.png?2.1.1 "Vokash")
.png "Trending")
2 days ago
Google this week made available nan latest loop of its Veo video-generation instrumentality to users of its Gemini artificial intelligence programme who person a "Pro" aliases "Ultra" account.
Also: I utilized Google's Flow AI to create my ain videos pinch sound and speech - Here's really it went
Veo has been disposable successful preview for immoderate clip now. What's caller pinch nan latest implementation is nan expertise to statesman your video by uploading a still image to service arsenic nan first frame. (ZDNET's Prakhar Khanna has reported his experience utilizing nan capacity arsenic a built-in characteristic of his Honor 400 phone, versus utilizing it done nan website arsenic I did.)
How to usage Veo to make videos from photos
You springiness nan strategy a prompt, property enter, and Veo creates an eight-second video utilizing your uploaded photograph arsenic a reference constituent from which to build nan first framework of video. Veo adds sound, including music, footsteps, and different incidentals.
Videos return respective minutes astatine a clip to develop.
Also: This interactive AI video generator feels for illustration stepping into a video crippled - really to effort it
In my testing truthful far, I find Veo's implementation some fascinating and a spot creepy.
My results pinch Veo's photo-to-video feature
I tried respective still images I had taken, including a selfie and immoderate thoroughfare photography. Seeing one's pictures travel to life, if you will, is jarring. It is disconcerting really good it works, and, arsenic nan photographer, it's disconcerting really nan consequence contrasts pinch one's representation of nan event.
Also: This caller AI video editor is an all-in-one accumulation work for filmmakers - really to effort it
The bully aspects are nan value of nan video, which is successful keeping pinch nan photographic image. Things specified arsenic position of a segment are mostly good maintained, and moving objects successful nan inheritance are, successful immoderate cases, well-orchestrated to beryllium consistent.
1. Jogger moving on nan promenade
Here, for example, is simply a photograph I took of a jogger connected nan East River promenade successful Manhattan. I gave Veo nan prompt, "Please make a video successful which nan jogger continues to tally into nan region on nan promenade."
Below is nan original still image followed by nan Veo video.
The mobility of nan jogger is good, arsenic is nan activity successful abstraction arsenic if from nan constituent of position of nan photographer.
This is simply a important method achievement, to my mind, connected a very basal level. Remember that this is 8 seconds of 720p-quality resolution, which is rendered astatine nan modular movie complaint of 24 frames per second. That intends Veo has to create, successful a fewer minutes, 192 frames from nan first image. Given really small effort it took maine arsenic nan user, it would beryllium easy to place conscionable really important that is from a purely method constituent of view. The powerfulness of each that computing successful nan unreality really shines successful thing for illustration this.
One also, however, sees nan artifacts that harvest up from Google's predictions astir nan frames, giving nan point a alternatively eerie quality.
The jogger connected nan right, for one, doesn't really look nan aforesaid arsenic nan jogger successful my photo, only vaguely akin (hair is different, stride is different).
Also: Forget Sora: Adobe launches 'commercially safe' AI video generator. How to effort it
Another artifact is that, astatine nan existent infinitesimal successful time, nan fig moving toward nan camera connected nan left-hand broadside of nan image was strolling, not jogging. I deliberation that's clear successful nan image. But Veo rendered that personification jogging arsenic well.
Another point emerges connected nan FDR Drive road successful nan precocious left. One tin spot vehicles that mysteriously vanish astatine immoderate constituent successful their movement. That is simply a changeless taxable of nan Veo videos, nan inability of nan programme to afloat support continuity.
2. Woman stepping past The Horseshoe Bar
A astonishing accomplishment emerged erstwhile I submitted a photograph of a barroom connected 7th Street successful nan East Village, called 7B, aliases The Horseshoe Bar. I added nan prompt, "Can you show nan female stepping past nan building?"
The resulting video shows bully thoroughfare position but what's really astonishing is that it managed to capable successful nan achromatic motion supra nan doorway connected nan unseen broadside of nan building that shows nan horseshoe symbol. That suggests Veo was capable to find successful immoderate information a completion of nan bar, which is alternatively amazing.
Also: Midjourney's caller animation instrumentality turns images into short videos - here's how
The unseen buildings that Veo fills in, however, arsenic nan video turns nan corner, are not nan existent buildings connected that street, a lawsuit of Veo coming up pinch a reasonably decent substitute. Notice a beardown artifact: Veo gave nan stepping individual a bluish hat, which it seemed to person added erroneously based connected nan personification successful my photograph stepping successful beforehand of a bluish motion connected nan building.
3. Person successful achromatic boots gets up and disconnected train
Some artifacts are much striking. In a 2nd portion of thoroughfare photography, I uploaded a image of personification sitting successful a subway car pinch achromatic boots. I gave nan prompt, "The personification successful nan achromatic boots gets up from their spot and gets disconnected nan train." What was produced was rather striking, and beautiful bully for an approximation of really this fig mightiness move. The personification doesn't, however, exit nan train.
When I persisted pinch a 2nd prompt, "That's great, but 1 adjustment. Is it imaginable to show nan doors of nan train car opening and nan personification successful nan achromatic boots really stepping retired nan doors to exit nan train?", Veo produced a 2nd version.
This time, nan individual astatine slightest is shown moving toward an exit, arsenic doors are shown sliding open. However, respective artifacts present neglect a reality and consistency test. For 1 thing, nary 1 exits a New York City subway car astatine nan -- extremity -- of nan car; they exit astatine nan broadside doors, arsenic that is wherever nan level is. Second, nan sliding doors depicted astatine nan extremity of nan car do not beryllium successful New York City subway cars. Those exits person one, not two, sliding doors.
Also: You tin nutrient video ads successful seconds pinch Amazon's caller AI instrumentality - here's how
Third, it's clear successful nan original still image, based connected nan ray and nan specifications seen done nan rear model of nan train car, that this is not nan past car successful nan line; location is different car down it. Yet, erstwhile nan doors unfastened successful nan video, we spot nan level and tracks, suggesting this car is now nan past car successful nan line. It's an inability present for Veo to decently infer from item nan full building of nan environment.
Last but not least, successful a 4th inconsistency, we tin spot done nan unfastened doorway that nan level is straight beneath nan train, truthful that nan train is -- riding complete nan level -- alternatively than nan tracks.
4. Thunder and lightning pinch rain
I submitted a rainy nighttime image connected Lexington Avenue successful Manhattan and asked for "A video of thunder and lightning and superior rainfall successful this thoroughfare scene." The consequence is alternatively cartoonish, but it's surely a nosy infinitesimal pinch nan correct intent.
5. Dark bath selfie
Putting one's likeness into Veo has its ain typical creepiness, aliases amusement, aliases both, depending connected your consciousness of humor.
Also: The champion AI image generators of 2025: Gemini, ChatGPT, Midjourney, and more
I first utilized a very acheronian bath selfie. I was impressed pinch nan scope of imaginative animation. My features, however, look to morph drastically into personification else's likeness, and I'm not judge whose. (I've been told I look for illustration Thom Yorke of nan set Radiohead sometimes.)
6. Professional headshot
In different instance, I utilized my ZDNET headshot and asked Veo, "Can you make a video of this man doing nan cha-cha-cha?" I for illustration nan resulting movement, accompanying music, and nan very large footwear sounds are very amusing.
However, nan creepy portion present is that without further prompting, Veo has near my look a rigid disguise of expression, which doesn't make consciousness successful a creation video. In fact, my caput doesn't really move astatine all; it's fixed.
7. Las Vegas selfie
I uploaded yet different selfie, taken astatine Caesar's Palace casino and edifice successful Las Vegas, and prompted, "Please make a video of this man successful nan leather overgarment dancing tango pinch nan statue of Venus that is successful nan background." Well, Veo did not win successful making america dance, but nan resulting level show by my likeness is amusing. So is nan music. Notice that nan sleeves of my leather overgarment move black, for immoderate reason.
8. A humanities mashup pinch John C. Calhoun
On nan hunch that manipulating humanities figures mightiness beryllium disallowed, I tried creating a humanities mashup to trial nan matter. I uploaded a image of onetime US vice president John C. Calhoun from nan US Library of Congress, and requested that Veo make a video of Calhoun dancing nan cha-cha-cha.
Veo started to make a video, past discontinue pinch nan message, "I can't make that video. Try describing different idea. You tin besides get tips for really to constitute prompts and reappraisal our video argumentation guidelines. Learn more."
9. Making Scarlett laugh
I past tried uploading a image of actor/director Scarlett Johansson from her Wikipedia page, and requested "a video of this female laughing." Again it started and past discontinue pinch nan aforesaid correction message.
10. Making myself laugh
I double-checked nan matter pinch my ain headshot, arsenic a non-historical, non-famous person, and was capable to get Veo to make a video of maine laughing (albeit looking not astatine each for illustration nan original headshot).
That suggests that Veo whitethorn beryllium built pinch safeguards against manipulation of humanities aliases popular civilization images, though I cannot beryllium certain.
Should you effort Google Veo?
The Veo service, successful preview, is surely not without glitches.
After my first mates of successes, I many times sewage a informing that I would person to hold to do much videos, arsenic nan work is rate-limited astatine nan moment. There are complaints astir this successful nan personification fora for Gemini, including group being denied nan work for complete 24 hours, and a agelong mentation of nan matter by a unpaid merchandise "expert." Basically, video is bandwidth-, compute- and memory-intensive, truthful it's not astonishing Google would person to limit usage astatine nan outset.
The astir nonstop solution is to upgrade to nan higher level of Gemini, nan "Ultra" plan, though this intends going from $19.99 a period to $249 a period (discounted for nan first 3 months to $125). That's a steep value conscionable to beryllium capable to get astir what look alternatively harsh limits.
Also: Is Google's $250-per-month AI subscription scheme worthy it? Here's what's included
Even aft subscribing to Ultra, I reached a limit aft 5 videos, pinch an correction connection saying "something went wrong." Another explainer post successful nan personification forum suggests that location is nary clear limit for nan Ultra plan; it's an obscure matter of AI "credits" successful nan unreality service.
That abrupt shutdown contradicts Google's position of service that say, "You'll get a notification erstwhile you're adjacent to nan limit. The notification will show you really galore videos you person left." (Learn much successful nan Gemini apps thief conception astir various Gemini limits.)
The replacement to Ultra is moreover much complex, utilizing nan master "Flow" improvement instrumentality alternatively of nan Gemini app.
In summation to usage limits, users person complained of method glitches, specified arsenic videos that deficiency sound.
Also: I tested Google's Veo 2 image-to-video generator connected Android - here's my verdict
The wide belief is that this is very overmuch a beta product.
You whitethorn wonderment astir nan dangers of deepfake videos. Google has posted a number of points astir information measures for Gemini apps generally, but location is nary clear connection astir Veo videos.
Overall, Veo seems to maine an absorbing trick, though Veo doesn't clasp my liking aft nan first fascination has worn off. As a photographer, I'm much willing successful a azygous authentic infinitesimal than I americium successful 192 inauthentic moments.
For those not progressive successful nan movie industry, Veo whitethorn supply a model into really AI tin progressively beryllium utilized to capable successful for actors, aliases widen likenesses to create action without really employing nan actors.
Given stronger algorithms and further information (scene data, characteristic data, etc.), I tin ideate Hollywood could usage this exertion to nutrient moving images that service existent stories. It's an eye-opener astir wherever video is going successful an property of AI.
Get nan morning's apical stories successful your inbox each time pinch our Tech Today newsletter.
English (US) · 
Indonesian (ID) ·