Luma AI Turns Your Photos and Prompts Into Shareable Video

Chapter 1

One Login, Nine Ways to Start a Video

Priya runs a small candle shop out of her apartment and wants a ten-second clip of a new scent for Instagram. She has no camera crew, no editing software, and about twenty minutes before she needs to post. A friend mentions Luma AI, and she expects one video generator – type something, get a clip. What she finds instead, behind the same login, is closer to nine tools: type a description and get a video, upload a photo and watch it move, merge two photos into a hug or a kiss, redraw a photo in an anime or classic-cartoon style, give a still portrait a voice, or take a video already made and reshape it for every platform it needs to reach.

None of this is obvious from the homepage alone, which mostly shows a text box and a short set of questions. Each mode actually lives on its own page, built around one specific thing a person wants to do with a photo or an idea, and nothing on any single page tells you the other eight exist.

This book goes through them in the order a person is likely to meet them – starting from nothing, moving to a single photo, then two photos, then changing how a photo looks, then giving it a voice, then reshaping a finished clip, and finally what all of this costs in credits and which plan fits how often generation actually happens.

By the end, the goal is not to know that Luma AI has many features – it is to know which single page to open for the specific photo or idea sitting in front of you, what it will ask you to upload, and where the site's own rules stop you before a credit is spent on something it will not generate at all.

What you can actually do here

Luma AI's nine generation modes live on separate pages with no single map between them. Here is what each one actually does, what it needs from you, and which chapter walks through it.

Starting from nothing

Two photos, one moment

Changing how a photo looks

Giving a photo a voice

Getting it everywhere, and paying for it

Chapter 2

Describing a Scene Instead of Filming One

Before a single frame is filmed, video usually needs three things: a location, a subject, and a camera. Luma AI's text-to-video mode replaces all three with a sentence.

The mechanics are simple. Write what you want to see, choose 5 or 10 seconds, and pick a 1:1, 16:9, or 9:16 frame depending on where the clip is going – square for a feed, vertical for Stories or Reels, widescreen for anything else. The tool generates a short video from that description alone, with no footage, actor, or camera involved anywhere in the process.

The one documented lever a person controls is the description itself. The site states plainly that the more specific the description, the closer the result – which is a different discipline from writing a caption. A caption describes a photo that already exists; a text-to-video prompt has to describe a scene that does not exist yet, in enough detail that the system has something concrete to build rather than something vague to guess at.

This is also the mode with no photo to fall back on if the result misses the mark. Image-to-video and the two-photo animations covered in the next two chapters start from something real; text-to-video starts from nothing but the sentence, which is exactly why the specificity of that sentence carries more weight here than anywhere else in the product.

Chapter 3

Making a Still Photo Move

A product photo already exists in most sellers' phones – lit, cropped, ready. What is missing is motion, and three separately named pages on the site solve that same problem: image to video, photo to video, and animate a photo. Each is the same underlying idea told from a slightly different angle: give it a still, get back a short clip of that still, brought to life.

The process does not change across any of them. Upload one image, choose a 5 or 10 second length, and pick a 1:1, 16:9, or 9:16 aspect ratio to match wherever the clip is heading – a square product post, a vertical Reel, or a widescreen upload. There is no timeline to learn and no plugin to install; the entire thing runs in the browser from upload to download.

This mode suits exactly the situation Priya was in: a photo that already says what it needs to say, needing only movement to stop scrolling past it. It differs from text-to-video in one practical way worth holding onto – here, the finished clip is anchored to a real photo already chosen, so the result stays recognisably close to that photo rather than depending on how well a sentence was written.

Portraits, product shots, and anything else that is already a good still photograph are the strongest candidates for this mode. A blurry or very small source image carries its flaws straight into the finished video, since nothing here repairs a photo – only something that moves it.

Chapter 4

Two Photos, One Moment, and the Rule Behind It

Two sisters have not been in the same country in three years. One of them finds Luma AI's hug animation and uploads a photo of herself and a photo of her sister, expecting nothing more than a novelty clip – and gets back a short video of the two of them embracing, built entirely from two photographs that were never actually taken together.

The mechanics are shared across three animations that live on the same page: hug, kiss, and fight. Upload one photo of each person, pick the animation, choose a 5 or 10 second length, and generate. Photos of at least 300×300 pixels work best, and anything larger than that is compressed automatically before the clip is built, so there is no manual resizing step to get wrong.

The site draws one line around all of this clearly, and it is worth reading before uploading anyone's photo but your own: these animations are meant only for people who have given their explicit, informed consent to have their likeness used this way. A photo of a celebrity, a public figure, or a private individual who has not agreed to it is not something this feature is meant to be used on – the content policy names this directly, alongside a broader ban on using the platform to create realistic likenesses of people without consent, particularly where the result could deceive, defame, or harass them.

For a reunion clip between two people who both know it exists and are happy to see it, none of this changes anything about how the feature works. It matters most for the case that will eventually occur to someone out of curiosity: using a stranger's or a celebrity's photo. That case is exactly what the policy exists to rule out, and it is checked with human review as well as automated screening on every upload.

Chapter 5

Redrawing a Photo Into Someone Else's Style

A photo taken on an ordinary phone can look like it was drawn by hand, in a handful of recognisable styles, without hiring an illustrator. The Styles tool takes an uploaded photo, redraws it in one chosen look, then animates the result into a short clip.

Five looks are documented across the captured pages: a soft anime-influenced style, a classic animated-film style, and looks associated with Simpsons, GTA, and claymation. No single page lists all five together – the Disney-style page mentions the others only as a one-line pointer at the end, and the Ghibli-style page does the same in reverse. Someone who lands on one style page through a search or an ad has no way of knowing, from that page alone, that four more looks exist on the same tool.

The mechanics match every other single-photo mode: choose a style, choose 5 or 10 seconds, generate. Photos of at least 300×300 pixels work best, the same guidance as the two-photo animations.

One real limit applies specifically here. The site's content policy states that these style filters produce original AI-generated interpretations inspired by broad artistic techniques – not reproductions of specific copyrighted characters or trademarked brand elements. A style transfer leaning on a recognisable character rather than a general artistic look sits outside what the tool is meant to be used for, and responsibility for that sits with whatever is typed or uploaded into it.

Chapter 6

Giving a Photo Something to Say

A spokesperson clip usually needs a person willing to be filmed saying something on camera. Luma AI's lip sync and avatar tools replace that person with a photo and an audio track: upload a portrait, or an existing photo or clip, add audio, and the system generates matching mouth movement so the subject appears to speak the words.

Two pages describe what is functionally the same capability from two directions. Lip Sync starts from either a photo or an existing video clip and adds a voice to a face already there. Avatar starts specifically from a single portrait and turns it into something closer to a presenter – a photo that talks, rather than a video with a voice added to it.

Both suit the same kind of use: an explainer, a spokesperson-style clip, or a short social post where filming an actual person is not practical or not wanted. Both also depend entirely on the audio being ready beforehand – there is no recording or scripting step inside the tool itself, only the matching of mouth movement to whatever audio is uploaded.

This is the one mode across the whole site where the input photo does not need to show motion, action, or a second person – a single still, calm portrait is the ideal source, since the entire result rides on the audio doing the work the photo cannot.

Chapter 7

One Video, Every Feed

A video shot or generated in one aspect ratio rarely fits every place it needs to be posted. A widescreen clip crops awkwardly into a square feed; a square clip leaves bars on a vertical Story. Reframe takes a video that already exists and reshapes it into 1:1, 16:9, or 9:16 while keeping the main subject in view, rather than making the creator crop it by hand in a separate editor.

This is the one mode in the whole product built specifically for repurposing rather than creating. Everything else in this book starts from a photo or a sentence and ends with a new clip; Reframe starts from a finished video and produces versions of that same video, sized for wherever it still needs to go.

It pairs naturally with everything upstream of it – a text-to-video clip generated at 16:9 for a website can be reframed into 9:16 for a Story without regenerating anything from scratch, which is the practical reason to know this mode exists at all rather than assuming a video is locked into whatever shape it was born in.

Chapter 8

What a Credit Actually Buys, and Which Plan Fits

Every generation on the site draws down credits, and the credit cost is decided by two things: which model is chosen, and how long the clip runs. The Standard model costs one credit for a five-second video; the Pro model costs two credits for the same five seconds. Length scales the cost further within either model – a ten-second Pro clip costs twice a five-second Pro clip.

What that means in practice is that the model choice, not the plan, decides how far a fixed number of credits goes in a given week. Someone generating mostly Standard-model clips gets roughly double the output of someone generating the same number of Pro-model clips from an identical credit balance.

The three plans – billed weekly, monthly, or yearly – differ from each other in almost nothing except the billing period and the price attached to it. Every one of them lists the same feature set: highest video quality, no watermark, custom aspect ratios, dedicated support, and both image-to-video and text-to-video access. There is no free-tier feature gate to navigate here, no plan that unlocks a mode another plan hides – the decision is entirely about how often generation happens, not which capabilities are available. The use map at the front of this book carries the exact current figures.

A person generating a handful of clips around one event – a single product launch, one family reunion – is better matched to the shortest commitment available. Someone posting on a weekly cadence, across product shots, style transfers, and the occasional two-photo animation, is the shape of user the higher-commitment plans are actually built around.

One more limit is worth knowing before buying credits ahead of time: the site reserves the right to delete an account, and any credits tied to it, after 180 days of no activity, with no refund offered. Buying a large credit balance and then not returning for six months is the one scenario where unused credits genuinely disappear – a reason to match the plan to how often generation is actually expected to happen, not just to how much cheaper the yearly rate looks on paper.

Questions readers actually ask

Contact / More useful information from RamthaMedia

Official source links:
Luma AI