First came the rise of AI chatbots, then image generators blew up. Now, tech companies are rushing to release AI video generators.
During the past year, nearly every major tech company has announced some kind of AI video model they’ve been cooking. Each company has its own timeline, which can make it hard to keep up with who’s done what. To save you from searching, I’ve run down every major AI video program and done some early testing on the ones available now.
I’ve spent a lot of time with these companies’ AI image generators, which are one of the building blocks in the development of AI video models. In my early testing, I’ve seen some of my favorite image generator features pop up in the video models, while others are noticeably absent.
AI videos are a huge leap forward in a company’s AI creative offerings, and they’re something worth keeping an eye on. This is especially true as the advancement of the tech comes at a time when the legality, ethics and other concerns continue to surround AI creative offerings.
This is everything you need to know about the major AI video generators.
What are AI video generators?
AI video generators are one of the latest ways tech companies are using generative AI technology. These text-to-video and image-to-video programs let you create short, audio-less clips. Most of the major tech companies have announced their own version, with some available for you to use now and some upcoming programs to keep an eye on.
Two AI video generators you can use now are Sora by OpenAI and Adobe Firefly. Both are paid programs that produce decent results and let you customize your shot with control panels. Runway, an AI start-up that co-created the Stable Diffusion image generator, is another AI video option with freemium plans.
Google’s AI video project Veo is currently in private beta, but you can take it on an early test run using a new feature on YouTube Shorts’ to generate an AI video background. Meta’s MovieGen is technically just a research concept, but it boasts the unique ability to synchronize audio with its AI video clips. We’ll have to wait and see whether Veo and MovieGen are released this year and how they stack up.
Sora joined the ChatGPT family at the end of 2024. It’s a pretty user-friendly program, but it doesn’t have the same conversational UI as Dall-E 3 — you can’t “chat” with Sora to make follow-up revisions. Instead, it’s more reminiscent of traditional AI creative services. You have a panel to customize your video’s dimensions, length and stylistic feel. You can enter a prompt or upload an image for Sora to animate, and you can use a few editing options to perfect your video from there. Sora videos also come automatically watermarked, designating their AI origins.
Sora is only available to paying ChatGPT users. If you’re a ChatGPT Plus user ($20 per month) you’ll get 50 priority generation credits per month, with videos up to 5 seconds long at 720p. Upgrading to the Pro tier ($200 per month) gets higher monthly credits, including 500 videos created with priority/fast generation and unlimited videos with relaxed generation. Pro subscribers can also create higher-resolution videos at a max of 1080p, extend the duration of their videos up to 20 seconds and have the option to download videos without the watermark.
OpenAI’s privacy policy states that it may train on your content unless you opt out. To do that in Sora, go to Settings > General, and turn off Improve model for everyone. You can also exclude your videos from public explore feeds in settings.
Firefly’s standalone AI video generator came out of a private beta in February, and it’s available for you to use now. If you’re familiar with Firefly’s AI image tools, the video generator set up will feel familiar. The left-side panel lets you customize your clip, and it has the added benefit of letting you select the kind of motion you want (zoom in and out, move right and left, etc). You can also select the camera angle you want, like if you wanted to mimic drone footage with an aerial view.
Some Creative Cloud plans include Firefly access, including if you’re paying for a single program or all of the Adobe apps. You can check and compare options here. If you don’t have an existing Adobe plan, you can try the Firefly standard plan ($10 per month) to create up to 20 videos a month. If you need more generation credits, the Pro plan ($30 per month) gets you up to 70 videos a month. Both Firefly plans come with unlimited AI image generation. Your Firefly videos will be 5 seconds long, at 1080p with no audio.
Adobe says that videos created with Firefly are commercially safe, and its AI policy states it will not train on your content. Firefly videos don’t have a visible watermark, but its content credentials are automatically attached to your work. Firefly models are trained on licensed and public domain content.
Google’s DeepMind research lab has been working on the company’s AI video model, Veo, for a while now. Google announced the project at its developers conference last year, showing off some of the possible use-cases in a clip starring Donald Glover. In December 2024, Google said it was moving Veo 2 into a private preview for approved users on Vertex AI, Google Cloud’s machine-learning platform. You can sign up for the waitlist here.
From what we’ve seen so far, Veo will be able to handle text-to-video and image-to-video prompts. We don’t know how long the clips will be or whether they’ll have audio. Google says videos will be created at a max of 1080p and will have Google’s SynthID watermark embedded. Google Cloud’s general AI policy states it does not train on customer data without permission. Google did not respond to requests for further information on Veo’s development and release.
If you don’t want to wait, you can get a taste of what Veo 2 can do on YouTube. Last month the company announced that YouTube Shorts creators could now create AI-generated backgrounds for their videos thanks to a new feature powered by Veo. In my initial testing, the creation process was smooth, but I’m reserving judgment until the program is fully formed and available for everyone.
AI enthusiasts might recognize Runway as the start-up that co-created the popular AI image generator Stable Diffusion. You might also recognize Runway from a landmark deal it made with a major film studio last fall. Lionsgate agreed to open up its catalog — thousands of hours of movies like The Hunger Games and John Wick and TV shows like Mad Men — to be used to create custom AI models for the studio to use. Even though Runway’s not technically one of the big tech companies, its AI video tools are worth mentioning.
During my brief testing of the service, I was impressed with the prompt-building tools and the general ease of finding my way around. I’ve also used the service before as part of Canva’s Magic Media app, which is convenient if you’re a Canva lover. You can use Runway for free on its web app, with 125 monthly credits — you’ll use about 20 credits with each generation, so it’s a pretty low limit. Upgrading ($15 per month or $144 annually) gets you 625 monthly credits, access to newer models and the ability to upscale videos to 4K and download without watermarks.
Runway’s terms of service says it can train its AI on your prompts and the resulting videos but doesn’t retain ownership over them. Its privacy policy also states that Runway may disclose your information to affiliates, business and marketing partners. The videos you make are automatically private.
Last but not least is Meta’s AI video model, Movie Gen. The project is only a research concept right now — it’s not publicly available, with no word on when it may be. Perhaps like with Google and YouTube, we’ll see some AI-powered features pop up first on its social platforms, Instagram and Facebook. (We already have a number of other AI features eating up space on our feeds.)
Thanks to a research paper Meta published, we know Movie Gen videos could be 1080p HD and up to 16 seconds long at 16 frames per second. The most notable thing going for Movie Gen is the possibility of synchronized audio. Meta said that Movie Gen could also be used to create sound effects, ambient noise and instrumental music up to 45 seconds long. There’s always a chance this feature doesn’t make it to the final cut, but it would give Meta a huge edge over all its competitors.
Meta’s AI models for its chatbot and image generator are trained on publicly available content from Facebook and Instagram (European users can opt out) and licensed information.
Add comment