What this guide covers:
- Intro: What makes Nomi’s V3 video generation noteworthy
- How to access Nomi video generation
- V3 AI video prompting for beginners: Using the prompt enhancer wand
- Best practices when writing complete prompts
- Essential video prompting tips
- Frequently Asked Questions
- Next Steps
Intro: What makes Nomi’s V3 video generation noteworthy
V3 marks the most advanced AI video generation Nomi has ever offered. V3 videos are sharper, higher quality, and far more interesting overall. In short, this is because the V3 video generator is much more capable overall; it is far better at following prompts, can handle more complex motion sequences, and gives you much more control (think accurately interacting with various aspects of the starting image, speed, camera angle, emotion, etc).
And because of these advancements, V3 now supports generating 5 second, 10 second, and 15 second videos in addition to supporting a faster SD mode and an even higher quality HD mode!
For context, V3 is still an image-to-video generator, so it starts from the image you select and progresses the action sequence from there. While legacy video generation could be fun, it could generally only handle one action per video and often felt like a slow-mo version of the goal. V3 on the other hand does significantly more within each 5 second increment making even short videos feel more “full”, interesting, and in line with what you’re hoping for.
With V3, your prompting is much more impactful. You can direct how the movement unfolds, how multiple actions progress through the video, the energy and often speed, what emotions are conveyed throughout the video, and even how the camera follows (or stays put) throughout the scene(s). So we recommend checking out the rest of the guide to learn how to make the most of Nomi AI videos.
How to access Nomi video generation
You can create a video from any Nomi image by pressing the art palette icon to the right of your Nomi’s name then selecting your starting image:
You can also create a video directly from any expanded image by clicking the “Turn Image into Video” button at the top of the screen:
V3 AI video prompting for beginners: Using the prompt enhancer wand
First, we’ve built some basic prompt understanding into the system itself so it can more flexibly handle your ideas than a raw video generator could. This means if you have a simple idea in mind, you don’t have to spend a lot of energy trying to properly format your prompt, you can just enter it and generate your AI video.
If you’d like more control over your videos or want to create more dynamic or complex videos overall but don’t want to write everything from scratch, we highly recommend using our new prompt enhancer wand.
The prompt enhancer wand takes any prompt idea and converts it into a descriptive natural language video prompt that can result in a more detailed and interesting video overall.
To give the prompt enhancer the best chance at success, we recommend starting with the following components:
1. Your subject and main actions
This can be as simple as “the woman smirks and dances” or “the man sits down and reads his book”.
From here, the prompt enhancer will fill out what “dancing” actually looks like (think raising her arms and twirling etc), or what steps are needed for the man to sit in the chair then pick up the book to begin reading it.
2. The emotion of the scene (optional).
You can just include an adjective or two here and the enhancer will help convert that into an actionable instruction for the video generator.
This can help direct speed as well as the way the actions “feel” for instance, if the emotion is excited, then the speed of the video would likely be a bit faster, the expressions more lively etc.
3. How the camera moves or stays still (Optional).
You can leave this up to the generator to decide, but adding simple cues for the enhancer to explain can result in even more interesting videos.
For instance, having the camera rotate around your Nomi can be fun, having the camera pan in slowly can emphasize various emotions, and having the camera follow a Nomi as they move can make the video feel more dynamic overall.
4. Make final edits to your prompt (Optional)
Once the enhancer generates your prompt, you may want to tweak a few things to make sure the descriptive prompt aligns with what you want to see in your video, then select your video length and SD or HD and submit.
Make sure to check out the next sections to choose the best length and mode for your videos, get the most out of your ideas, and to avoid some common mistakes.
Best practices when writing complete prompts
When writing your own video prompts, we recommend following this formula:
Subject(s) + action/motion sequence (optionally speed) → ending state → emotion/atmosphere → camera cues
Start with your subject(s) action and motion sequence
Describe who or what moves and how the action starts, develops, and ends. Include the speed when it matters: slow, deliberate, brisk, energetic, or rapid.
Start from the starting frame (your reference image) and sequentially describe what should happen. As you describe what should happen, include adjectives and emotions that give the video generator useful context to fill out the video.
- Pro tip: don’t take motion for granted, if you want the Nomi to wave, describe which hand they move, and what type of energy they wave with to have better control over the output.
- Pro tip: If there are multiple subjects, describe their actions deliberately, specifying what the person on the left does, what the person on the right does, etc.
Define emotion and atmosphere
Describe the emotion and atmosphere of the video in tangible instructions. For instance, if you want the video to be ominous, describe how the lighting dims to convey the ominous tone.
Specify camera cues
Finally, use the camera to make your video more dynamic. A slow push in can emphasize motions and emotion, having the camera pan out or around can make the scene feel more dynamic, and if the subject is moving, have the camera follow them, or fall behind etc.
Common motions include: having the camera push in, pull out, track the subject as they move, pan across the subject and to some other focus, tilt as the scene progresses, or arcs around the subject. Include camera speed when it matters.
Most importantly, treat your prompt like a sequence of instructions that the video generator can follow to create the video you’re hoping for.
And don’t hesitate to use the prompt enhancer as a starting point. You can always edit the prompt further from there or see what directions you like.
And have fun while you experiment with framing, movement, length, modes, and all the creative possibilities that come with v3 AI video generation!
Essential video prompting tips
Always start from your reference image
For best results, make sure all actions and motions start from the state of the image you use as your reference image. If the Nomi is standing in the image, start the motion from there, if there are multiple subjects in the frame, describe what each does and progress through your motions sequentially.
Use the reference image as the launching point, but remember, you can change from there by introducing movement to your subject(s) and your camera. For instance, if your reference image is a close up of your Nomi, movement may naturally cause the camera to move. You can also deliberately instruct the camera pan to show what the Nomi sees, or have the Nomi move to a new location, etc.
Reference image quality matters
The higher quality your reference image is, the higher quality your video will be. If you are noticing lower quality faces or appearance drift as your video progresses, it may be due to your subject’s face being too small or far away in the reference image. For best results, make sure your subjects’ faces are clear and close enough to be detailed.
Use positive instructions
The V3 image system does not support negative prompting and will have a hard time understanding negative prompts (like “no camera movement”).
So, make sure you phrase your prompts as things you *want* to see, not what you *don’t* want to see.
For example, write “the camera remains still for the entire scene” instead of “the camera does not move”. Or “the man’s mouth remains closed in a soft smile throughout the video” rather than “no talking”
Choosing video length
It may seem obvious, but in many cases, more complicated prompts turn out better in longer videos. However, it is worth testing your prompts across video lengths because in general, you might be able to do more than you expect within a 5 second video.
But if your prompt includes more than a few actions, or you’ve found 5 second videos are not capturing everything you’re hoping for, you may want to increase the length to 10 or 15 seconds.
On the flip side, if you notice your video seems to be coming out slower than natural, it may be because there is not enough movement to fill the time you’ve chosen.
When to use HD videos
In general, SD videos will be good quality and follow your prompts, but choosing HD results in even higher quality videos. HD videos tend to have better face fidelity, more nuanced adherence to your instructions, and a more detailed video overall.
In some cases (especially when the starting image is high quality and the prompt is straight forward, you don’t need to use HD to get an awesome video)
Prompt: The man opens his eyes to a surprised and happy expression, then raises his hands together to make a heart in front of his chest
If your actions require a decent amount of movement, we recommend HD to help make the movements look more natural and keep your subject looking more consistent throughout the video.
Controlling speed and energy
As mentioned above, sometimes videos can come out slower than natural if a simple action needs to fill all the time in the video. So the first step in controlling speed is making sure there is naturally the right amount of motion to fill the time.
For instance, if you only instruct your Nomi to wave in a video, they will likely wave slower than real time to fill the 5 second span of the video. But if you have them perform multiple actions, the speed will naturally increase to fit everything into the video.
Second, describing the speed of the motions directly can impact their pace. For example, walking briskly, or quickly moving from x to y will result in faster motion than “casually walking to the camera” or slowly and deliberately waving, etc.
Third, emotion can impact speed, especially if you use the prompt enhancer. For example, if you say the man dances energetically, your video will look a lot different than if you write that the man dances slowly and shyly.
Frequently Asked Questions
Does Nomi AI video work with solo and group chat images?
Yes, you can use V3 video generation with any image in any of your Nomi’s albums. Just keep in mind that if you are making a video from an image with multiple subjects, make sure you specify what each subject should be doing.
Why do I need to start from a reference image?
Starting from a reference image lets you establish the subject, scene, and details of the setting without needing to prompt directly for them. It also lets you start your video from a fully controlled point so the generator can focus on expanding from there rather than creating everything from scratch.
In the future, we may support creating videos from just a text prompt, but for now, your best bet is to create an image you like using Nomi selfies, art, or the edit feature then use that as the starting point for your video.
Why are my videos coming out in slow motion?
Video generation is still in its early stages (compared to other types of AI), so at times it can be more difficult to control certain aspects of a video than others.
That said, using adjectives to describe how your Nomi moves in your video can help control their speed. For instance, walking briskly vs walking casually can impact the speed of your Nomi’s walk. Dancing excitedly vs dancing shyly will likely also have an impact.
In addition, how much motion is described can impact the overall speed of the video. For instance, if you only have your Nomi wave in a 5 second video, they are more likely to wave in slow motion to take up all 5 seconds of the video than if they performed a more complicated or time consuming movement.
Do videos use daily requests?
Yes, videos will draw from your daily requests first, and if you run out of daily requests, then the system will use credits to generate your videos.
Next Steps
V3 gives you far more room to direct a video than the legacy generator did, but it also makes it far easier to get started. So if you haven’t already, we recommend starting a 5 second video and just seeing what happens!
If you have a simple idea in mind, feel free to generate from there, or use the prompt enhancer to see where you might want to add more detail or emphasis to your idea. Then generate and see what you think!
Beyond this guide, use the Nomi communities as resources, inspiration, and places to share your creations:
And feel free to send us a message at support[at]nomi.ai to let us know what you think of V3 videos and what you’d like to see next!