How to Frames: Mastering Start and End Frames in Google Flow (2026) — Transcript
Full transcript
- 0:00Of all the different ways you get to
- 0:01create AI videos, the start and the end
- 0:03frame are the most reliable. Welcome
- 0:05back to this channel and on this one we
- 0:07will explore the frames tool on Google
- 0:09Flow. This is a straightforward direct
- 0:12guide. I'm not going to diverge into
- 0:13different topics. I'm just going to show
- 0:15you what the frame is, how it works with
- 0:17[music] examples.
- 0:20Everything starts with your frames.
- 0:22Think of this as your first most
- 0:24important step. If the image quality or
- 0:26the idea is poor, well, the video will
- 0:29be too. To get the most out of Flow and
- 0:32frames, [music] you really need to
- 0:33sharpen your image generation skills
- 0:36first. We will create a few examples,
- 0:38but for the first example, [music] I
- 0:39want to create the images right here
- 0:41with you. I will go at the bottom. I'm
- 0:43using Nano Banana Pro 16:9 and one
- 0:45single generation. If you want to use
- 0:47Nano Banana 2, this will work as well.
- 0:50And [music] this is my prompt. It's just
- 0:52a full body shot of a young woman just
- 0:54walking on a trail. We have soft natural
- 0:56light and subtle shadows and then we
- 0:59finish with a natural texture candid
- 1:01shot. I just want a lady walking in the
- 1:03forest. That's it. If you want the
- 1:04prompts and images I'm using for these
- 1:06examples, let me know down in the
- 1:07comments and I will send you a link. We
- 1:09get it back and this is what I get. It's
- 1:11just a woman walking in the forest.
- 1:13[music] Fine, this will work. Now, then
- 1:16you need to create the end frame. This
- 1:18will be the starting frame. And to
- 1:19create the first frame from the starting
- 1:21frame, you have [music] two choices. You
- 1:23can use ingredients or go to the edit.
- 1:25And I'm going to do both, why not? I'm
- 1:27going to go right here, just take this
- 1:28image and then drop it as an ingredient.
- 1:31Remember we are using Nano Banana.
- 1:33[music]
- 1:33Since she's not looking at the camera,
- 1:35what I want is for her to look at the
- 1:37camera and do something like that.
- 1:39Simple. This is my prompt, medium close
- 1:41shot. The subject is holding both hands
- 1:43up towards the viewer. Her expression is
- 1:46joyful, smiling, looking directly at the
- 1:49lens. And then I provide the instruction
- 1:51for the, you know, the arms. Let's run
- 1:53the prompt and hopefully we get
- 1:54something
- 1:55something consistent. While this one is
- 1:57going, you have a lot of different
- 1:58methods of creating different images by,
- 2:01you know, using sheets or using
- 2:03references or using ingredients or
- 2:05elements. Again, you have a lot of
- 2:07different methods. This is not the main
- 2:09topic of this video. I've created other
- 2:11videos about subject consistency. If you
- 2:13want to take a look, go to the cards or
- 2:15check my channel. We get it back a
- 2:17response and we get something, you know,
- 2:19super consistent. It looks like it's
- 2:21her, but most importantly, you need to
- 2:23take a look at the the background, at
- 2:25the setting. On the end frame is
- 2:26something super different in the back or
- 2:28the subject, you will be getting these
- 2:30weird transformations, which in this
- 2:32case is something that we don't want.
- 2:34So, this is one way of creating the end
- 2:36frame from the starting frame. The other
- 2:39way is by going directly to the image
- 2:41and then running an edit. We don't need
- 2:43to add ingredients. I'm just going to
- 2:45use the same prompt that we used a
- 2:47minute ago. I'm just going to run the
- 2:49task. When we get it back, we can see
- 2:51it's again still pretty consistent. Now,
- 2:53if you ask me which one which method I
- 2:55like the most, I like the edit a little
- 2:58bit more than creating it with [music]
- 3:00ingredients, but as you can see, both
- 3:01methods will work. Should you spend a
- 3:04lot of time perfecting your start and
- 3:06ending frame? 100% yes. Crappy frames
- 3:10equal [music] crappy videos. And when
- 3:12you are creating the video, no amount of
- 3:14prompting will fix those errors. I will
- 3:17use this image. If you want to add it to
- 3:19your main dashboard, you just need to
- 3:20click on save to project and then make
- 3:23sure that you select this one before you
- 3:25exit. And once back on your dashboard,
- 3:27we can see that we have both images. And
- 3:30I can say both are, you know, are fine.
- 3:33Once you have your frames and you're
- 3:35100% sure, your job is to prompt
- 3:39whatever is that happens in between. I
- 3:41will use this one as my starting. I'm
- 3:43going to go at the bottom and select
- 3:44video and then use frames 16:9 one
- 3:47single output. Now, we have a new model,
- 3:49the fast, the light [music] and the
- 3:51quality. You know what? I'm going to use
- 3:52the light, why not? I'm going to take
- 3:54this as my starting frame. I'm going to
- 3:56I can just drag and drop to the starting
- 3:58frame or I can click on this one and I
- 4:01can select it from the menu. Just
- 4:03completely up to you. But I like to drag
- 4:05and drop. So, this one will be my
- 4:07starting and this one will be my ending
- 4:09[music] frame. You have an icon right
- 4:10here and with this one you can just swap
- 4:13the starting for the ending and the
- 4:14ending for the starting. And this is an
- 4:16important feature. We will talk about
- 4:18this in a minute. Of course, we see the
- 4:20images and we know exactly what we want.
- 4:23Nevertheless, it's a good idea to let
- 4:24the model know what we want from [music]
- 4:27this sequence. I'm going to go right
- 4:28here to the prompt and my prompt will be
- 4:31a woman is walking in the forest. The
- 4:33camera follows her with this low
- 4:35cinematic movement as she walks towards
- 4:37the camera. [music] So, we are pretty
- 4:39much describing the frames. Let's just
- 4:41run the task and see if we get something
- 4:43cool. Okay, so let's see.
- 4:45She's walking. She needs to look at the
- 4:47[music] camera.
- 4:48And that's weird, but she needs to open
- 4:51her arms. Now, for the most part, it's
- 4:54fine. [music]
- 4:54The VO it's adding some artistic
- 4:57direction, that leaf, but again, for the
- 5:00most part [music] it's a good
- 5:01generation.
- 5:03All right.
- 5:05Okay, so as you can see, it's pretty
- 5:06simple to work with. You add your
- 5:08starting, your ending and then you
- 5:10prompt what happens. Things that you
- 5:11need to be aware of, the background and
- 5:13the subject must match. In this case, we
- 5:16have a consistent subject and then the
- 5:18background it's consistent as well. If
- 5:21the frames are way too different from
- 5:23each other, you will be getting wild
- 5:25transformations. Unless you want this
- 5:27trippy five type of transformations. If
- 5:30that's what you want, then go for it.
- 5:32When you create a video, it has a
- 5:33limited duration. If you prompt for her
- 5:35to jump, laugh, spin, clap and then drop
- 5:39a line, that will fail or it might fail.
- 5:42The model will still try to deliver all
- 5:44of that, but it's better if you keep it
- 5:46concise. If you have a full body shot at
- 5:48the start, a good idea is to stay close
- 5:51to the full body shot. You can provide
- 5:53completely different shots from the
- 5:55starting and the ending, but again, most
- 5:58often than not, you will be getting
- 6:00somehow inconsistent generations.
- 6:02Sometimes you will be getting something
- 6:04awesome and sometimes you will be
- 6:06getting something, you know, not so
- 6:08awesome. And that's just completely
- 6:09fine. The thing is that it will cost you
- 6:12more credits. So, when you provide
- 6:14completely different frames, you can
- 6:16really expect some crappy generations.
- 6:19Just be aware.
- 6:21Okay, now that you know a little bit
- 6:23more about frames, let's explore some
- 6:25scenarios. The zoom is one of the safest
- 6:27motions that you can execute with
- 6:29frames. To do this, you need a zoom in
- 6:31and zoom out image. This is my example.
- 6:34I have a man on a mountain, let's say,
- 6:36and then the same man with arms open and
- 6:39looking up. You'll notice that the
- 6:41person in my images is in a slightly
- 6:43different pose. Now, I've created this
- 6:45on purpose. If the subject stays in the
- 6:48exact same pose, I mean, from a close
- 6:50shot and then just a full body shot,
- 6:53when you create the video and you zoom
- 6:54in or you zoom out, it will be super
- 6:57static. It will create a flat lifeless
- 6:59video, just scaling the image up and
- 7:02down. So, you need to make both shots
- 7:05dynamic. You can change a little bit of
- 7:06the lighting of what's going on in the
- 7:08back. Maybe if you have cars, you can
- 7:11add cars in the background or anything
- 7:12in the background that is moving. Maybe
- 7:14the wind blows and the hair or the
- 7:17clothing moves. Let's do it. I'm going
- 7:18to take this as a starting frame and the
- 7:21other one as an ending frame. And of
- 7:23course, we always need to prompt for
- 7:25something. A man standing on the peak of
- 7:28a mountain and the camera zooms out as
- 7:30he opens his arms. Again, we are pretty
- 7:33much describing the obvious, but you
- 7:37kind of need to do this. It's a way of
- 7:38reinforcing the model what we need to
- 7:41do. I'm going to still run it on light.
- 7:42Let's see what we get back. While this
- 7:45one is going, what you can do is to go
- 7:47right here at the bottom and you can
- 7:48just click and it will swap the frames.
- 7:51Now, we have the other motion. Instead
- 7:53of zooming out, we will be zooming in.
- 7:56I'm going to change the prompt, of
- 7:58course, for the camera zooms in as he
- 8:00closes his arms. Let's run it. We have
- 8:03the first one. Let's just take a look.
- 8:07Zoom out. Man opens his arms. It looks
- 8:10[music] very natural. And we end on the
- 8:14ending frame.
- 8:15Cool.
- 8:21You can see some issues with the video,
- 8:22but remember that we are using the cheap
- 8:25version of VO. If you run the quality,
- 8:27you will be getting something better.
- 8:29Let's take a look at the other one.
- 8:30Zoom in.
- 8:32It zooms in.
- 8:38Well, that was weird. Of course, this is
- 8:40a bad generation and the prompt and the
- 8:42images are not the problem. It's just
- 8:44the way it is. We got a bad generation.
- 8:47What can we do? Well, the only thing
- 8:48that you can do is to run it again.
- 8:51That's it. Let's see if we get it right
- 8:53on this generation. And uh that's a
- 8:56little bit weird.
- 8:58But for the most part, [music] it's
- 8:59fine.
- 9:02Since we are prompting for camera
- 9:03motion, you can and you should be
- 9:06specific. By default, if you don't
- 9:08prompt for anything, the model will
- 9:10create a slow type of movement. If you
- 9:12want more energy or maybe a different
- 9:13type of movement, you need to prompt for
- 9:15it. You can use keywords like dolly,
- 9:18pull, arc, crane or fast. I'm going to
- 9:21try it right now. I will start with the
- 9:23man, you know, the close-up and then the
- 9:25other one will end. And my prompt is
- 9:28going to be a man standing on the peak
- 9:30of a mountain we can know. The camera
- 9:32dollies out fast from the subject to
- 9:34reveal the mountain. Cool. I'm going to
- 9:37run this generation. And while this one
- 9:39is going, what if you prompt for a
- 9:41different camera motion? What we are
- 9:43doing right now is prompting the zoom in
- 9:45and the zoom out, which is obviously
- 9:47what we have on the first frame and the
- 9:48last frame. But what if I go right here
- 9:51and do something completely different?
- 9:53We still have the man on a mountain, but
- 9:54then I'm saying a smooth sweeping drone
- 9:57shot that pulls away into the clouds.
- 9:59And hopefully you can see the problem
- 10:01with this prompt. Because we provided an
- 10:03ending frame, the model is forced to
- 10:05finish on that very specific shot. If
- 10:07[music] you prompt for a completely
- 10:09different final scene, it will try to do
- 10:12it and then it will be forced to finish
- 10:14on the [music] ending frame. Depending
- 10:16on the motion that you're prompting,
- 10:18this can look great because it will
- 10:20create a transition or you will be
- 10:22getting something just super weird. It's
- 10:24just like throwing a coin. It's like
- 10:2650/50. Prompting for something crazy or
- 10:28something that maybe they model will not
- 10:31understand can give you a good result,
- 10:33but most often not it will not. [music]
- 10:36So, you need to have the extra credits.
- 10:38I'm going to run this generation and
- 10:39while this one is going, we can take a
- 10:41look at the other one. On this one, the
- 10:42camera needs to dolly [music] out fast
- 10:45to reveal the background. Let's see what
- 10:47we get back.
- 10:50It is doing that. We said that we wanted
- 10:52to reveal the vast mountain range. And
- 10:55it's doing that, but it
- 10:56>> [music]
- 10:56>> at the same time it's just failing
- 10:58because it's just not stopping on the
- 11:00ending frame.
- 11:01And to [music] finish, because again
- 11:03it's needs to be forced to finish on
- 11:05that ending frame, it will create a
- 11:07transition. And this is the transition.
- 11:11Fine.
- 11:12Not a good generation, but again not bad
- 11:15generation. This is still very usable.
- 11:17Now, let's see the other one. So, in
- 11:19theory it should be going up into the
- 11:21clouds and then at some point go back.
- 11:25Let's see.
- 11:26And notice it's a little bit weird.
- 11:28You're just getting
- 11:29>> [music]
- 11:29>> clouds.
- 11:31And then the transition, of course.
- 11:37If we run this generation like 10 times,
- 11:40I'm pretty sure that we will be getting
- 11:42maybe two that are usable and the rest
- 11:45will [music] be unusable. So, when you
- 11:47have a very specific starting frame and
- 11:49a very specific ending frame, [music]
- 11:50trying to prompt for something that
- 11:52makes sense. If you provide something
- 11:54complex, you will be getting something
- 11:56weird and a transition. This idea
- 11:58applies to any type of motion. Just to
- 12:00give you one more example, I have this
- 12:02woman and you have an idea of what I
- 12:04want to do. [music]
- 12:05I want to take this woman as a starting
- 12:07frame and then I want to push in and we
- 12:10want to see her eyes with tears. This is
- 12:13my prompt. The camera performs a fast
- 12:15dolly push in. Keywords fast [music]
- 12:18dolly push in towards the eyes of the
- 12:21woman. Again, I'm reinforcing what we
- 12:23want to do. It's pretty obvious to push
- 12:26in towards the eyes, but it's always a
- 12:28good idea if you prompt for it. I'm
- 12:31going to run this generation. While this
- 12:33one is going, we are not providing a lot
- 12:34of information of what will happen in
- 12:37between. [music] And we are not saying
- 12:39what we don't want. So, the AI might
- 12:42fill the gaps with some effects or maybe
- 12:45some movement from the subject or maybe
- 12:48something completely weird. If you want
- 12:50for the subject to stay static and the
- 12:52background to remain consistent and
- 12:54remain the same, you need to go to the
- 12:57prompt and say the subject [music] is
- 12:59completely static, the background stays
- 13:01consistent while the camera pushes in.
- 13:03We we get it and this should be a very
- 13:05simple generation. But after running a
- 13:07lot of different generations, I'm kind
- 13:09of a let's say 50% sure that we will be
- 13:12getting a weird transition. I'm going to
- 13:14go right here.
- 13:15We have the motion
- 13:17and [music] there we go. That's the
- 13:19weird part.
- 13:20Nonetheless, for the most part it's it's
- 13:23good. It's fine.
- 13:26But it's [music] adding a little bit of
- 13:27VFX.
- 13:29If you don't want for the VFX, you just
- 13:31need to say no VFX, no anything, just
- 13:34[music] keep the background consistent
- 13:35and on and on and on. So, you need to
- 13:37prompt for this.
- 13:39Fine.
- 13:40As you can see in this video, I'm trying
- 13:41to show you different examples, but I am
- 13:43also expecting to get errors. So, you
- 13:46can see where you can find some
- 13:48glitches. The next most reliable way is
- 13:51a static shot.
- 13:53You have an establishing shot, which is
- 13:56the where, and then the ending frame is
- 13:58the what will happen. As an example, I
- 14:00have this scene with an with a couch,
- 14:02just an empty couch. And on the next one
- 14:05is the complete unchanged scene, but
- 14:07then we have a woman sleeping on the
- 14:10couch. As before, we can assume what
- 14:12will happen. I want an empty couch and
- 14:14then I want the woman
- 14:16>> [music]
- 14:17>> sleeping on the couch. This is my
- 14:19prompt. A woman enters the room and
- 14:21smoothly lies down on the green couch.
- 14:24Her weight sinking into the cushions
- 14:26naturally. So, yeah, it's kind of simple
- 14:28and this is a resource that we use the
- 14:30whole time. We have the shot of
- 14:32something and then something happens on
- 14:34that something. It doesn't need to be a
- 14:36room. It could be an empty car. We have
- 14:37a shot of an empty car and then with the
- 14:39next frame is going to be someone inside
- 14:42the car. In this case, instead of a car,
- 14:44we have a room. An empty room, [music]
- 14:45someone walks in and performs an action.
- 14:48Another thing that you can do, if you
- 14:50don't care about the subject and how the
- 14:53subject is, because this is part of
- 14:55providing an ending frame. We are sure
- 14:57that she is sleeping on this very
- 15:00specific position, you can, of course,
- 15:02just remove the ending frame. And then
- 15:04when you prompt for something, the model
- 15:06will just fill the gaps for you. And I'm
- 15:08going to just run a test. I'm going to
- 15:10run this generation and in the meantime,
- 15:12we have the other one. Going to play it
- 15:13back and I'm 100% sure that this
- 15:16generation will be a good generation.
- 15:18Going to play it back.
- 15:20And it's a very simple [music]
- 15:22generation. She needs to lie down
- 15:25and
- 15:27good vibes.
- 15:34Let's go back to the beginning and
- 15:36There we go. [music] She goes in. Oh,
- 15:39nice couch. I'm going to take a nap.
- 15:43And that's it.
- 15:45Yeah, it's a pretty safe way of working
- 15:47with frames. Let's see the other one. We
- 15:50have no idea what is the woman, what
- 15:52will happen and where she's going to go.
- 15:57Nonetheless, it still is a good
- 15:58generation.
- 16:00Maybe this part
- 16:03it's not and it looks a little bit
- 16:05weird. It is just a nice touch.
- 16:07It's that she's grabbing the blanket
- 16:11from a different from here.
- 16:13And of course it fails. That's why
- 16:16providing the ending frame will always
- 16:18guarantee [music]
- 16:20that you get most likely good
- 16:23generation.
- 16:26Although again, nothing is guaranteed.
- 16:28As I mentioned before, it doesn't need
- 16:29to be a room or maybe the inside of
- 16:31something. It could be a very close shot
- 16:34of something else. On this example, I
- 16:36have this exhaust and then of course the
- 16:38car, you know, we turn on the car the
- 16:40car and we see the sparkles and the
- 16:42flame. I know that this is a little bit
- 16:44unrealistic, but I know [music] nothing
- 16:47about cars. I'm going to drop the
- 16:48starting frame and I'm going to go to
- 16:50the ending frame. And we just need to
- 16:52prompt what happens in between. [music]
- 16:54Static camera, close-up of a car
- 16:56exhaust. I'm going to run the generation
- 16:59and see if we get something cool. While
- 17:01this one is going, nothing stops you
- 17:03from using just the first frame and
- 17:06trying to be creative. Just like we did
- 17:08with the couch. But again, you can
- 17:10expect some glitches. Just to give you
- 17:12an example, I'm going to go to this
- 17:14woman and I'm going to drop the [music]
- 17:16starting frame as the starting frame.
- 17:18And I'm going to prompt for something
- 17:20way more challenging. The camera
- 17:22performs a full 360° orbit around the
- 17:25subject, [music] finishing with her
- 17:27flashing a big smile and a cool
- 17:30energetic pose. So, we are being a
- 17:32little bit more creative and let's say
- 17:34experimental and try to get something
- 17:36super dynamic. This is actually fine,
- 17:38but you need to really remember that
- 17:40doing things [music] like this will cost
- 17:42you a lot of credits. You might get it
- 17:44on the first try, but most likely you
- 17:47will need to run it multiple times until
- 17:49you get what you want. I'm going to run
- 17:51this and on the meantime, I will go to
- 17:54this one. Let's see if we get something
- 17:56cool
- 17:57or something weird.
- 17:58We get the fire and the sparkles.
- 18:02That was a little bit weird, but it's
- 18:04fine. It's usable.
- 18:06We don't need to use the whole clip.
- 18:08Sometimes you just need parts.
- 18:10Maybe we can just stop it right here and
- 18:11then move on to a different shot.
- 18:15And then we get the sparks, of course.
- 18:17Usually, almost everything that you get
- 18:19it's usable. You just need to take the
- 18:21different [music] parts and then go to
- 18:23the editor and see if you can create
- 18:25something cool. If you can't, doesn't
- 18:26matter. You can go back, create a new
- 18:28generation until you get something that
- 18:30you think it will work and then take it
- 18:32back to the editor. Let's see the 360.
- 18:34And again, this is just like flipping a
- 18:36coin.
- 18:40>> [music]
- 18:40>> Okay.
- 18:42It was actually pretty cool. We still
- 18:45got to get that
- 18:46effect.
- 18:47>> [music]
- 18:51>> It wasn't that bad. And this was the
- 18:53first generation. So, sometimes you will
- 18:55be getting on the first and sometimes
- 18:57you will need to run it multiple times
- 18:59until you get it. That's the way it
- 19:01works. Right, so this was an
- 19:03introduction to frames and flow. But
- 19:05this applies to any other platform.
- 19:07There's so much we can do right here
- 19:09like 2D animations, looping, time lapses
- 19:13and a lot more. If you want for me to
- 19:15upload videos showing you different
- 19:16examples that we can create with frames,
- 19:18let me know down in the comments.
- 19:20[music] All right, so if you liked all
- 19:21of this and you learned something new,
- 19:23remember to like, subscribe and leave a
- 19:25comment. Don't be a stranger.
- 19:35>> [music]
- 19:41[music]
About this transcript
This page contains the full transcript of How to Frames: Mastering Start and End Frames in Google Flow (2026) by AI Mind Revolution, generated from the public captions YouTube serves with the video. The transcript has 3,740 words across 543 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.