# How to Make AI Videos: A Step-by-Step Guide for 2026

Canonical URL: https://promptwise.com/blog/how-to-make-ai-videos/

1. [Home](https://promptwise.com/)
2. [Blog](https://promptwise.com/blog/)
3. How to Make AI Videos That Look Planned, Not Random

# How to Make AI Videos That Look Planned, Not Random

By [Daniel Riley](https://promptwise.com/about/) · Published 24 October 2026

**In short:** To make an AI video that looks planned, decide what it is for, pick a model for the shot (Seedance 2.5 is the best all-round start), set up a start frame, then write the prompt, review, add sound and edit. This guide covers each step with prompts to copy and a 20-second worked product ad. Every step works in PromptWise with the same 40+ image, video and audio models.

Most first attempts at AI video go the same way. Somebody types a sentence about a dog on a beach, picks whatever model the app suggests, presses Generate and gets eight seconds of something that moves. The dog has the right number of legs, mostly. The camera floats about for no reason. Nothing much happens, and there is no obvious way to turn it into a video anyone would actually post.

Making AI videos that look planned is a different job, and it is closer to making a short film than to typing a search. You decide what the video is for, choose a model that suits the shot, set up the first frame, write the shot as a director would describe it, then review, fix, add sound and cut it together. None of these steps is hard. Skipping them is what makes AI video look like AI video.

This guide walks through that whole process in order, using the tools in PromptWise. It covers which model to start with for which job (with clear opinions), the settings that matter, prompts you can copy word for word, a worked example of a 20-second product ad, and a troubleshooting section for when clips go wrong. Everything here works with the 40+ image, video and audio models in PromptWise, including Seedance 2.5, Kling 3.0, Veo 3.1 and Nano Banana Pro.

[Start creating](https://app.promptwise.com/sign-up)

## The short answer

If you only read one part, read this. To make an AI video:

1. Decide where the video will be posted, how long it should be and what one thing it needs to show.
2. Pick a video model for the job. Seedance 2.5 is the best all-round starting point.
3. Make a start frame with an image model such as Nano Banana Pro, so you approve the look before you pay for motion.
4. Write a video prompt that describes one shot: the camera, the action in order, the light and the mood.
5. Set the aspect ratio, resolution and length, then generate two or three takes.
6. Add voice, music or sound effects, cut the best takes together and export.

The rest of this guide explains each step properly, with the reasons behind it. That matters, because when a clip fails you need to know which step to go back to.

## What you need before you start

You need surprisingly little. An idea, a rough sense of where the video will live, and an account with access to a few video models is enough to begin. A paid editing suite is useful at the end but not essential for short social clips.

What helps most is a short written plan. It doesn’t need to be a full script. Three lines will do: who the video is for, what it should make them feel or do, and the single image they should remember. “Coffee drinkers on Instagram. Make them want a flat white on a cold morning. Steam rising off a cup by a frosted window.” That is already more direction than most AI videos ever get, and it will shape every prompt you write later.

It also helps to accept one thing early. AI video models produce short clips, not finished films. Seedance 2.5 and Kling 3.0 Motion Control make clips of up to 30 seconds in PromptWise, but most shots are easier to control at around 5 to 10 seconds. A 45-second ad is therefore several clips edited together. Planning in shots from the start saves a lot of credits and frustration.

## Text to video or image to video?

There are two ways to make an AI video, and choosing between them is the first real decision.

Text to video means you describe the shot in words and the model invents everything: the subject, the setting, the framing, the light and the motion. It is quick and it is the best way to explore ideas, because you can try ten very different concepts in the time it takes to set up one properly. The weakness is control. The model decides what your character looks like, and it will decide differently every time.

Image to video means you first make (or upload) a still image, then use it as the start frame and describe only what should move. The model already knows the composition, the faces, the product and the colours, so your prompt can focus on camera and action. It takes a few more minutes, but the results are far more predictable, and it is the only sensible route when the same person or product has to appear in several clips.

The recommendation here is simple. Use text to video while you are working out the idea, then switch to image to video for anything you intend to publish. PromptWise supports both from the same screen: the video settings bar has a Start / End Frame option, so you can set the opening image and, where you want a specific ending, a closing one too.

## Step 1: Decide what the video is for

There are three decisions you must make about your video before you consider choosing a model or writing a prompt: the platform on which it will appear, the length of the video and the video’s purpose, or job.

First decide on the platform the video will be used on, and think about the shape that it will take. Most vertical videos are made in 9:16 for phone screens (TikTok, Instagram Reels, YouTube Shorts and Stories ads), while videos for YouTube, websites and presentations are best made in 16:9 widescreen. Square 1:1 or tall 4:5 videos are also well suited to social feeds. Choose the aspect ratio before you make anything, the start frame included, because the model will compose the shot for the exact frame you specify. If you generate a 16:9 video and then crop it to 9:16, most of the picture will be thrown away, and heads are often cut off at the edges.

How long is your video? A 6-second bumper might consist of only one shot, but a 15-second video can be made up of two or three shots, and a 30-second ad can contain four to six. Plan each shot before you generate anything, listing them out as simple one-line statements.

The purpose of your video will also help to determine the shots you need. A product advert should show the product correctly in at least one of the shots. A brand mood piece should look consistent throughout. An explainer video needs space in the shots for the text to be read and the voiceover to be heard.

## Step 2: Pick a video model

PromptWise has 20 video models, which is wonderful once you know them and a little overwhelming on day one. The good news is that most people only need three or four. Here is where to start, and why.

**Seedance 2.5 is the default.** If you are not sure, start here. The app describes the Seedance family as “cinematic video creation with sound”, and Seedance 2.5 is the model PromptWise gives full access to on every plan. In the app it makes clips of 4 to 30 seconds, and it generates sound along with the picture, so one generation can cover a whole short social clip, audio and all. It handles scenes with a couple of beats well, so it is a good first choice for ads, social clips and short story moments.

**Kling 3.0 is the pick for people and motion.** When a shot depends on a person walking, turning, dancing or reacting, Kling 3.0 is the one to try first. It makes clips of 3 to 15 seconds at up to 4K. Kling 3.0 Turbo makes clips of the same length and is the sensible way to draft a Kling shot before paying for the full version. When you need to guide exactly how a character moves, Kling 3.0 Motion Control is built for that, with clips of up to 30 seconds.

**Veo 3.1 is for shots where the sound has to be exact.** Seedance 2.5 makes sound too, so that alone isn’t a reason to switch. PromptWise describes Veo as “precision video with sound control”, though, and when a door has to slam on the right frame or a short spoken line has to match the lips, Veo 3.1 is where to go. It comes in three versions, Veo 3.1 Quality, Veo 3.1 Fast and Veo 3.1 Lite, so you can sort out framing and timing on a lighter version before a final Quality run.

**Lighter models are for drafts.** Seedance 2.0 Fast and Seedance 2.0 Mini make 720p clips of 4 to 15 seconds and are good for testing whether a prompt works at all. Keep the prompt identical when you move up to the final model, so the test actually tells you something.

The other models are worth knowing about once the basics feel easy. Hailuo 2.3 Pro is described in the app as “high-dynamic, VFX-ready cinematic video” and is a fun one for energetic, stylised shots. Seedance 2.0 makes clips at up to 4K. Gemini Omni Flash and Gemini Omni Flash 1.1 cover “multimodal video generation and editing”. Grok, Flux 3 Video, LTX, MiniMax H3, Seedance 1.5 Pro, Kling Omni and Kling 2.5 Turbo each read prompts a little differently, which makes them useful for getting a second opinion on the same shot.

| Job | Start with | Draft on | Why |
| --- | --- | --- | --- |
| General social clip or ad | Seedance 2.5 | Seedance 2.0 Fast | Long clips with sound, handles a few beats |
| People, faces, performance | Kling 3.0 | Kling 3.0 Turbo | Natural motion, 3 to 15 seconds |
| Directed movement, dance, gestures | Kling 3.0 Motion Control | Kling 3.0 Turbo | Built for guided motion, up to 30 seconds |
| Dialogue or effects timed to the action | Veo 3.1 Quality | Veo 3.1 Fast or Lite | Precise sound control in the same generation |
| Stylised, high-energy effects | Hailuo 2.3 Pro | Seedance 2.0 Mini | Made for dynamic, VFX-style shots |
| A second opinion | Kling Omni, Grok, LTX | Your usual draft model | Different readings of the same prompt |

One honest caveat: these are starting points based on how the models are positioned and how they behave in practice, not a lab benchmark. The best way to choose is to run one prompt on two or three models and keep the take you like. Because every PromptWise plan includes every model, that comparison costs only credits, not another subscription. A full head-to-head of Seedance, Kling and Veo deserves its own article, so this one stays practical.

## Step 3: Make the start frame

Begin with a great still frame to start your video. This frame can then be animated by the video model, and it is much easier to add motion to a frame that has already been chosen than to invent an entire video.

Images also cost less than video to generate, so it makes sense to experiment with stills first and save video generations for frames you already like. Current credit costs for each model are on [PromptWise’s pricing page](https://promptwise.com/models-and-pricing/).

**Nano Banana Pro is the best default for start frames.** It is Google’s image model, it goes up to 4K in PromptWise, and it follows detailed descriptions well. For quick variations, Nano Banana 2 and Nano Banana 2 Lite are faster. Seedream 5.0 Pro and GPT Image 2.5 Sunburst are strong alternatives when you want a different look from the same prompt, and Realism Pro (Soul 2) is worth a try for photorealistic people. The [Nano Banana Pro page](https://promptwise.com/models/nano-banana-pro/) covers settings and pricing in more detail.

Describe the image to someone who cannot see it. Write natural sentences that cover the subject, where it is, the lighting, how close the camera is and the kind of image it should be, such as an illustration or a realistic photograph. Rather than entering a long string of keywords for the start frame of a coffee ad, try writing a few natural sentences such as:

Prompt

Copy

```
A close-up photograph of a white ceramic cup of flat white coffee on a wooden windowsill on a cold winter morning. The window behind it is frosted at the edges, and a blurred snowy street is visible through the glass. Soft blue daylight comes from the window, and a warm lamp glows from the right. Gentle steam rises from the cup. The image is vertical, realistic and shot with a shallow depth of field.
```

Set up the still frame to match the aspect ratio of the video you want to create (9:16 for this coffee ad). Generate a few versions, then pick the cleanest composition with the most room for motion. The steam rising from a coffee cup needs somewhere to rise to, and a person walking needs space to walk. Images packed to the edges of the frame give the video model nowhere to go.

If you are animating a real product, start from a good photo of it instead. The [AI product photography guide](https://promptwise.com/blog/ai-product-photography-replace-photoshoots/) covers how to build clean product stills, and the [guide to AI image prompts](https://promptwise.com/blog/how-to-write-ai-image-prompts-that-actually-work/) goes much deeper on lighting, lenses and composition.

## Step 4: Write the video prompt

A video prompt is not a caption, and there is a simple formula that gives much better results than one: answer these five questions, as cleanly as possible and in the order listed.

First, how is the camera placed and how does it move? Write down the shot size (close-up, medium shot, wide shot) and the camera movement (slow push-in, tracking shot to the left, gentle orbit and so on). The camera can stay locked-off if there should be no movement and you don’t want the model to add any. If you leave the camera out, the model will usually add a slow aimless pan or zoom of its own.

Second, what is the action in the scene and how does it play out? This forms the body of the prompt and describes a short sequence of actions (for example, “She lifts the cup, pauses, then takes a sip”). A prompt that simply states a scene (for example, “A woman drinking coffee”) gives the model nothing to do.

Third, the shot’s location. Keep this very short if you are working with a start frame, since the image already shows the setting. Working from text alone? Then describe the place, the time of day and the weather.

Fourth, the light. Say where the light comes from and the feeling it conveys. A low sun behind a subject is different from cold daylight through a window. Neon from a sign high above is different again. The light in your prompt is what makes a clip feel filmed rather than rendered.

Fifth, add a sentence describing the mood and pace (calm and unhurried, or quick and energetic). If the model makes sound, finish with a sentence describing what should be heard.

Here is an image-to-video prompt for the coffee frame from Step 3:

Prompt

Copy

```
Close-up, with a very slow push-in towards the cup. Steam curls upward from the coffee and drifts across the frosted glass. After a moment, a hand enters from the right, wraps around the cup and lifts it gently out of frame. The warm lamp light flickers slightly. The mood is quiet and cosy, and the pace is slow.
```

This leaves out a number of things you might expect to see in a video prompt. For example, it does not re-describe the cup, the windowsill or the snow, because the start frame already contains them. Repeating the whole scene in the video prompt often results in the model changing something you were happy with.

Two habits make prompts much more reliable. Keep each clip to one main action and one camera move, because asking for a walk, a turn, a wave and a zoom in five seconds is more than any model handles cleanly. And write in plain sentences rather than lists of adjectives. “Cinematic, 8K, ultra-detailed, masterpiece” adds nothing a modern model needs. There is a full breakdown of camera, lighting and timing vocabulary in the [cinematic AI video prompts guide](https://promptwise.com/blog/cinematic-ai-video-prompts/), and the habits that waste credits are covered in [5 AI prompt mistakes killing your results](https://promptwise.com/blog/5-ai-prompt-mistakes-killing-your-results/).

[Start creating](https://app.promptwise.com/sign-up)

## Step 5: Set it up and generate in PromptWise

The following is a step-by-step guide to generating your first video in the app, starting from a completely blank screen.

To create a video, open Studio and switch to video mode. Then select a family of models (Seedance, Kling, Veo and so on) from the model picker on the long bar at the bottom of the screen, and select a specific model from that family. The model picker menu displays the maximum resolution and the clip length for each model. For example, Seedance 2.5 creates clips of 4 to 30 seconds, while Kling 3.0 creates clips of 3 to 15 seconds.

Next to the model picker is the Start / End Frame section. You can select the image you created in Step 3 as the start frame. If needed, you can also select an end frame (for example, for a video of a product arriving in a finished position).

Next to the start and end frame option is a button to set the video format (shown as something like “16:9 / 720p / 8s”). It displays the current settings, and these can be changed within the model’s range. It’s best to keep the resolution low and the length short when creating a draft. The highest settings should be used for the final version of a take.

After this you can enter your prompt. Wise, the PromptWise AI assistant, is also on hand to help develop your idea and write the prompt with you.

Note: the Generate button displays the number of credits a generation will use before you press it. Credit costs vary depending on the model, length and resolution of the video. If a generation fails, its credits are refunded automatically.

Generating two or three takes for a prompt, rather than relying on a single take, is highly beneficial. The second or third take often ends up being the best for a shot, and you can really get a sense of what the prompt is asking for by comparing the different takes. Depending on your plan, several generations can run at the same time (Starter: 2, Plus: 4, Pro: 6, Max: 8), so you don’t have to wait for one generation to finish before starting the next.

## Step 6: Review, fix and regenerate

Watch every take at full size and at least twice. The first time, watch it as a viewer: does it feel right? The second time, watch it as an editor and look for the specific faults that will give it away. Hands that merge into objects. Faces that shift slightly between frames. Text that melts. Objects that pass through each other. A camera that changes its mind halfway through.

When a take is close but not right, change one thing and generate again. If the camera moves too much, adjust the camera sentence only. If the action is rushed, simplify the action or lengthen the clip. Changing three things at once can fix the problem, but you will not know which change did it, and you will be guessing again on the next shot.

Some faults are much easier to fix in the start frame than in the video. If a face looks wrong in every take, the face in the still is usually the cause. If the composition feels cramped, regenerate the still with more space. Going back one step costs a couple of credits and saves many more.

Know when to stop, too. If five takes of the same shot keep failing in the same way, the shot is probably asking for too much. Split it into two simpler clips, or try a different model. A shot that Kling 3.0 struggles with may work first time on Seedance 2.5, and the other way round.

## Step 7: Add voice, music and sound

Silent AI video feels unfinished, even on platforms where most people watch with the sound off. There are two ways to add audio.

The first is to generate it with the picture. Seedance 2.5 does this, and so does Veo 3.1. For most clips Seedance is enough: a long take comes back with the sound already on it. If the timing is fussy, say a door that has to slam on cue, use Veo 3.1, which the app sells on “sound control”. Either way, finish your prompt with a sentence about what the viewer should hear: “The sound of rain on the window, a kettle clicking off in the background and soft café chatter.” Keep any spoken line short and put it in quotation marks.

The second is to make the audio separately with the PromptWise audio models, which gives you more control over each layer. For a voiceover, ElevenLabs Multilingual V2 is described as “ultra-realistic multilingual text-to-speech”, and Seed Audio 1.0 Speech makes “narration and dialogue from a script”. You pick a voice from the menu, paste your script and generate. For music, CassetteAI Music and Seed Audio 1.0 Music both work from a description of the genre, mood and tempo or instruments. A prompt as simple as this one works well:

Prompt

Copy

```
A warm, slow acoustic guitar piece with soft piano, gentle and hopeful, for a cosy winter morning coffee advert.
```

For sound effects and ambience, try ElevenLabs Sound Effects or Seed Audio 1.0 Sound Effects, again described in plain words: “a ceramic cup set down on a wooden sill, with a quiet room tone and distant traffic.”

Write the voiceover script before you finalise the shots. A 15-second ad has room for about 30 to 40 spoken words at a natural pace, and if the script runs long you will need longer clips or fewer words. Getting the timing right on paper is far quicker than re-generating video to fit.

## Step 8: Edit, export and post

When you’ve picked out the best versions of your shots and audio and downloaded them from PromptWise, it’s time to assemble them in an editor of your choice. This could be CapCut, Premiere Pro, DaVinci Resolve or even the video editor on your phone.

To polish an AI-generated video in the editor, trim the very start and end off each shot, as that is often where the motion is weakest. Then cut on movement, so if one shot ends with a hand picking up a cup, the next shot starts in the middle of the hand lifting it. Keeping a single colour look throughout is often all the colour grade a video of this type needs. Adding captions for viewers who watch social video with the sound off is a good idea as well. Just remember to add any text in the edit, and avoid asking the video model to render text for you.

Export your video at the same resolution you created it in, and in the aspect ratio of the platform where you intend to publish it (for example, 9:16 for Instagram Stories). Everything you create in PromptWise carries a full commercial AI licence, so you can use your videos in ads and client work and publish them on the social platforms of your choice.

## Ten AI video prompts to copy

These are written for image to video, so each assumes you have already made a matching start frame. They also work as text-to-video prompts if you add one or two sentences describing the setting. Each one sticks to a single shot, a single camera move and a clear order of events.

**Product reveal.** Medium close-up with a slow orbit to the right around a perfume bottle on a black stone plinth. A thin ribbon of mist drifts across the bottle from left to right. Halfway through, a narrow beam of warm light sweeps across the glass and catches the label. The mood is luxurious and still, and the pace is slow.

**Food close-up.** Extreme close-up, locked-off camera. Honey is poured slowly onto a stack of pancakes from above the frame. It spreads over the top pancake and begins to run down the sides. Soft morning light comes from a window behind. The mood is warm and indulgent.

**Lifestyle walk.** Wide shot with a steady tracking move to the left. A woman in a long camel coat walks along a quiet London street of white terraced houses, glancing at her phone and then looking up and smiling. Low autumn sun from behind her creates a soft rim light. The pace is relaxed and natural.

**Travel establishing shot.** Very wide aerial shot, slowly gliding forward over a turquoise bay at sunrise. Small fishing boats rock gently on the water, and the first light catches the cliffs on the right. Thin mist sits over the sea. The mood is calm and expansive.

**Fitness.** Medium shot from a low angle, with a handheld camera that follows the action. A man finishes a set of push-ups on a rooftop at dawn, stands up, breathes out and looks out over the city. Cold blue light with a warm sunrise on the horizon. The pace is energetic, then settles.

**Interior or property.** Wide shot with a slow push-in through an open doorway into a bright Scandinavian living room. Sheer curtains move slightly in a breeze. Sunlight falls across a pale wooden floor and a green sofa. The mood is airy and peaceful.

**Fashion detail.** Close-up, locked-off camera. A hand runs slowly along the sleeve of a chunky cream knit jumper, and the fibres catch the light. Soft, even studio light with a gentle shadow on the right. Quiet and tactile.

**Music video moment.** Medium shot with a slow dolly-out. A singer stands in a dark warehouse under a single overhead light and slowly lifts her head towards the camera as dust drifts through the beam. Deep shadows and a slight blue tint. The mood is moody and dramatic.

**Explainer background.** Wide shot, locked-off camera. A clean, minimal desk scene with a plant, a closed laptop and a mug. A slow shaft of sunlight moves across the desk over the length of the clip. Plenty of empty space on the left for on-screen text. Calm and simple.

**Nature.** Close-up with a very slow push-in on a single raindrop hanging from the tip of a fern leaf in a forest. The drop swells, trembles and falls out of frame. Soft green light, misty background. The sound is gentle rain and birdsong.

If you use one of these on a model that generates sound, keep or add the final sound sentence. If not, leave it off, because it won’t add anything.

## Worked example: a 20-second product ad from scratch

This is how the workflow is applied to a real ad project: a 20-second vertical ad for a reusable water bottle, for Instagram Reels.

You start with the ad plan. The target is commuters and gym-goers, so the ad should look fresh and sporty. The one image to remember is cold water poured into a sweating steel bottle. That single image is then developed into a four-shot sequence for the 20-second ad: a close-up of the bottle in a kitchen in morning light, a hand filling the bottle with ice water, the bottle swinging from a person’s backpack on the walk to the station and, finally, a logo card made in the edit.

The start frame prompts are then run in Nano Banana Pro at 9:16 to generate a few versions of each frame. The product is described in exactly the same words every time (for example, “a matte forest-green stainless steel water bottle with a bamboo lid”), and the three best results are kept. You can also use your own product photos and generate only the backgrounds around them.

Next, the start frames are animated. Seedance 2.5 animates the first and third shots at 720p, for about 5 to 6 seconds each. Kling 3.0 Turbo is used to draft the pouring shot first, as hands and very close physical action are the sorts of shots Kling 3.0 is the first choice for. The winning prompt for the pouring shot is then run on Kling 3.0. The pouring prompt is:

Prompt

Copy

```
Close-up, locked-off camera. A hand tilts a glass jug and pours ice water into the open bottle. Ice cubes tumble in and the water rises to the top. Fine droplets form on the outside of the steel. Bright, cool morning light from the left. Fresh and crisp.
```

Two or three takes per shot are usually enough. For shots that aren’t working, especially liquid, change only the action sentence in the prompt to slow the action down or move a little further away.

Next comes the sound for the ad: a 20-second voiceover of about 45 words and a short, upbeat background track. The voiceover is generated with ElevenLabs Multilingual V2 and the music with Seed Audio 1.0 Music. The sound of the water pouring is a separate sound effect made with ElevenLabs Sound Effects, since it is the moment the whole ad hangs on.

The generated clips are then trimmed and cut together on movement. The same light grade is applied to all of the shots, captions are added so that people watching without sound can still follow the ad, and the logo card closes it out. In total, three generated video clips, a handful of stills and three generated audio pieces make up a single 20-second ad, all from one subscription, with no file transfers between tools.

## How to keep characters consistent across clips

Making one good short video is easy. Making a longer video from several shots, where the same person still looks like themselves in every clip, is where most AI video falls apart.

A robust solution is image to video with a fixed reference. Create one approved still of your character and build all the start frames from that image, instead of re-describing your character in each prompt. When you do have to describe them, paste a fixed character line (for example, “a woman in her early thirties with short curly auburn hair, a yellow raincoat and small gold hoop earrings”) into each prompt unchanged, and change only the shot, camera and action for each clip. Do the same with your light and colour sentence, so the shots match when they are cut together.

For a recurring character across multiple posts (for example, an AI presenter or another brand persona), build the character from reference images and reuse the same approved stills every time. When the character needs to speak, upload the image and an audio reference, either your own recording or an Elevenlabs voice, to Seedance 2.5, which lip-syncs the clip to that audio. UGC Factory makes ready-to-use UGC ads in a few clicks, for ads with realistic AI creators talking to camera.

## How many credits does an AI video take?

In PromptWise, every generation of images, video or audio is paid for in credits. The amount depends on the model, the length of the clip and the resolution, and the Generate button always displays the exact number of credits needed before you actually press it. That is the only figure you should rely on for a specific setting.

Starter costs $19 a month for 275 credits. Plus is $39 for 800 credits, Pro is $79 for 1,800, and Max is $149 for 3,800, with larger Max 2 ($249, 6,500 credits) and Max 3 ($349, 9,500 credits) options. Prices are in US dollars and exclude VAT, and paying yearly saves up to 45%. What each model uses per image or clip is listed on [PromptWise’s pricing page](https://promptwise.com/models-and-pricing/).

What does that mean for this guide? Say one idea takes three start frames, half a dozen Kling 3.0 takes and a voiceover. On Starter that leaves most of the month’s credits untouched, so there’s room for a second and third idea, or for doing the first one properly.

Three habits make credits go much further. Draft on lighter models (Seedance 2.0 Fast, Kling 3.0 Turbo or Veo 3.1 Lite) and run only the winning prompt on the final model. Experiment on stills, where each try costs far less than a video. And generate at the length you need rather than the longest the model allows. Credits refresh every billing month, and unused credits don’t roll over, so it pays to use them steadily. If you run low mid-month, you can top up in the app or upgrade anytime in settings. All plan details are on the [PromptWise pricing and models](https://promptwise.com/models-and-pricing/) page.

[Start creating](https://app.promptwise.com/sign-up)[see plans](https://promptwise.com/models-and-pricing/)[see plans](https://promptwise.com/models-and-pricing/)

## Troubleshooting AI video

### The camera drifts around for no reason

The prompt didn’t say how the camera moves, so the model invented something. Name one move with a speed and direction, such as “slow push-in” or “steady tracking shot to the left”. If nothing should move, say “locked-off camera”.

### Nothing really happens

The prompt describes a scene, not an event. Add a small beat: someone turns, a light changes, steam rises, a door opens. A clip needs a reason to exist.

### The motion looks rushed or jerky

Too much is happening for the length. Cut the action down to one main movement, lengthen the clip within the model’s range, or split it into two shots.

### Faces or hands look wrong

Faces that are small in the frame and fiddly hand gestures are where models struggle most. Move the camera closer, simplify the gesture and fix the face in the start frame first. For people-heavy shots, try Kling 3.0.

### The product changes shape or the logo melts

Video models are not reliable at keeping fine text sharp while things move. Start from an accurate still, keep product motion gentle (a slow orbit or push-in rather than a spin), and add logos and text in the edit.

### The clip changes the start frame too much

The video prompt is probably redescribing the scene. Cut it back to camera, action, light and mood, and let the start frame carry the rest.

### The style changes from clip to clip

The prompts are worded differently. Reuse the same character line and the same light and colour sentence, word for word, in every prompt of the sequence.

### The vertical version feels cramped

A wide shot has been squeezed into 9:16. Use closer shot sizes for vertical video and favour vertical movement such as a tilt up or a crane down.

### The generation failed

It happens occasionally. If a generation fails, its credits are refunded automatically, so just try again, perhaps with a simpler prompt or a different model.

## Make it repeatable with Flows

Workflows are useful once they’re working, but the real value comes from running the same workflow again with a different product, a different character or next week’s campaign. Flows is the PromptWise node-based creative canvas for exactly that. It connects prompts, images, videos and models into workflows you can reuse, refine and scale.

A typical flow for this guide connects a start-frame prompt to Nano Banana Pro, then passes the resulting image to Seedance 2.5 or Kling 3.0 with a video prompt. For the next version of the ad, you simply change the product description in the prompt and run the entire chain again, rather than rebuilding every step. For teams making many variations of the same ad, most of the time savings come from re-running a chain of connected nodes instead of building a new workflow from scratch each time.

You can also use PromptWise from chat clients like Claude, code editors like Cursor or any other MCP client, by connecting it under Account Settings → MCP.

## Frequently asked questions

### How do I make AI videos?

Start with a short plan: the platform, the length and the one thing the video must show. Make a start frame with an image model such as Nano Banana Pro, then animate it with a video model such as Seedance 2.5 using a prompt that describes the camera, the action in order, the light and the mood. Generate two or three takes, add sound and edit the best takes together.

### What’s the best AI video generator for beginners?

Seedance 2.5 is the best all-round starting point in PromptWise. It makes clips of 4 to 30 seconds, with sound, and every model is on every plan. For people and natural motion, try Kling 3.0. For dialogue or effects that have to land on an exact moment, try Veo 3.1.

### Is it better to make AI videos from text or from an image?

Use text to video to explore ideas quickly. Use image to video for anything you plan to publish, because starting from an approved still keeps faces, products and colours consistent and lets your prompt focus on camera and motion.

### What do I need before I start making AI videos?

Very little: an idea, a rough sense of where the video will be posted and an account with a few video models. What helps most is a three-line plan: who the video is for, what it should make them feel or do, and the single image they should remember. An editing app is useful at the end but not essential for short social clips.

### How do I fix an AI video that doesn’t look right?

Watch every take twice, then change one thing and generate again: adjust only the camera sentence if the camera moves too much, or simplify the action if it feels rushed. If a face looks wrong in every take, fix the start frame instead. If five takes fail the same way, split the shot into two simpler clips or try another model.

**More answers:** credit costs and which plan to pick are on [PromptWise models and pricing](https://promptwise.com/models-and-pricing/) · who owns what you make is covered in [plans and billing](https://promptwise.com/plans-and-billing/) · [adding a voice to AI video](https://promptwise.com/models/elevenlabs/)

## Start making AI videos

Making AI videos well comes down to a handful of decisions made in the right order. Know what the video is for. Pick a model that suits the shot. Approve the look as a still before you pay for motion. Describe one shot at a time, in the order it happens. Then review, fix one thing, add sound and cut it together.

PromptWise puts 40+ image, video and audio models in one web app on one subscription, including Seedance 2.5, Kling 3.0, Veo 3.1, Nano Banana Pro and the ElevenLabs voice models, so you can go from first frame to finished edit without switching tools.

[Start creating](https://app.promptwise.com/sign-up)[See plans](https://promptwise.com/models-and-pricing/)

## Related reading

- [How to Make AI UGC Ads: Hooks, Scripts and Examples (2026)](https://promptwise.com/blog/how-to-make-ai-ugc-ads/)
- [Nano Banana Pro Prompts: 31 Copy-Paste Prompts by Job (2026)](https://promptwise.com/blog/nano-banana-pro-prompts/)
- [Seedance vs Kling vs Veo: Which AI Video Model to Use in 2026](https://promptwise.com/blog/seedance-vs-kling-vs-veo/)
- [AI Image Prompts: How to Write Prompts That Work](https://promptwise.com/blog/how-to-write-ai-image-prompts-that-actually-work/)
- [Cinematic AI Video Prompts for Kling 3.0 and Veo 3.1](https://promptwise.com/blog/cinematic-ai-video-prompts/)
- [AI Product Photography: Replace Expensive Photoshoots](https://promptwise.com/blog/ai-product-photography-replace-photoshoots/)
- [5 AI Prompt Mistakes Killing Your Results (and Fixes)](https://promptwise.com/blog/5-ai-prompt-mistakes-killing-your-results/)
- [PromptWise pricing and models](https://promptwise.com/models-and-pricing/)