Did Minimax Solve Character Consistency?

The Chinese AI Video company's new 'Subject Reference' Feature is VERY good

Author

AI 4Humans
21 Jan

Today on AI For Humans The Newsletter!
Consistent characters From Minimax
New open source Chinese models are making AI political
And Apple decides to turn it’s AI off & back on again
Plus, our can’t-miss AI feature of the week!

Welcome back to the AI For Humans Newsletter!

It’s a busy week in the AI space and we’ll dive into what DeepSeek’s new R1 model means for the future of open-source AI and OUR future at large… but it’s another Chinese AI company that got our attention this week:

Hailuo AI introduced their ‘Subject Reference (S2V-01) Model’ to the Minimax AI video platform this week which promises and delivers on the following:

  • Single image to AI video model consistency

  • Much lower cost (following a trend from other Chinese AI models)

  • Faster results than from the previous model

There are a lot of great examples floating around the internet already, and we specifically appreciated Levelsio’s write-up. He’s been working with a number of AI models for the start-ups he’s created (btw, he’s got a great backstory and it’s worth listening to this interview with him) and to see him as excited about this got us even MORE enthused.

I tried it a few times this morning via Replicate and the first results were ok but I can see how this will def start to make it much easier to make full blown AI films vs just smaller edits. A heads up, each Replicate gen of this costs $.50, so it could get expensive fast.

Maybe don’t use a super smiley version of yourself?

Character consistency is truly the holy grail of AI video (well, that and maybe needing fewer generations to get a better result). Can’t wait to dig in further.

See you Thursday!

-Kevin & Gavin

3 Things To Know

An Open Source Model As Good As o1!?
This week’s big headline looks to be the release of Deepseek-R1, an open source model that matches state-of-the-art o1 in most domains, along with 6 more open source models targeting GPT-4o and Claude 3.5 Sonnet. On the day a new leader of the free world is signed in, a Chinese company sends a not-so-subtle shot across the American AI bow. Ironically, Deepseek in many ways is delivering on the open source ethos OpenAI was originally founded on.

Sam Altman Feeds The Tease Machine
After sharing that a new model, o3-mini, would be coming in a couple week’s time Altman told the Twitter AGI mob to slow their roll. He clarified they don’t plan to release AGI next month, but various leakers hype accounts have been teasing unannounced breakthroughs, adding to speculation that OpenAI’s upcoming private briefing with US officials is to unveil a new breakthrough.

Meanwhile, at Apple
Notification summaries have been disabled because Tim Apple says I just can’t trust these new computers (our quote). It’s another AI stumble from the world’s largest company, and comes at a time when models have never been better. Even X/Twitter’s Grok implementation has gotten pretty darn good, using AI to explain trending topics, and add context from the web to a tweet.

Today’s best single shot text-to-image tool is probably Ideogram, and the best indie film stills probably come from Runway Frames, but old reliable Midjourney still has a few tricks up it’s sleeve.

This week we saw a very cool medieval animated short from X user @thefuzzysignal:

They shared in the replies the visuals were consistent scene-to-scene thanks to Midjourney’s ability to maintain a consistent style over multiple prompts.

It’s a straightforward feature to use:

Here we used the prompt 1950s diner, low contrast cinematic, teal and orange colors to generate some ideas.

Once you have something you like, simply click “Use Style” and change up the prompt to produce as many as you need with the same visual aesthetic.

Are you a creative or brand looking to go deeper with AI?
Join our community of collaborative creators on the AI4H Discord
Get exclusive access to all things AI4H on our Patreon
If you’re an org, consider booking Kevin & Gavin for your next event!