, but Google has not announced it.