Veo 3<\/a> A good video? Learn how to talk first<\/strong><\/p>\nFirst, gotta get the basics. A well-written cue is the key to generating a good video.<\/p>\n
The more specific you can be in your cue words, and the more plain language you can use to describe them, the easier Veo 3 will be to understand and the easier it will be to generate the video you want.<\/p>\n
Veo3 Seven Elements of Cue Words<\/strong><\/p>\nMaster these 7 elements to generate professional-grade videos<\/p>\n
\n- Subject.<\/strong>\u00a0Who, or what, is in the scene? Is it a person, an animal, an object, or a landscape.<\/li>\n
- Context.<\/strong>\u00a0Where is the main body? Indoors? City streets? Or in the forest?<\/li>\n
- Action.<\/strong>\u00a0What is your subject doing? Walking, jumping, or turning his head?<\/li>\n
- Style.<\/strong>\u00a0What kind of visuals do you want? Cinematic, animated, stop-motion, and so on.<\/li>\n
- Camera motion.<\/strong>\u00a0Describe how the camera moves. Is it an aerial, flat, overhead, or low angle.<\/li>\n
- Composition.<\/strong>\u00a0How was the image composed? Is it a wide angle, close-up, or something else.<\/li>\n
- Ambiance.<\/strong>\u00a0Emotion and light. You can say \"hot-colored,\" \"blue light,\" or \"night.\"\u3002<\/li>\n<\/ul>\n
Change your cue word every time you generate it.<\/strong><\/p>\nIf you've ever played with a model like Midjourney or Flux, you probably know that running the same cue word more than once (i.e., with a different seed seed) will come out a little differently each time.<\/p>\n
Veo 3 is different.<\/p>\n
For the same cue word, even a very simple one, Veo 3 outputs very similar results.<\/p>\n
You might get someone who looks the same and wears the same clothes in about the same place.<\/p>\n
Important note: This is actually quite good when the results you're generating are a bit glitchy (like a bit of a problem with coherence or audio). You can run it again with a different seed and you'll probably get the results you want.<\/strong><\/p>\nBut if you're still exploring models and looking at possibilities, running with the same hint several times is a waste of money\u3002<\/p>\n
If you're not quite sure what you want, then start with a couple of cue words in completely different general directions. If you already know which elements you want, then make those elements more specific.<\/p>\n
for example<\/p>\n
\n- What the woman looks like (hair color, hairstyle, skin tone).<\/li>\n
- What she's wearing.<\/li>\n
- Where she is.<\/li>\n
- How she laughs.<\/li>\n
- Why is she smiling.<\/li>\n<\/ul>\n
Role consistency<\/strong><\/p>\nGenerally speaking, it's pretty hard to generate a video with a video model that doesn't have features like start frames or scene elements to try to keep the characters consistent.<\/p>\n
These are features, however, that the Veo 3 will soon have.<\/p>\n
Until then, we have other ways too.<\/p>\n
This is because one of the features of Veo 3 is that similar prompt words will generate similar characters. So, as long as you keep the character descriptions consistent across different generation tasks, you'll usually get a character that looks the same.<\/p>\n
This means that you can mess with a list of character descriptions and reuse them as is in different prompt words.<\/p>\n
Sample Role Description:<\/strong><\/p>\n\"John, a 40-year-old man with short brown hair, glasses, blue jacket, looks like he's got something to think about.\"<\/p>\n
The more unique and specific these descriptions are, the more visually coherent Veo 3 will be between the different scenes generated separately.<\/p>\n
So, you can create some role reference sheets and write down the exact words used to ensure consistency.<\/p>\n
![\"3a1ceffdj00sy506t006sd000u000b3m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/3a1ceffdj00sy506t006sd000u000b3m.jpg\" "\\"3a1ceffdj00sy506t006sd000u000b3m\\"")
<\/p>\n
How to write cues for audio<\/strong><\/p>\nBecause Veo 3 brings audio with it every time it generates a video, you'll also have to tell it what you want to hear.<\/p>\n
Consider these elements below:<\/p>\n
\n- What people say (in conversation).<\/strong>\u00a0What the people in the scene are saying.<\/li>\n
- Ambient noise (ambient sound).<\/strong>\u00a0Background noise in a scene, such as the sound of a busy street, a noisy office, a bustling caf\u00e9, etc.<\/li>\n
- Sound effects or out-of-scene sounds.<\/strong>\u00a0Like the phone ringing.<\/li>\n
- Music that may be required for the scene.<\/strong>\u00a0Things like intense movie soundtracks, upbeat pop songs, and so on.<\/li>\n<\/ul>\n
How to write dialog prompts and how to avoid subtitles<\/strong><\/p>\nWith Veo 3 you can create all kinds of interesting characters. They can talk, tell jokes, gesticulate, and sometimes act.<\/p>\n
But if you want them to talk, you have to tell it in the cue word.<\/p>\n
There are two ways to write dialog prompt words:<\/p>\n
1. Explicitly.<\/strong><\/p>\n\"A man says, \"My name is Ben.\"<\/p>\n
2. Implicitly.<\/strong><\/p>\n\"A man told us his name.\"<\/p>\n
Both ways, a video of a man talking will be generated. The first, will use the exact lines you give; the second, will let the model decide what to say, for example, the model will give you a name on its own.<\/p>\n
Note: Veo 3 does not support Chinese lines. To make the characters speak Mandarin, describe them in English:<\/strong><\/p>\nSays in Mandarin: \"[Speaks English or Chinese]\"<\/p>\n
Write your own dialog<\/strong><\/p>\nIf you're going to explicitly tell the model character what to say, try to keep the dialog short.<\/p>\n
Preferably the kind that can be said in about 8 seconds.<\/p>\n
If you try to cram too much into the dialog, the character might talk fast.<\/p>\n
If you let the character say too little, there can be awkward silences, or the character can say a bunch of AI gibberish that doesn't make much sense (like the second example below).<\/p>\n
Without clear guidance, the model can't make up all the words it needs on its own.<\/p>\n
![\"1e45c720j00sy506t0081d000u000cnm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/1e45c720j00sy506t0081d000u000cnm.jpg\" "\\"1e45c720j00sy506t0081d000u000cnm\\"")
<\/p>\n
Let Veo 3 write the lines.<\/strong><\/p>\nIf you're not good at writing conversations, it'll be a big help to use \"indicative conversations.\" And you can always record your favorite output video and use it in the new hint\u3002<\/p>\n
Here we have Veo 3 generating a video of a stand-up comedian telling a joke.<\/p>\n
For the first time, we let Veo 3 decide on the content of the jokes.<\/p>\n
For the second time, we had Veo 3 follow the joke given in our cue word.<\/p>\n
![\"0c472abcj00sy506t00add000u000bum\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/0c472abcj00sy506t00add000u000bum.jpg\" "\\"0c472abcj00sy506t00add000u000bum\\"")
<\/p>\n
You see, Veo 3 can help you fill in all the dialog as long as the prompts are given the right words and the context is given.<\/p>\n
Do you want to try out how flexible Veo 3 can be in terms of dialog? Try these prompt words below:<\/p>\n
\n- A stand-up comedian tells a joke.<\/li>\n
- Two people discussing a movie.<\/li>\n
- A man arguing on the phone.<\/li>\n
- A woman tells her life story.<\/li>\n<\/ul>\n
Don't let the subtitles get in the way.<\/strong><\/p>\nVeo 3 should train a lot of videos with embedded subtitles, so you'll often see badly spelled, wrong content subtitles in the generated results.<\/p>\n
They often ruin a generation.<\/p>\n
However, there are a few simple ways to avoid this:<\/p>\n
\n- Put the lines you want to hear behind the colon, like, \"A man says, \"I'm Ben,\" instead of \"A man says, \"I'm Ben.\"<\/li>\n
- Add \u201cno hints\u201d to the hint - in Veo 3, this negative hint works\u3002<\/li>\n
- if it doesn't work, say it again and again: \"no subtitles, no subtitles!\"\u3002<\/li>\n<\/ul>\n
Pronunciation.<\/strong><\/p>\nSometimes you'll find that models get the pronunciation of certain words wrong.<\/p>\n
The easiest way to do this is to take these words and use theSpell it out phonetically.<\/strong>.<\/p>\nPronunciation examples:<\/strong><\/p>\nOriginal language: \"Continue to read, understand fofr and Shridar's guidance on video production\"
\nAmended: \"Continue reading, understand foh-fur and Shreedar guidance on video production\"<\/p>\n
He who speaks, he who listens.<\/strong><\/p>\nWhen you're writing multiple characters' dialog in prompts, Veo 3 sometimes gets confused and doesn't know which line is spoken by whom.<\/p>\n
This is especially common when character descriptions are more similar and Veo 3 can't quite tell who's who.<\/p>\n
At this point in the cue, be specific about the speaker:<\/p>\n
Example of a multiplayer dialog:<\/strong><\/p>\n\"The woman in pink said, \"But I'm the one in pink.\"
\n\"The man with the glasses answered: No, I'm the one with the glasses.\"<\/p>\n
Music should be included in the cue<\/strong><\/p>\nAs with the rest of the video, if you want music in the scene, it's in the cue word.<\/p>\n
You can be extraordinarily articulate, describing the type, style and mood of music you want.<\/p>\n
Of course, you could be vague and let Veo 3 decide for itself.<\/p>\n
Style Control<\/strong><\/p>\nIf nothing is said, Veo 3 generates that well-produced live-action video by default.<\/p>\n
Just imagine that smooth and professional demo, commercial, or MV.<\/p>\n
If you want it not to run in that direction, you have to add to the cue wordhairstyle<\/strong>.<\/p>\nStyle control example:<\/strong><\/p>\n\u201cIn a [style name] way: a man with a beard in a flannel shirt and old jeans sits on his legs by a beating bonfire, and Amber's light falls softly and dances in the quiet forest open beneath him, with loose needles. He was standing just outside the edge of the fire, a giant grizzly bear, calm and still, with a warm light in his fur, and a fire in his eyes, and a strange intelligence. Shake hands, like old friends.\"<\/p>\n
You'll notice that not only has the overall graphic style of the video changed, but even the character movements and interactions have changed along with it.<\/p>\n
In all of these examples, the audio is pretty much the same, and we didn't cue the audio individually, so it doesn't change much between styles.<\/p>\n
![\"c543264bg00sy506s0571d000id00cxm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/c543264bg00sy506s0571d000id00cxm.gif\" "\\"c543264bg00sy506s0571d000id00cxm\\"")
<\/p>\n
camera movement<\/strong><\/p>\nAs you might expect, like other video models, the Veo 3 responds well to common camera motion cue words.<\/p>\n
Using these words, you will be able to control the actions and images in the video:<\/p>\n
\n- eye level<\/strong><\/li>\n
- high angle<\/strong><\/li>\n
- worms eye<\/strong><\/li>\n
- Dolly shot<\/strong><\/li>\n
- zoom shot<\/strong><\/li>\n
- Pan shot<\/strong><\/li>\n
- tracking shot<\/strong><\/li>\n<\/ul>\n
![\"f40dfe2dg00sy506r04mmd000hs00bwm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/f40dfe2dg00sy506r04mmd000hs00bwm.gif\" "\\"f40dfe2dg00sy506r04mmd000hs00bwm\\"")
<\/p>\n
Self-generated video: Veo 3 makes you \"for real\"<\/strong><\/p>\nThe Veo 3 is amazingly good at generating the kind of selfie videos that really look like they were taken with a cell phone.<\/p>\n
We have found that there are specific phrases that consistently unlock this effect.<\/p>\n
For example, a hint begins with a \"a self-generated video\" and is much better than a simple description of a person holding a camera\u3002<\/p>\n
in addition,Getting the arm in the frame<\/strong>It's the key to ensuring realism.<\/p>\nThe example of the gorilla is a good example of this, as it says in the hint: \"Press the camera to the length of the arm. His long and powerful arm is clearly visible.\" So the video looks more like a real selfie than a simple close-up\u3002<\/p>\n
Natural eye contact is also helpful. In the Tokyo example, \"I look at the camera from time to time, and then turn to the fun stand,\" the natural look is better than staring at the camera\u3002<\/p>\n
Here are two examples of how these techniques work:<\/p>\n
Cue word one:<\/strong><\/p>\nA self-portrait explored by a traveler in Tokyo's crowded street market. She's wearing an old cowboy jacket, full of excitement. The afternoon sun casts a beautiful shadow between the booths. She talks while she tastes street snacks, occasionally looks at the camera and turns to the fun stand. It's a bit of a granular, very cinematic. She said in an English accent, \"Well, you have to taste this in Tokyo. This octopus pellets is a terrible thing, and the stallor just told me that it's a three-generation business of their family.\u201d Finish her thumbs up\u3002<\/p>\n
Cue word two:<\/strong><\/p>\nA hand-held, self-styled shot from the perspective of a gorilla in the jungle. A giant silverback gorilla stretches the camera to the length of the arm. His long and powerful arms are clearly visible in the picture, and his face is perfect. The gorilla said, \"I'm just trying to find out if this is really working, and I'm going to send it back to TikTok, basically, it's kind of cute, and it might be deleted.\"<\/p>\n
Selfie video tips:<\/strong><\/p>\n\n- The hint begins with \"a self-generated video of...\"<\/li>\n
- Getting the arm in the frame is the key to realism<\/li>\n
- Natural eye contact is also helpful<\/li>\n
- AND THEN, \"THE IMAGE HAS A LITTLE BIT OF A GRANULAR SENSE, A VERY FILMY SENSE.\"<\/li>\n<\/ul>\n
![\"a2ff1db6g00sy506r05qbd000mw008am\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/a2ff1db6g00sy506r05qbd000mw008am.gif\" "\\"a2ff1db6g00sy506r05qbd000mw008am\\"")
<\/p>\n
THE TOKYO EXAMPLE ALSO REVEALS A LITTLE TRICK: ADDING THE \"IMAGES WITH A LITTLE BIT OF A GRAVITATIONAL SENSE, VERY CINEMATIC\" DESCRIPTION, WHICH CAN GET THE RESULTS OUT OF THE OVERLY \"CLEAN\" AI SENSE\u3002<\/p>\n
Eventually, the video will look more like a real cell phone shot.<\/p>\n
Veo 3 How do I get vertical video?<\/strong><\/p>\nCurrently, Veo 3 doesn't directly support vertical video, and can only generate 16:9 horizontal screens.<\/p>\n
However, you can make a screen video, and then use a model like Luma's Refame Vidio to \"outprint\" it\u3002<\/p>\n
Reframe Video lets you cram any video (up to 30 seconds) into it, and then it generates a new vertical video with the ratio you specify. All output is 720p.<\/p>\n
Good news: Veo 3 will soon support vertical video natively. Also, Veo in Vertex-ai \/ Aistudio does support vertical video.<\/strong><\/p>\nPhysics: Veo 3 simulation is so realistic!<\/strong><\/p>\nVeo 3 does a great job of simulating real physics effects.<\/p>\n
It keeps objects moving and interacting correctly, while also applying a variety of different styles.<\/p>\n
This means that even if the video is converted to a different art style, the model retains the natural trajectory of the object, ensuring that physics-based animations like falling, bouncing, and fluid motion remain physically correct.<\/p>\n
![\"ad1752d1g00sy506q01j0d000hs00csm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/ad1752d1g00sy506q01j0d000hs00csm.gif\" "\\"ad1752d1g00sy506q01j0d000hs00csm\\"")
<\/p>\n
Video quality: Pull up to 4K 60fps with one click<\/strong><\/p>\nVeo 3 outputs 1280p x 720p video by default.<\/p>\n
\ud83d\ude80 Recommended Tool: We recommend using Topaz Lab's Video Upscaler to take your video straight to 4K resolution and still nail 60 frames per second.<\/strong><\/p>\nX-Platform Search Tips<\/strong><\/p>\nAlso, you can type Veo 3 Prompt on X to check out all the great Veo3 prompts.<\/p>\n
![\"d5c40e3fj00sy506r002kd000p700fvm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/d5c40e3fj00sy506r002kd000p700fvm.jpg\" "\\"d5c40e3fj00sy506r002kd000p700fvm\\"")
<\/p>\n
Some are still written in Jason format, which is usually more controlled and accurate.<\/p>\n
![\"5aec5526j00sy506t00ccd000jy00j5m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/5aec5526j00sy506t00ccd000jy00j5m.jpg\" "\\"5aec5526j00sy506t00ccd000jy00j5m\\"")
<\/p>\n
Summary of core points<\/strong><\/p>\nThe difference between a bland video, and a stunning video, is your cue word.
\nWith Veo 3, you're not just describing what's happening, you'reDirecting a scene<\/strong>.
\nHigh-quality video takes the subject, the background, the action, the shots, the audio, and the atmosphere and layers them on top of each other.
\nThink like a filmmaker and Veo 3 will follow your lead.<\/p>","protected":false},"excerpt":{"rendered":"This time we look at the free access to the gemini pro account and the Veo3 tips. The use of Veo3 requires at least a gemini pro account. In Google Flow, 1,000 credits per month is provided, and every Veo3 video consumes 100 credits. How do you get the Gemini Pro account free of charge? How do you get the gemini pro account?The educational benefits link: https:\/\/gemini.google\/students\/(gemini pro 15 months free)? Temporary<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[149,144],"tags":[5321,436,6976,5592],"collection":[],"class_list":["post-37968","post","type-post","status-publish","format-standard","hentry","category-jiaocheng","category-baike","tag-ai","tag-gemini","tag-veo-3","tag-5592"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/37968","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=37968"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/37968\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=37968"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=37968"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=37968"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=37968"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"d3c3ca58j00sy506t005bd000u000k2m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/d3c3ca58j00sy506t005bd000u000k2m.jpg\" "\\"d3c3ca58j00sy506t005bd000u000k2m\\"")
<\/p>\n![\"01404451j00sy506u00lrd000u000k9m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/01404451j00sy506u00lrd000u000k9m.jpg\" "\\"01404451j00sy506u00lrd000u000k9m\\"")
<\/p>\n![\"bff84f26j00sy506t003ld000u000l3m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/bff84f26j00sy506t003ld000u000l3m.jpg\" "\\"bff84f26j00sy506t003ld000u000l3m\\"")
<\/p>\n![\"e357489cj00sy506u0092d000u000kfm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/e357489cj00sy506u0092d000u000kfm.jpg\" "\\"e357489cj00sy506u0092d000u000kfm\\"")
<\/p>\n![\"21749d20j00sy506t003qd000u000mom\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/21749d20j00sy506t003qd000u000mom.jpg\" "\\"21749d20j00sy506t003qd000u000mom\\"")
<\/p>\n![\"f1df93c4j00sy506t0037d000u000omm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/f1df93c4j00sy506t0037d000u000omm.jpg\" "\\"f1df93c4j00sy506t0037d000u000omm\\"")
<\/p>\n![\"73ea758cj00sy506s002cd000u000ehm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/73ea758cj00sy506s002cd000u000ehm.jpg\" "\\"73ea758cj00sy506s002cd000u000ehm\\"")
<\/p>\n![\"81212a5ej00sy506s003cd000u000g1m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/81212a5ej00sy506s003cd000u000g1m.jpg\" "\\"81212a5ej00sy506s003cd000u000g1m\\"")
<\/p>\n![\"951575d4j00sy506r001xd000pi00fgm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/951575d4j00sy506r001xd000pi00fgm.jpg\" "\\"951575d4j00sy506r001xd000pi00fgm\\"")
<\/p>\n![\"524d8610j00sy506u00hdd000u000lmm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/524d8610j00sy506u00hdd000u000lmm.jpg\" "\\"524d8610j00sy506u00hdd000u000lmm\\"")
<\/p>\n![\"c6bb7383j00sy506u00r9d000u000n5m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/c6bb7383j00sy506u00r9d000u000n5m.jpg\" "\\"c6bb7383j00sy506u00r9d000u000n5m\\"")
<\/p>\n![\"201a587ej00sy506t005wd000u000ixm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/201a587ej00sy506t005wd000u000ixm.jpg\" "\\"201a587ej00sy506t005wd000u000ixm\\"")
<\/p>\n![\"7e215b56j00sy506u00rid000u000lrm\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/7e215b56j00sy506u00rid000u000lrm.jpg\" "\\"7e215b56j00sy506u00rid000u000lrm\\"")
<\/p>\n![\"6783f2f6j00sy506u00rkd000u000jom\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/6783f2f6j00sy506u00rkd000u000jom.jpg\" "\\"6783f2f6j00sy506u00rkd000u000jom\\"")
<\/p>\n![\"3b12908fj00sy506v00jdd000u00209m\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/3b12908fj00sy506v00jdd000u00209m.jpg\" "\\"3b12908fj00sy506v00jdd000u00209m\\"")
<\/p>\n