# Google Just UNLOCKED the Nano Banana of AI Video (Gemini Omni Deep Dive)

https://www.youtube.com/watch?v=7HhlSu3pPvU

[00:00] Google just unlocked Gemini Omni, making this the nano banana moment for video.
[00:05] Imagine uploading any footage and transforming it into dozens of different styles, the effect shots, and endless creative possibilities, all while keeping your subject and lip sync consistent.
[00:17] That's what Gemini Omni does.
[00:19] This is a powerful new addition for creators, but if you're using it the wrong way, you're going to waste a ton of credits, and that's where I step in.
[00:28] I'm Jack.
[00:30] I've spent over a decade creating visual effects for many of the biggest brands on the planet, and now I'm using AI to reinvent creative workflow.
[00:37] In today's video, we're taking a deep dive into Gemini Omni, breaking down exactly how it works, when it shines, and where it falls short.
[00:47] And of course, I'll share all of my prompts and materials for completely free.
[00:51] So, by the end, you'll know exactly how to get the most incredible results from this model.
[00:55] If you do enjoy today's content, be sure to whack a like on there for me and consider subscribe.
[00:59] All right, let's
[01:01] Get into it.
[01:04] So, as you guys saw in the intro, Omni Strong Suit is video editing.
[01:06] We can upload footage and, for example, completely change the lighting, replace a product with impressive objects tracking, or even change the subject's appearance, like, I don't know, removing my beard.
[01:18] This generation is pretty much exactly what I look like without facial hair, which is crazy.
[01:23] And also all the proof I need that this beard is not going anywhere.
[01:26] To really dive into Google's new video model, I spent a ton of credits to carry out a variety of different tests.
[01:36] And as I take you guys through these, you're going to learn a bunch of tips and tricks along the way, which will stop you making mistakes and ultimately make your credits go further.
[01:46] I'll also perform a direct comparison to Sea Dance 2 towards the end of the video, so make sure you stick around for that.
[01:53] The first thing I wanted to test was an effect which is going viral all over social media right now.
[01:58] And that's using Omni to turn an object into glass.
[02:01] But I
[02:05] wanted to take this one step further by uploading a clip where I'm not only talking to test the lip sync, but also rotating my phone and asking Omni to understand that we can only see through the phone screen.
[02:17] The back of the phone remaining as a solid object.
[02:20] pretty interesting test this.
[02:23] So over on Higsfield, we want to go up to the top lefthand corner, hover over the video tab, and then click into Gemini Omni Flash.
[02:31] There are two different ways to use this tool.
[02:33] The best in my opinion is elements.
[02:36] And this is where we can upload one video reference and up to five image references.
[02:40] And below we can enter our text prompt, select the duration and aspect ratio.
[02:45] And it really is as simple as that.
[02:47] Now, you do also have the option to use the more traditional route, which is start and end frame, but I find this a little bit restrictive.
[02:53] Generally speaking, we want to take full advantage of the omni reference model.
[02:58] So, I uploaded this footage of myself and I engineered a
[03:05] prompt using Claude.
[03:08] We'll go ahead and break down this prompt in a second, but first, let me show you the result that we got back.
[03:12] All right.
[03:14] So, I want to test if we apply this effect whilst I'm talking.
[03:17] Does it still work?
[03:20] Does the lip sync hold up?
[03:23] So, a pretty incredible result here where we're testing the models world understanding.
[03:28] Considering my face is being covered up by the phone in the source footage, Omni is doing a really great job of keeping my appearance consistent when we're seeing through the glass.
[03:37] It's even adding these really nice reflections and subtle kind of fingerprint smudges to the glass, which really just helps to sell the effect.
[03:46] And of course, it's having no issue in understanding that when we show the back of the phone, we don't want to be able to see through it.
[03:56] If I was going to be hyper critical, you'll notice that there is this kind of small black shape on the left hand side of the glass, which kind of morphs around and behaves in a bit of a strange
[04:07] way.
[04:10] But for my very first generation with Gemini Omni, I thought this was really quite impressive.
[04:14] Now, the prompt I used was a bit of a twoparter.
[04:18] So, the first section is essentially a bunch of constraints.
[04:23] We're telling Gemini Omni what we don't want to change.
[04:28] And for the bottom section, I'm describing the effect that I'm after.
[04:32] So you guys can actually go ahead and reuse this framework if you wanted to by swapping in your own desired effect at the bottom.
[04:38] And if you did want to go ahead and try any of this out, you'll find all of my prompts and materials available on my free school community.
[04:44] Link for that down below in the description.
[04:47] I also did a similar test where I held up my guitar here and applied that glass effect.
[04:55] And once again, the results are really quite impressive.
[04:58] The only issue is that when I eventually rotate my guitar, you can't see the scratch plate, the strings, and all of those bits.
[05:05] So, there is a bit of a lapse in terms of
[05:08] understanding there.
[05:09] But, I love the way that this kind of beveled section on the back of the guitar is causing this distortion to my face when we're looking through it.
[05:16] Next, I wanted to test out restyling realworld objects.
[05:19] So, we have this pretty interesting footage here where I have my glasses in my hand.
[05:26] and I hold them up close to the camera and I rotate them and then put them back on.
[05:31] And this is quite complex visually because the glasses are in sharp focus with a really shallow depth of field behind them.
[05:38] And interestingly, we've also got this extra level of magnification when we're looking through the lenses.
[05:44] So, I was really interested to see what would happen if I asked Gemini Omni to swap between different types of glasses.
[05:49] So very similar prompt approach once again and this is the result that we got back.
[06:08] Really impressive stuff here.
[06:11] Omni is dealing really well with those changes in magnification and depth of field as well as showing us that it can be very good at object tracking.
[06:21] All of the glasses look realistic.
[06:23] Nothing is breaking here.
[06:23] Really cool.
[06:26] And I did a similar test where I pick up a Stanley cup and rotate it.
[06:29] And I asked Omni to preserve its shape and branding, but morph it into a variety of different materials.
[06:38] Again, the object tracking, reflections, and shadows look pretty amazing.
[06:44] Omni has ended up adding an extra logo on the back side of the Stanley Cup, though.
[06:49] But I love the fact that when the material changes, we can actually see that in the reflection of the glass table as well.
[06:56] And here's another effect that's just all over social media that I had to go ahead and try out.
[06:59] I uploaded some footage of me holding and rotating this brown folded envelope to see if Gemini Omni could
[07:10] Turn this into a Pokémon card.
[07:13] And of course, I wanted to go for the PSA 10 Charizard.
[07:17] And I had a pretty interesting approach to this.
[07:19] So, I actually wrote front and back on the corresponding sides of the envelope.
[07:23] And I also made use of the Omni reference model by uploading the front view of the card as well as the back view and labeling them with the hope that Omni would match that up to the front and back of our folded envelope.
[07:38] Once again, a pretty similar approach in terms of prompt.
[07:43] Let me go ahead and show you the result that I got back.
[07:52] Now, obviously, I was pretty brutal with this test.
[07:57] A lot of the results you're going to see on social media are actually playing it pretty safe.
[08:01] I really like to put these tools through their paces.
[08:04] So, I'm rotating that envelope a bunch of different times to make this a real challenge.
[08:09] For the most part, Omni actually holds up really
[08:11] well.
[08:13] The object tracking and the lighting on the card are pretty close.
[08:15] It's just right towards the end here that Omni loses its grip and understanding of where the front and the back of the card are supposed to be and we just continually see that Charizard artwork again and again.
[08:29] There are also some text legibility issues here on the card itself.
[08:32] So, a bit of a mixed bag this test, I would say.
[08:35] Sticking with the theme of object tracking, I wanted to see if I could upload some footage of me holding and rotating a serial box and have Omni convincingly swap out the artwork.
[08:48] So, this is less of a flashy V effect style shot and more of a workflow geared towards advertising.
[08:54] First, I wanted to see what Omni was capable of creatively.
[08:56] So, I went ahead and dropped in this really simple prompt here.
[09:01] And this result is really cool.
[09:04] I love the design that Omni has come up with here.
[09:09] And the object tracking once again
[09:11] working really well, but there are those text legibility issues once again.
[09:16] It seems like with large bold text, Omni performs well, but when we start to have that really kind of small type face, that's when we get that classic AI garbled up look.
[09:28] And in fairness, this is a problem when it comes to most AI video models.
[09:30] cling.
[09:33] Probably the best alternative for small text detail, I would say.
[09:36] I was also interested to see if I could get on me to essentially prepare the serial box for pack replacement.
[09:43] In advertising, it's often the case that there will only be one product that gets filmed and then in post-prouction, VFX artists will actually replace the artwork on each face of the box so that the ad can be reversed to work anywhere in the world.
[09:59] So, I wanted to see if I could add these tracking markers that could be useful for a visual effects workflow.
[10:06] So, again, this isn't some flashy V effect shot, but more of a look at a realworld
[10:12] use case in the creative industry.
[10:15] Very impressive in terms of the object tracking and the lighting once again.
[10:19] Now, moving on to some impossible visuals.
[10:21] I wanted to try to recreate an effect that you're seeing everywhere right now.
[10:26] So, I uploaded this footage of me holding out my hand towards my laptop screen.
[10:31] And the idea is that I wanted Omni to bring the bird to life so that it jumps out of the screen and sits on the palm of my hand.
[10:38] The issue I kept running into here is that for some reason Omni kept changing the content on my laptop screen.
[10:47] You can see that we've got Photoshop open instead of the Higsfield UI.
[10:52] The bird has completely changed, including that baseball cap.
[10:57] There's all sorts of kind of stranges happening here, even though the effect itself is quite impressive.
[11:03] After a bunch of different tests, the only way I could get the result I was after was actually just using the start and end frame feature inside of Gemini Omni.
[11:12] Now, you can see that the coloring of
[11:14] the bird is still changing a bit.
[11:16] That's kind of a shame.
[11:18] Uh, but the overall effect is looking pretty great, I have to say.
[11:20] Just to be clear, both times when I was using the omni reference approach versus the start and end frame approach, I was using the exact same very simple text prompt that you can see on the screen here.
[11:31] Useful bit of information to pass on to you guys if you're struggling to get the result that you're after.
[11:37] Start and end frame may be the way to go.
[11:39] I also wanted to test out this footage here of me holding up this empty glass, rotating it so that it's upside down.
[11:46] I went back to my more technical prompt structure here where we're swapping in the effect at the bottom and I'm asking Omni to fill the glass with water, but also make sure that when I rotate the glass, the water doesn't fall out.
[12:00] And the result was really quite impressive.
[12:02] The fluid dynamics look very believable.
[12:07] The only thing I was a bit on the fence about is that once that glass is rotated, has the volume of water been reduced a little?
[12:15] Now, my wife pointed out that the top of the glass is of course wider than the bottom.
[12:20] So, maybe this is actually accurate.
[12:22] You know what? You guys let me know down below in the comments what you think.
[12:26] All the same, I think there is a lot to be impressed with here.
[12:29] Now, we have to move on now and take a look at the competition, which of course is Sea Dance 2.
[12:36] So, first I wanted to recreate the glass phone effect.
[12:38] So just a reminder of the result we got from Gemini Omni Flash and using the exact same video reference and prompt.
[12:46] This is the result we got from Sea Dance 2.
[12:49] So whilst it's kind of generally working, there are actually some problems here.
[12:55] The glass itself has that kind of white tint to it.
[12:57] When we rotate here, we can still see the top of my phone, which we did not ask for in the prompt.
[13:05] So actually arguably Gemini Omni has done a better job here which is kind of surprising.
[13:13] Another test which you guys would have seen actually right at the
[13:17] start of the intro was when I used Gemini Omni to make this kind of opening title animation in like a plaster scene style.
[13:24] So I wanted to try to recreate this with Seance 2.
[13:27] I was really quite impressed with the results with Gemini Omni.
[13:31] And here is the result that we got back from Sea Dance 2.
[13:34] Again, it's pretty good, but for some reason, that kind of opening section where the plaster is seen as being like formed or molded comes across as a little bit gross to me.
[13:45] Generally speaking, I actually prefer the output we got from Gemini Omni.
[13:49] And then I wanted to test out our bird V effect shot.
[13:52] So, just a reminder of the result that we got back when we were using Google's new video model.
[13:59] You'll remember that I did have some problems getting towards this end result and ended up having to use a start and end frame workflow.
[14:05] So, here is the result that we get back from Seance 2.
[14:08] I was able to just use their Omni reference workflow here.
[14:13] And whilst I think the result is really quite
[14:17] impressive, I feel like the bird has changed into like a kingfisher.
[14:23] I'm not good with my bird breed knowledge, but to me it looks like maybe it's actually changed the uh type of bird here, even though the visual effects themselves are really quite impressive.
[14:37] And I also at this stage thought it'd be great to carry out a comparison against Gemini Omni and Sea Dance 2 Mini because of course Mini is more cost effective and a bit closer to Gemini Omni generally speaking.
[14:50] That's when we're talking about the amount of credits you have to spend and the time for the generation.
[14:56] So, you can see here that the hat falls off.
[15:01] And there are a couple of other issues.
[15:02] Again, it feels like the breed of the bird is maybe changing.
[15:05] Generally speaking, a decent result.
[15:07] It's just losing the hat there is kind of breaking the illusion, I would say.
[15:12] Next up is the monkeys, which you guys would have seen in the intro as well.
[15:16] So, the source footage is of course just me
[15:19] talking to camera, reading out the intro script.
[15:21] Then we went ahead and used this very simple prompt here to add in monkeys all over the place just causing carnage in my office.
[15:30] So, this is the result that we got from using the same video reference and prompt within Sea Dance 2.
[15:37] And I have to say that the visual fidelity is so much sharper than what we had inside of Gemini Omni.
[15:43] Of course, that's locked to an output of 720p with Seance 2.
[15:46] We can actually go all the way up to 4K.
[15:51] Now, this is just 1080p, but everything feels a lot sharper.
[15:54] And the way that the monkeys are moving and everything looks great, especially that one in the top leftand corner moving over my little neon sign.
[16:03] That looks really pretty impressive.
[16:06] Same video reference and prompt inside of Sea Dance Mini.
[16:08] Again, the result is pretty great.
[16:11] A little bit less sharp.
[16:14] And there's this kind of moment towards the end of the video where all of the
[16:19] monkeys randomly kind of decide to go down into the floor.
[16:22] I don't know, maybe they found some food or something and they're kind of piling on there.
[16:28] Um, but not quite as impressive a result.
[16:30] A bit closer to what we would have seen with Gemini Omni.
[16:32] We also have the glasses test.
[16:35] So, a reminder of what we were able to achieve with Google's latest video model here.
[16:39] Very nice result.
[16:42] So, a comparison here with Seance 2 and I have to say it's dealt with the shallow depth of field and the magnification in the lens as well, but I don't actually like the result as much.
[16:54] Yes, it is sharper, but it seems a little bit less kind of visually interesting and less creative than what we got back from Gemini Omni.
[17:01] You guys also let me know what you think of this comparison.
[17:03] And then we also have the Seance 2 mini as well.
[17:05] Pretty close to the result we got from Seance 2, I have to say.
[17:11] So if you guys were looking to kind of cut down your credit cost, might actually be worth trying out Seance 2 fast or mini
[17:21] options as well.
[17:23] Then we have the Stanley Cup shot where we were changing the materials using Gemini Omni.
[17:28] generally a great result, but it was adding on that extra logo on the back face of the product.
[17:32] With Sea Dance 2, there's a lot of compression for some reason on that glass table.
[17:38] I'm not too sure how well that's going to translate across uh when you guys are watching this YouTube video, but it's looking like really nasty kind of like digital artifacting, which we didn't have with Gemini Omni.
[17:52] Pretty interesting.
[17:54] The morph of the kind of material on the Stanley Cup does look good.
[17:57] It's also interestingly adding that logo to the back face of the Stanley Cup.
[18:02] However, it's kind of like on its side in a weird place.
[18:08] So, it's added an extra logo, but also changed its orientation.
[18:13] So, I would actually say that Gemini Omni wins on this test.
[18:16] And then for Sea Dance 2 Mini, I think the result is pretty
[18:21] similar to be fair, but again, we're getting that extra logo on the back and there's a point on the wooden face where it's just completely garbled and just not making any sense whatsoever.
[18:31] So, to give you guys a little bit more context, Seance 2 is currently around four times more expensive than Gemini Omni.
[18:42] Even Seance 2 Mini is around 1.5 times more expensive.
[18:46] Although these prices do tend to fluctuate slightly depending on offers that might be going on at any given time.
[18:53] So after this really varied amount of tests, what do I actually think of Gemini Omni?
[18:59] Well, let's start with the good.
[19:01] This model is pretty cheap, especially when you compare it to Sea Dance 2.
[19:06] It's also really quite fast with the majority of my generations coming back within 1 to 2 minutes.
[19:11] But of course, like any AI tool, it isn't without its problems.
[19:15] Now, I touched on this earlier, but Omni is locked at 720p
[19:23] output resolution, and that is really quite low.
[19:28] When I wanted to use my generation's full frame in the intro, I did feel the need to go ahead and upscale them.
[19:32] So, to do this, you can carry it out inside of Higsfield.
[19:35] I selected Topaz Video.
[19:37] Their latest model, Starlight Precise 2.5, is really, really quite good.
[19:43] But obviously, this comes at an extra credit cost, so this isn't going to work for everyone.
[19:48] Now, we do need to keep in mind that the full name of this model is Gemini Omni Flash, essentially meaning that it's the light version of the tool.
[19:58] So, I personally would expect to see a pro version dropping hopefully in the next 2 to 3 months.
[20:03] not as much of a problem, but we are locked to a 10second generation limit as well.
[20:09] And as you guys may or may not be aware, when Seance 2.5 drops in the coming weeks, we'll be able to generate 30 secondond long videos.
[20:20] So,
[20:23] there is a big difference there, but
[20:25] also not really a deal breaker there
[20:27] because there are plenty of shots that
[20:29] are featured in films and adverts that
[20:33] are below 10 seconds long. And something
[20:35] that is a bit of a kicker in my opinion
[20:38] is poor prompt adherance. Sometimes even
[20:42] with a really well-written prompt, Omni
[20:45] is just going to do its own thing. It
[20:47] just doesn't really care what you're
[20:48] after. So you've either got to deal with
[20:51] the result that you get or go ahead and
[20:53] regenerate. So an example here where I'm
[20:55] asking Gemini Omni to show me
[20:58] levitating. So I lift up off the ground
[21:00] and then gently fall back down. It just
[21:03] didn't do that at all. Even though the
[21:05] prompt is very clear, very simple, uh it
[21:08] just ends up with me looking really
[21:10] embarrassing. Basically just giving me
[21:12] back the source footage. Another example
[21:15] with the serial box test where you can
[21:18] see that in that opening part of the
[21:20] shot, it's actually gone ahead and
[21:22] removed my glasses and flipped the
[21:24] orientation of my baseball cap. Again,
[21:27] things that I definitely didn't ask for
[21:29] inside the prompt. And not so much a
[21:32] prompt adherance problem, but still a
[21:34] good example is where I actually
[21:35] uploaded some footage where I was using
[21:38] AI to build this kids animation. And I
[21:40] was asking Omni to go ahead and remove
[21:43] all of the characters, which you can see
[21:45] in this one here, from this particular
[21:47] shot. Whilst it does a pretty good job,
[21:50] you can see that the character is
[21:52] randomly like moving that train, which
[21:54] of course you wouldn't want. And I
[21:56] actually went back to my more kind of
[21:57] complex prompt structure here to ask
[22:00] Omni to remove the logos from the rug,
[22:03] which you can see on this generation.
[22:05] And it did do a really great job, but
[22:08] for some reason at the end of the
[22:09] generation, you get this kind of
[22:11] stuttering where it's like looping the
[22:13] footage. So that's another bit of a
[22:15] strange problem there. I also input this
[22:18] source footage again from the kids
[22:20] animation. And all that I was asking
[22:23] Omni to do was to swap out the outfit
[22:27] for the character so that he has these
[22:29] pajamas and a cute night cap. But you
[22:32] can see it's just completely shifting
[22:34] the style. It's changing the
[22:36] environment, uh, the bedding, everything
[22:39] has just ended up being swapped out,
[22:41] which of course is not what I asked for.
[22:43] I'd imagine you guys watching probably
[22:46] haven't touched VO3.1 for some time now,
[22:49] but the prompt adherance or kind of lack
[22:52] thereof does feel quite similar to that.
[22:54] Now, this isn't every time, but to feel
[22:57] like you're using VO3.1 when you've just
[23:00] come from using Seance 2 can feel like a
[23:04] bit of a step down at times. Then again,
[23:06] Gemini Omni is much cheaper and faster.
[23:11] So, it's kind of swings and roundabouts
[23:13] here. I'd love for you guys to let me
[23:15] know how you're getting on with Gemini
[23:17] Omni. If you have enjoyed today's video,
[23:19] be sure to whack a like on there for me
[23:21] and consider subscribing. Thank you so
[23:22] much for watching, guys, and I'll catch
[23:24] you in the next one.