Adobe today launched its competitor to OpenAI’s Sora, a text-to-video AI generator that Adobe is positioning as a possible alternative to licensing stock footage. It costs at least $10 per month with the “best value” option at $30 per month, but it’s nowhere near good enough to warrant either price.
Powered by Adobe’s Firefly video model, the generative video feature is currently in beta but is still being monetized. Current Creative Cloud subscribers can test it out with two free generations but after that, Adobe requires a Firefly subscription.
The $10 per month option grants 2,000 generative credits, which is enough for 20 generated videos, while the $30 per month option grants 7,000 credits, which is enough for 70 video generations. Videos are always five seconds long (no longer) and max out at 1080p resolution. It is important to note that these are “introductory” prices that will change on March 15. Adobe has not been clear what it will change to.
When a customer attempts to subscribe, a notice informs them that the pricing “will automatically renew monthly until you cancel. Price subject to change at renewal.” Change to what? Adobe does not say. This language only appears during the purchase process and is not explained in Adobe’s public FAQ.
Additionally, Adobe doesn’t offer other tiers above these, such as an unlimited tier, so the only option for those who exhaust their pool of credits is just to buy another subscription to Firefly and stack them on top of each other. At least the annual subscription option does provide a discount over the month-to-month option, which isn’t always the case for Adobe software.
Keeping that pricing structure in the back of mind, let’s examine what Adobe believes this tool is going to be useful for.
“Today, we are expanding this vision with the launch of the Adobe Firefly Video Model, the only generative AI model that is IP-friendly and commercially safe, and the expansion of creative control in the Firefly web app which empowers anyone, at any skill level, to explore, ideate and bring their creative vision to life,” Adobe says.
“This isn’t just another video tool —– we’ve revolutionized the creative workflow and watching our community generate custom b-roll that exactly matches their vision, crafting dynamic visual effects, and designing motion elements, all while maintaining complete creative control. And there’s more — whether you need to bridge content gaps in your timeline or enhance existing footage with new elements, Firefly Video Model empowers you to achieve your creative vision with efficiency and precision.”
So there are two proposed use cases: one is stock footage that is more specifically tailored to something that might not exist and the other is for what boils down to moving storyboards.
To be frank, the idea that Firefly is good enough to replace stock footage in any circumstance is laughable. PetaPixel generated four videos with varying prompts to see how it handled them. Actually, we only generated three since the first one required a re-do since we wanted to see a variation of it. That burned both of our test generations on one account.
The first was created using the prompt: “an asian-american camera reviewer on youtube in the snow talking about camera equipment while holding a camera.”
When that produced a female host, we changed the prompt to: “an asian-american male camera reviewer on youtube in the snow talking about camera equipment while holding a camera” to see its attempt at a man.
Using our second Creative Cloud account, we prompted it with two more, the first being: “an aerial view of the san francisco skyline at sunset. the camera moves in a straight line from one side of the city to the other as the sun slowly sets.”
The last was generated with the prompt: “a wolf running through a verdant green field, morning light, with mountains and a forest in the background. the camera follows the wolf as it runs.”
None of these are usable in exchange for stock footage. The first two which use a supposed human are ghastly. In no world would these even remotely pass as real. These are, without a doubt, garbage. Even Coca-Cola’s trash-tier AI-generated slop ad from last year is better, and that was a low bar.
The last two are better, but they still don’t look believable. The San Francisco skyline might pass for usable at first glance for anyone who had not seen it before but compared to reality, the perspective is wrong. The main issue is that the buildings are not the right height in relation to one another (let alone most of the less iconic ones are just incorrect).
Adobe has a lovely stock video that is, basically, exactly what I asked Firefly to create and serves as an example of what the AI should be aiming for:
Yes, the AI-generated video is technically cheaper but it’s also unusable for anyone with better than legally blind vision. Beyond that, the real video clip is available beyond 1080p resolution, which is something Firefly can’t yet match.
The wolf is, I think, the best generation. Additionally, real stock footage of wolves is scant and often pretty bad. That said, what Firefly created looks like a moving painting, not like a video clip that would be usable for anything an editor would purchase stock footage for. And, again, Adobe Stock has a good option that basically fulfills the demand of the prompt and does so significantly better than Firefly did.
The secondary use case is using these tools to “ideate” ahead of an actual shoot. In that case, sure — you could argue these get the general vibe. However, if I wanted to give an idea of a sweeping shot of San Francisco to a team, I could just use that watermarked stock footage video. Same for the wolf. Maybe it could be argued the “camera reviewer in the snow” is more specific than what is available on Adobe Stock, but at the same time, the clip below gets the job done pretty well, in my opinion.
It’s also — and this is important — free. Why would anyone pay for a worse, eldritch horror version of the above when the preview Adobe allows anyone to download is free to access?
I can see the argument that I shouldn’t be so hard on Adobe Firefly Text-to-Video here because it’s a beta and it’s not nearly the “final form” that it will eventually reach. Even so (and I am not convinced it will get to the point where it can ever replace real footage), why then is Adobe charging for it? Charging for a beta that produces low-resolution, inaccurate, unusable, and extremely short clips feels like something thought up in a board room by bureaucrats who have never once even spoken to an artist.
Firefly’s text-to-video tool is not good enough to be used for what Adobe says it’s there to be used for and I’d have this opinion even if it was free. But it’s not. It’s expensive, especially considering videos are difficult to get right on the first go-around and multiple attempts will often be required. In no world is this worth $10 a month, let alone $30 or more. And that’s an important note: starting in one month, it will cost more — that is for certain.
Adobe already says that the $30 tier isn’t going to be the top-end option.
“Additionally, a new Firefly Premium plan designed for professionals looking to generate video content on a regular basis will be coming soon to provide even more audio and video capacity for high-volume creators and teams,” the company said in a press release.
When it’s not worth using for free, I’m not sure it matters how much it will cost.
Source link
