Image to Video: Getting Your Pictures Ready
Three doors a picture can go through
There is no single “add image” box. A picture goes in through one of three doors, and each has its own count, size limit and plan requirement.
| Door | Where | How many | Size and format | Plan |
|---|---|---|---|---|
| Photos for an ad | Home page, Ad Video tab, Add images | Up to 4, all of one subject | Under 10 MB each once converted, 16 MB in all | Any account. Reading the photos is free |
| A character, place or prop picture | Step 2, Upload & Replace on the card | One per card | JPG or PNG; about 9:16 for a person, 16:9 for a place or a prop | Paid plan |
| The asset library | Assets, Upload asset | As many as your storage allows | JPG or PNG, up to 20 MB each | Paid plan; counts against the plan’s storage |
The one door that is missing is a single still that becomes the first frame of a clip. Pictures here are references for who and where, not frames to animate, which AI image to video: four methods explains. Everything below is about getting a picture through whichever of the three doors you need.
What your browser does to every picture
Whatever you pick, the same thing happens before a single byte is sent. The browser opens the file, applies the camera’s rotation so an upright phone photo stays upright, paints any transparent area white, scales the picture down if its longer side is above 4,096 pixels, and saves it again as a high-quality JPEG. There is one path and no shortcut, even for a file that was already a JPEG. Renaming photo.webp to photo.jpg gets you nothing, because the site never trusts the name: it works from the picture the browser has just encoded.
The reason is practical. The image and video services further down the line accept JPEG and PNG only, so converting at the door means a file is never accepted at upload and then found unusable three steps later. The price is one extra round of JPEG compression, which is invisible at the sizes involved.
| What you pick | What happens |
|---|---|
| JPEG | Opened and saved again as a JPEG. Size may change a little. |
| PNG | Transparent areas are painted white and the result is saved as a JPEG. A cut-out product on a clear background arrives on white. |
| WebP | Any browser that can show it can convert it, so it works. |
| HEIC / HEIF (iPhone) | Works in Safari, which can open it. Most other browsers cannot, and then the file is refused the moment you pick it. |
| SVG, camera RAW | Refused when you pick them. The browser cannot turn them into a photo. |
| A PDF or a document | Refused as not an image. Export the page you want as a picture first. |
| A picture still over 20 MB after conversion | Refused as too large. Compress it or crop it and pick it again. |
Refusals appear on the spot, before anything is uploaded or charged. What you can feed in lists the other kinds of input, such as scripts and video.
The limits, in numbers
| Limit | Number | Applies to | What happens past it |
|---|---|---|---|
| Long side | 4,096 px | Every picture, in your browser | Scaled down before upload |
| File size after conversion | 20 MB | Every upload | Refused as too large |
| Photos in one ad | 4 | Ad Video tab | “You can use up to 4 images”; the extras are dropped |
| Size of one ad photo | 10 MB | Ad Video tab | “That image is too large — use a JPG or PNG under 10 MB” |
| Total size of the ad photos | 16 MB | Ad Video tab, all photos together | Refused as too large in total |
| Width ÷ height of an app or website screenshot | From 0.4 to 2.5 (2:5 tall to 5:2 wide) | Ad Video tab, screenshots only | “This screenshot is too tall or too wide — crop it” |
| Short side of a reference picture | 512 px | A picture handed to the video model | Scaled up to 512 on the short side (adds no detail) |
| Long side of a big reference picture | 2,048 px | Reference pictures over 2 MB | Scaled down before the video model sees them |
The last two rows are quiet repairs rather than rules. A small picture is enlarged so that a segment is not refused for being tiny, and a very large one is reduced because pixels beyond what the finished video can show are wasted. Neither will rescue an extreme shape: pictures much wider than about 4 to 1, or much taller, are not enlarged and are likely to be refused at generation with an “invalid input” message. Invalid input: what to check walks through that message.
Which pictures work, and how to prepare each
| Your picture | It works as | Prepare it like this |
|---|---|---|
| A product on a plain background | The hero of an ad | Fill most of the frame, in focus, and put it first: the first photo is the featured subject and is also the picture on the end card |
| A dish, drink, garment, car or room | An ad photo | Same: clear, in frame, in focus. Extra photos show another angle, a detail or the place it lives |
| A shopfront or restaurant, customers included | An ad photo | Leave the people in. They are removed and the place is kept, and you will see a note saying so |
| An app or website screenshot | An ad photo, used as it is | One screen only. Crop a full-page capture to a phone or computer shape. A screenshot with a real face in it is refused |
| A selfie or snapshot of a person | A character picture | One person, face visible and turned toward the camera, even light. Waist-up usually beats a distant full-length shot. Avoid group photos, heavy filters, strong coloured light and sunglasses |
| Artwork you drew as a character sheet | A character picture, Use this image as is | Full figure, plain background, about 9:16 |
| A photo of a real place | A scene picture | Show the space as a whole. If it has people or a strong time of day in it, choose the optimise option so those do not follow you into every shot |
| A celebrity, a franchise character or a brand mascot | Nothing | The IP check refuses them when a clip is made, so there is no point preparing the picture |
A blurry, tiny or ambiguous photo is refused at the reading step with “We couldn’t work out what is being advertised here”, and nothing is charged for a refused picture. Which photos work for an ad · Using your own photo.
One thing to remember about a snapshot
On a character, place or prop card the picture becomes the reference for every shot that shows it, which is why Optimize into a reference sheet exists. A snapshot brings its background, its lighting and its pose with it, and the video model will treat all of those as part of the character. The optimise option redraws the face, hair and clothing from your photo in the project’s own look on a clean background; Use this image as is keeps the photo untouched and is for artwork already made to be a sheet. Clips generated before you uploaded keep the old look until you regenerate them. Your own reference images: what changes.
The first time you upload anything, a window titled Before you upload asks you to confirm that you have the right to use the material and that anyone whose face or voice is in it agreed. You confirm once.
A two-minute checklist
- Pick the door. A promo for one thing is the ad door. A person, a room or an object that must look the same in every shot is a card in step 2.
- Crop to the subject. For a screenshot, crop to a single screen; keep its width-to-height between 2:5 and 5:2.
- Order the photos. The hero goes first. Four is the most; one clear photo is enough.
- Do not resize it yourself. Anything up to a 4,096 pixel long side is fine, and the browser handles the rest. Do keep each ad photo comfortably under 10 MB.
- Clear the picture of things you do not want in every shot, such as a busy background or a colour cast.
- Check you may use it, and that anyone recognisable in it agreed.
- Write the words that go with it. A card picture has no description of its own, and the description is what the shot text reads, so use AI Describe and correct the result.
- Pick the file and read the message. If it is going to be refused, it happens now, for free.
What the picture becomes
Once a picture is in, it is a reference: it tells the model what a person, a room, an object or a product looks like, and each shot is a new picture composed from the shot text. Your subject appears in new shots, not your photo set in motion. For the ad route that means photo to video AI, up to four photos into a 15, 30 or 60 second spot. The wider routes, and what a free account can do with a picture, are in Image to video AI.
FAQ
What image formats can I use for image to video?
Any format your browser can open, which in practice means JPEG, PNG and WebP, plus HEIC in Safari. The browser converts everything to a JPEG before it is sent. SVG and camera RAW are refused.
What is the maximum image size?
The long side is reduced to 4,096 pixels in your browser, and a converted file over 20 MB is refused. For an ad, each photo must be under 10 MB and the photos together under 16 MB.
How many pictures can I use?
An ad takes up to four photos of one subject. A character, place or prop card takes one picture. The asset library takes as many uploads as your storage allows.
Will the video start from my exact picture?
No. The picture is a reference for who or what appears. Each shot is composed from the shot text, so you get your subject in a new shot.
Why was my image refused?
The usual reasons are a format the browser cannot open, a file still over 20 MB, an ad photo over 10 MB, a screenshot with a real face or an extreme shape, or a picture the reading cannot make out.
Does a small picture hurt the result?
A picture with a short side under 512 pixels is scaled up so it can be used, but scaling adds no detail, so a sharper source gives a sharper reference.
Do I need a paid plan to upload a picture?
Not for the photos of an ad. Putting your own picture on a character, place or prop card and uploading to the asset library are paid-plan features.
Can I use a photo of a person?
Use only pictures you have the right to use. A photo of a real person may be flagged, and recognisable public figures and franchise characters are refused when a clip is made.
Run the checklist, then pick the file
Refusals show up the moment you pick a file, before anything is charged.
Updated November 2, 2026