AI Filmmaking Is Moving Fast – Here’s What’s New

0:00 There's a brand new AI video tool that gives you a ton of camera control. Google is bringing AI video to the biggest screen in the world. And a new tool allows you to type in a prompt and get a TV show in return. But is that show any good? Let's find out in this week's episode of AI Film News. I want to kick things off by giving a special shout out to everyone who attended the second annual AI filmmaking party at NAB 2025. We had an amazing time watching robots battle it out in the arena and also AI filmmakers battle it out to create their own trailers in just a few hours. It was really fun to meet the entire community that flew in for the

0:47 event. We had an amazing time and can't wait for next year. Okay, let's get to the news. So, the first big news that I think you should know about is a brand new AI video tool called Higsfield AI. The thing that makes Higsfield extra unique is the fact that it allows you to have a ton of camera control over the outputed generations. There's a pretty cool example from Uncanny Harry, who is an artist over at Promise, of a few different action scenes, and you can see that there's some pretty dynamic camera movements. Dave Clark put together this shot here of this character. after basically getting up and walking away. He created this in conjunction with a Midjourney 7 image and the Higsfield video tool. But instead of me just showing you awesome examples from the community, let me show you how to use

1:33 the tool. To get started, all you have to do is go to the Higsfield homepage. You'll see they have a big list of a lot of the camera movements that you can utilize inside of the tool right from the homepage. And this includes classic camera movements like the crash zoom in, the arc, and of course tentacles. So to get started with the tool, just go to the create button in the top left. And I should note that you have the ability to generate images, and you can take images and turn them into video. So let's go ahead and test out the image generation first. So, we'll go to text to image and we'll say a cinematic liveaction shot of a glamorous woman walking down the street in Victorian England. And after a few seconds, it generated this image here. And I got to say, I'm really

2:20 unimpressed. While I think the composition is kind of nice and the lighting, there's so much artifacting that I really don't think the result is that compelling. And just a quick comparison, we did the exact same prompt inside of MidJourney and got this image, which looks incredibly realistic. There's tons of information in the background, just high-res details. So, I definitely think that MidJourney 7 is a far superior tool for creating images for your video work. Now, that being said, image generation is not why people are excited about Higsfield. It's all about video generation. So, let me show you how to use that tool. To create a video using Higsfield, just go to image to video and then there's a spot to upload your image. And for our example, we're going to use this image of this

3:05 cat just hanging out on this couch here. And we'll go ahead and upload that. And you'll see there are all of these motion controls that you can now select for the actual camera movement of your overall scene. So, for this one, we'll do a crash zoom in. And for our prompt, we'll say crash zoom in on cat's face. Keep it super basic and go ahead and click generate. And after a few minutes, we have this video here of this cat. And as you can see, the crash zoom in on the cat's face actually looks pretty good. And the camera movement is very compelling. It actually makes me think that they took keyframed or motiontracked real life camera movements and baked them into the model to give you just more realistic motion. specifically in the camera movement. But

3:51 that's just one example. Let's see a few of the other camera movements. So, we have this image here, a closeup of this woman's eyes, and let's select the super dolly out effect and go ahead and render that. And it gives us this result, which I guess looks pretty good. It does seem like it gets pretty soft. And I'm not entirely sure about the frame rate there. It does seem a little skippy. and the uh just overall fidelity and resolution is not the best. So, I definitely think you'd want to throw this through Project Starlight from Topaz Video AI just to get maximum quality. Next up, we have this image of this car and we wanted to animate it with the handheld camera shake effect and it gave us this here and it looks okay. I don't think those wheels are

4:38 moving so uh somehow it's just like magically levitating across the ground there, but the uh overall camera movement does look pretty good. We have this image of this Roman landscape. And we want to do a superlapse effect. And here is the result. And the crowd looks just absolutely terrible and weird. It all morphs in a very strange way. So, a really bad generation from that one. We wanted to do a bullet time effect. So, we have the shot here of a dragon and like this guy looking at the camera. And it slowly spins around. It's in slow-mo. Looks okay. Hey, I don't think there's a lot of life in that dragon, so not the best result there. But, you know, the camera movement is fairly dynamic. We have this shot here of this woman, and

5:24 we selected the dirty lens preset, and it created this shot, which we didn't even prompt for her to like spit out water or anything like that. I guess it just like added it in. So, uh I have no idea what happened there. Maybe she just like heard one of the funniest jokes of all time. Uh, but a really bad result from that one. And then finally, we wanted to do the fisheye effect with this image. Here was the result from Hicksfield. Looks pretty good. It looks fairly realistic. That woman just is kind of looking around. Fisheye camera handheld from below. Looks pretty good. We wanted to compare that to the same generation inside of Runway. So, we prompted it with a fisheye camera lens handheld. And inside of Runway, it

6:10 generated this result here, which I think is a bit more realistic. I like that it's actually liveaction speed, which just kind of helps to sell the overall effect. Obviously, there's some weird warping. Some of the cars, there's not a lot of continuity there, but overall, I think the results that we get from Runway are better. So, you can check out the pricing for Hicksfield over on their website. They have a free trial, which gives you free generations. It's not great, but they do have basic plans starting for as low as $9 a month. Always make sure you select that monthly plan. Don't subscribe to an annual plan from any AI tool because next month a different tool could be better and then you're stuck with an annual subscription to something that is obsolete. So, it seems like with that $9 a month plan, it

6:56 gives you about 15 video generations, which makes Hicksfield not necessarily one of the cheaper AI video options on the market. I'm not going to lie to you, I think outside of just the core camera movements, the overall generations that you get from Higsfield aren't my favorite. I think when you compare the generations from Luma or Runway, Google Vio or Cling, the Higsfield AI generations simply aren't up to par. That being said, I could see projects where you need specific camera movements, whether they are music videos or perhaps some advertising projects. In those instances, you may want to use Hicksfield. I just recommend testing it out for yourself to see which generations you prefer. In video

7:41 restyling news, the team at Korea AI came out with a new feature that allows you to restyle video footage. Essentially, this is similar to old tools like Runway Gen 1 where you can upload a video clip and it will just completely change the style to something else. Now, the thing that makes this tool specifically unique is it allows you to control the outputed generations through four different settings. The first is depth where it will create a depth map for your video and then restyle the depth. The second is the edges where it will go in and try to find the edges inside of your video and generate from there. The third is pose where it basically understands the pose and the dynamics of your character. And the fourth is video where it really will

8:29 just analyze the video itself and try to output a video that is in the same compositional and color style. And the cool thing is that they have quite a few presets over on their website that you can play around with. Let me show you how to use it. So to use this new tool, all you have to do is go to the Korea website and click on the video restyle button right there on the homepage and go ahead and upload your video. For our example, I'm going to use this video clip of this woman talking to someone else. It's just a very standard overthe-shoulder shot. And go ahead and upload. Now, after you uploaded your video, there are a few different settings that you can select here. So, you have depth, edges, pose, and color. It really just depends on the type of project that you're working on. I think

9:14 for our example, because I want it to be a little closer to the original video, I'm going to go ahead and select color. And we'll go over here to our prompt box. And instead of me typing in a prompt, I'm going to select a style here. So, you can just go to the style button and you'll see all of these different style presets, which is pretty cool. And if you go to the featured tab at the very top, there are a ton of different preset styles that other artists have uploaded. I think are just really cool. Let's go ahead and select this 10 dreams aesthetic, this like 1950s tin toy. And for our prompt, we'll say a woman talks. Basically, that's the action in the video. And go ahead and click generate. So, it'll take a few minutes to generate your results. Here's

10:00 the first one. You see the woman's just kind of talking there. Seems like she's kind of made out of tin. Not exactly. And then we have the second video clip in a different style, kind of just more painted, which I think is pretty cool. And then finally, we have the woman stylized to look like Vincent Van Go in that kind of chasing Vincent style if you watch that feature film. And uh yeah, it's kind of horrifying, but also kind of awesome at the same time. Now you will notice in these generations that character continuity is very tricky and I think that's a huge limitation whenever you are working with these style transfer tools. Now I do want to show you a different generation that we ran inside of Korea. So I have this video clip. It's basically just a time lapse of a city here and I did want to change it into a few different styles. The first one is anime and it generated

10:47 this one here which actually is kind of awesome. The second is cyberpunk and it generated this one. Not bad. I'm not sure what kabi sat means. I don't know what language that is, but uh pretty interesting result there. And then finally, we have a black and white storyboard version, which I actually think is pretty cool. I think the idea of animating your storyboards and just kind of giving yourself a rough animatic before you go to production is just kind of an awesome use case for AI technology. Now, obviously the core elements inside of this scene changed, but it did retain the overall composition. In other AI video generation news, the team at Runway came out with Runway Gen 4 Turbo. So, couple weeks ago, they came out with Runway Gen

11:33 4. We think it's one of the most powerful AI video generators that's ever come out. This update is a version of Runway Gen 4 that is not quite as good, but much faster when it comes to generating the results. You can generate a 10-second clip in only 30 seconds, and the results are actually pretty decent. There was a really interesting research project that came out this last week that really highlights one of the futures of entertainment. Now, I am talking about the idea of typing in a prompt and watching something in return. Up to this point, there have been a few tools that have ventured into prompt to entertainment, but the results really haven't been that great. Well, this research team is looking to change that. This brand new research project is

12:19 called 1 minute video generation with test time training. And essentially, the researchers used Tom and Jerry training data to allow it to create custom episodes of Tom and Jerry using their platform. Essentially, they were able to combine some tools together that can allow you to type in a prompt and get an episode of Tom and Jerry in return. Now, are these episodes great? I don't think so. But let's take a look at a few of the examples. So, we have this establishing shot here. It's basically taking place in a business building. Tom is walking into the office. You see, it intelligently understands what the character would be doing. I'm guessing this is like an elevator or a hallway. puts down his briefcase. He's working at

13:05 a computer, which I think is way after the original Tom and Jerry aired. And then you can see Jerry's here and he's chewing on the mouse cable. It intelligently understands that the screen should basically go to static. And then he sees Jerry down here. He gets mad. Jerry runs away and like disappears in a weird way. They chase one another and then Jerry runs into a hole. Tom hits his head and basically is like confused. And then he's always late for a meeting. He has to go to a meeting and his boss is there and he doesn't have the presentation. His boss is mad. He gets in trouble and naturally Jerry is laughing at him from his home. Now, there's a lot you can critique from that video. The continuity of the characters, there's weird warping. A lot of the

13:52 movements don't make a lot of sense. But really think about what we just watched. AI understood how to create an establishing shot. It understood setting up the story and where the character lives. It introduced both the conflict and the resolution all in a single prompt. And all of that being said, I don't want to understate how important this innovation is. This is one of the first AI video examples that really understands character development and story continuity as opposed to just individual prompts that are prompted one after another. Now, of course, this is a terrible end result and it would never resonate with a mass audience, but this is the worst version of this that will ever exist. So, I just want to reiterate that storytelling skills are the most

14:39 important skill you can have as a creative AI artist. That's why in our AI courses at Curious Refuge, we not only teach the technical side of AI storytelling, but also the storytelling and artistic side of developing taste, characters, and overall stories that resonate with an audience. It's really cool. We partnered up with Emmy-winning screenwriter Bernie Sue to introduce storytelling training in our advanced AI filmmaking course. If you want to learn more about that course or any of the other training we have at Curious Refuge, be sure to click the link below this video. AI video is making its way to the largest screen in the world. And I am of course talking about the sphere in Las Vegas. The news is basically the

15:25 team at Google has worked in partnership with the sphere to utilize AI tools to bring the Wizard of Oz to their ridiculously large spherical screen. Essentially, the film had to be upscaled to 16K. And so, they created their own proprietary version of Topaz Video to upres and get maximum quality from the video clips. They also had to change the aspect ratio from 4x3, which is the original Wizard of Oz, to something much wider so that it could fit on the overall screen. And for that technological problem, they were able to use Google's AI video tools to do outpainting to create more pixels on the screen. This film is set to premiere on August 28th at the Sphere. You better

16:11 believe if we're back in Las Vegas in the near future, we are going to check out the film. The team at Bite Dance has announced a brand new AI video tool that allows you to take a video with movement and character acting and basically take an image and animate the image using the source video. There are a few different examples over on their website. You see we have this woman singing here and it was converted over to a different style using an image. And you can see using the tool, there's just a lot more fidelity versus other AI tools that do a similar result like runway act one. We also have this video clip here of this woman talking to the camera. And you can see the style transfer works on animated characters. And then also this second option, which I don't think is very realistic, but it did a pretty good job.

16:59 We have this example here of this woman listening to her headphones. You can see there's just a lot of object permanence and continuity that is really impressive compared to some of the other AI video tools that do a similar result. In AI competition news, there are a few big competitions that I think you should know about. The first is the latest installment of Runway's Gen 48 competition. Now, if you're not familiar with this competition, essentially it's a 48 hour film race to create the best film possible. This edition will basically take teams of one to three filmmakers and have them create a 1 to four minute film. Each team will receive 300,000 runway credits and the winner will get $5,000 and a million runway

17:46 credits that they can use on their projects. And all winning filmmakers will get access to their creative partners program. And speaking of runway, they are hosting their third annual AI film festival in both New York City and Los Angeles in just a couple of months. The New York City version will happen on June 5th and Los Angeles will happen on June 12th. I think this is one of the best events in the AI filmmaking community. So, if you have a chance, I highly recommend making it out to the event. And the last bit of AI competition news that I'm really excited to announce is the second annual Reply AI Film Festival happening at the same time as the Venice International Film Festival. So, if you're not familiar with this event, essentially the finalists get flown from around the

18:32 world to Venice to the Venice International Film Festival to see their project play. It's one of my favorite AI film events, and Shelby and I will be judges for this year's competition. And it's also cool because there's a prize pool of over €30,000 euros available. You'll find a link below this video to learn more about the genres and how you can submit your films. There's also a bit of news that I think is kind of interesting that you probably have heard about. If you've been following X or LinkedIn and you're in the AI space, you've probably heard of the term vibe coding. Essentially, vibe coding is using prompts to create code to create whatever you want. Typically, it's a video game, and there are a few really interesting examples online of video games that people created using just

19:19 prompts alone. We have this Settlers of Katan real-time strategy game, which is kind of awesome. There's an 80s retro futuristic endless driving game where you collect these coins, and this Halo inspired game that was created with Gro 3, Claude, and 3JS. Now, I'm not going to get into the weeds and show you how to create one of these games yourself inside of this video, but I should note that if you want to get into it for yourself, people are utilizing Grock, Windsurf, Cursor, and Claude 3.7 Sonnet. But if you want us to make a tutorial, be sure to leave a comment below. Obviously, we like to talk about AI film, but if you'd like us to venture a bit into some coding stuff here, I would be happy to show you how it's done. In AI event news, there are a ton of events

20:06 and meetups popping up around the world. We have a Curious Refuge Barcelona meetup happening on April 11th. There's a New York City Curious Refuge meetup happening on April 13th. Runway is hosting a Montreal meetup on April 17th. And we have a Curious Refuge meetup in Paris on May 7th. And that brings us to our AI films of the week. The first film that I want to shout out is called Katana by Matthew Samson. It was created using Higsfield and Clling. And they worked around the lip-sync problem through utilizing robotic voices. It's kind of in this eastern Bladeunner aesthetic, which is pretty cool. And I think the curation of the individual shots is fairly impressive. So, good job on that project, Matthew. The next film

20:53 that I want to shout out is called Merks by Dave Clark. While it's not necessarily a deep story, it's a pretty interesting example of utilizing both MidJourney 7 and Luma Ray 2 to create some action movie scenes. And a lot of the shots seem realistic, especially with some of the camera shake. It's almost indistinguishable from a liveaction camera. And I think like always, it's interesting to see Dave's vision in the space because it seems like it's something like Burning Man in space mixed with an action film, which I would 100% watch. And the final film I want to shout out is called Lonely Little Frame. I'm not entirely sure who the artist was behind this project, but it was created by the Runway team to basically highlight their brand new Runway Gen 4 model. I think this is a

21:40 really good example of when you utilize both really good sound design and color grading to help cover up some of the technical limitations when you're working with AI video. Obviously, there are some issues with continuity inside of the film, but generally speaking, I think that there's a lot of character that rises to the surface because they were able to utilize sound and color to help tell a larger story. And then finally, I came across some interesting news this last week where the team at Universal Orlando is basically slapping on costumes to Boston Dynamics Robots to create characters for their park. For example, we have this video here of Toothless from How to Train Your Dragon, and you can tell it's like one of those

22:26 like doglike Boston Dynamic Robots, but they did a good job with putting an animatronic head on it, and it really seems like it has a ton of character to it. I think utilizing technology like this to increase guest and audience experiences is just super fun. But let's hope that these robots behave or we might be looking into an extended universe for Five Nights at Freddy's. Okay, thank you so much for watching this week's episode of AI Film News. Of course, if you want to get AIL news sent directly to your email inbox each week, you can subscribe for free by visiting curiousrefuge.com. And if you want to get the latest AI tutorials and news directly here on YouTube, be sure to like and subscribe.

23:11 Thank you so much for watching this week's episode. We'll see you next time.