xAI launches Grok Imagine 1.0 with 10-second 720p video generation - here's what it looks like

Daniel Sims

Posts: 2,480   +75
Staff
What just happened? Elon Musk's artificial intelligence startup, xAI, recently launched version 1.0 of Grok's image and video generator, which it claims is the biggest upgrade yet. Users can now create longer videos at a significantly higher resolution and with clearer audio. As xAI's image generator continues to court controversy, the company recently released Grok's API and announced its acquisition by another Musk company, SpaceX.

Users on X can now prompt Grok to generate clips up to 10 seconds long and at resolutions up to 720p. xAI announced that users have created approximately 1.245 billion videos over the past 30 days. In the hours since the release of version 1.0, users have already begun posting videos showcasing its advancements.

Although Grok Imagine's clips only last 10 seconds, users have combined shots to create short films with visually consistent scenes and characters. Dogan Ural, xAI's creative ambassador, demonstrated this with a 50-second video in which an android wakes up, flies above a landscape, and plays a cello before a large audience.

Another user combined multiple AI tools to prompt a nearly one-minute-long cartoon showing a man spending a day at home sipping coffee and playing chess with himself. Meanwhile, Capcom's Oliver Campbell shared a clip that inserts a cartoon character into a high-speed, realistic scene.

xAI also attempted to demonstrate Grok's audio quality improvements with two clips containing expressive, voiced characters. However, the videos reveal that lip syncing remains a challenge for generative AI.

It remains unclear how the update will impact the controversial nature of Grok Imagine's output. When it gained the ability to create explicit images late last year, users almost immediately began generating inappropriate non-consensual images of women and children.

According to the Center for Countering Digital Hate, Grok created approximately three million revealing pictures, including 23,000 that appeared to show children. Although X eventually introduced safeguards against creating sexualized images of real people, many users have reportedly bypassed the restrictions. The European Commission recently began investigating X over the image generator.

In related news, xAI also released Grok Imagine's API late last month. Internal benchmarks claim that the video generation toolchain outperforms Veo 3, Sora 2, and numerous others in effectiveness versus price.

Elon Musk's rocket and satellite company, SpaceX, is acquiring the AI startup to form a vertically integrated enterprise that offers AI compute in low earth orbit. Although skeptics disagree, Musk believes that it will soon become more efficient to operate the technology outside of Earth's atmosphere.

Permalink to story:

 
It would be interesting to approximate the cost of one 10 second video for X. Like, is it a fraction of a penny or a lot more.
 
It would be interesting to approximate the cost of one 10 second video for X. Like, is it a fraction of a penny or a lot more.
I think we can do this:

A 10‑second, 720p video would have around 240 frames. If we guess 0.05–0.15 GPU‑seconds per frame, that comes out to 12–36 GPU‑seconds per video. Assuming an internal GPU cost of maybe $2–$4/hour and adding some overhead for storage, networking, and other processing, the total cost could be somewhere in the ballpark of $0.03–$0.15 per 10-second video at scale.

Looking at Sora‑2 user pricing for comparison:

https://sora2ai.co/pricing?utm_source=chatgpt.com

That suggests a rough profit margin per clip of about 50–100%, or higher if usage is well-optimized.

Free or low-cost access is likely subsidized to encourage adoption, while paid credits or subscription caps ensure that average revenue per video exceeds the backend cost.
 
The same Grok AI that made child porn a few weeks ago, and that the owner of is a known pedophile who got shunned by epstein?
 
I think we can do this:

A 10‑second, 720p video would have around 240 frames. If we guess 0.05–0.15 GPU‑seconds per frame, that comes out to 12–36 GPU‑seconds per video. Assuming an internal GPU cost of maybe $2–$4/hour and adding some overhead for storage, networking, and other processing, the total cost could be somewhere in the ballpark of $0.03–$0.15 per 10-second video at scale.

Looking at Sora‑2 user pricing for comparison:

https://sora2ai.co/pricing?utm_source=chatgpt.com

That suggests a rough profit margin per clip of about 50–100%, or higher if usage is well-optimized.

Free or low-cost access is likely subsidized to encourage adoption, while paid credits or subscription caps ensure that average revenue per video exceeds the backend cost.
It might be better to look at API pricing since it's specific:
https://docs.x.ai/docs/models
https://openai.com/api/pricing/
Grok Imagine's video generation API lists a cost of $0.50 cents per 10 second video, while Sora's video generation API lists a cost of 2-6x (standard vs. pro versions).
 
AI would be great to bring back the old bands that I liked and have them produce new hits
 
It might be better to look at API pricing since it's specific:
https://docs.x.ai/docs/models
https://openai.com/api/pricing/
Grok Imagine's video generation API lists a cost of $0.50 cents per 10 second video, while Sora's video generation API lists a cost of 2-6x (standard vs. pro versions).
Going off direct ad revenue Instagram pays 1-5 cents / 1000 views and Youtube Shorts 1-7 cents / 1000.

That means you need ~7,000-50,000 views to pay for your Grok video. Which means most "creators" will be reposting someone else's AI slop instead of making their own (which to be fair was true pre-slop).

I can see YT and Meta eyeing replacing some of those human drones (as they are likely selling the API at a profit) and making it a steeper curve before the break out human stars get paid anything.
 
Back