Photos and videos: give the build your own pictures
Stock photos make every AI-built site look like every other. With Anticode you can add your own photos and videos to a request or a change, up to eight at a time. Each file is checked by its first bytes and stored under your account. The model looks at your photos and uses them where they fit. It can't watch videos, so it places each one by its name and the words of your request.
Why your own pictures matter more than a better prompt
A photo of your actual shopfront, your actual dishes or your actual team tells a visitor more than any headline, and it's the one thing an AI can't invent honestly. When you add your photos to AI website builds, the model also learns from them: the colours of your interior, the mood of your products, what's in your window. The Architect is told not to ask about things your photos already show.
Adding files
- Where: the picture button in the request box on the dashboard, or the change box in the builder. You can also drag files onto the box, or paste an image.
- What: JPEG, PNG, WebP or GIF photos up to 5 MB each; MP4, MOV or WebM videos up to 50 MB each.
- How many: up to eight with one build, change or Architect reply.
- Before you start: an account with a confirmed email, since files are stored on your account.
Each file uploads as soon as you add it and shows as a tile with its progress and a remove button, so you can take one out before sending. What goes with the build is the file's id, not the file again.
Checked by what's inside, not what it's called
The type of a file is worked out from its first bytes, never from its name or what the browser claims. A file called photo.jpg that isn't really a JPEG is refused with a message listing the formats that are accepted. Files are stored under your account and served with a policy that stops them running anything, and with byte ranges so videos can seek and play in Safari.
What the model does with photos and videos
| Photos | Videos | |
|---|---|---|
| Does the model see it? | Yes, as an image, while the photos fit in one request | No; it gets the file name and type |
| How it's placed | Where the content of the photo fits the page | Where the name and your request suggest |
| How it's used | As an image or background, with your file kept as is | In an app build, as a muted video that plays inline, looping if it's a background, never as the only way to read anything |
Because videos are placed by name, a descriptive name helps a lot. hero-loop-kitchen.mp4 tells the model far more than IMG_4471.MOV. Say where you want it, too: "use the kitchen video as the background behind the headline".
If you send many large photos, the model is shown as many as fit in a single request and the rest are listed by name, so put the most important ones first. The model is told to use your files by exactly the addresses given and not to swap other images in for them.
Where uploads are used
Uploads travel with model-written builds and changes, and with the Architect's conversation before a build. In an app they're one of the few things the sandbox lets through: the app's policy allows images and video from its own site and nothing else. When you export an app to Next.js, the files are included in the project's public folder. Pages from the design engine can also use a curated photo library with written alt text, if you don't have pictures of your own yet.
Limits
- Storage per account. Each plan has a storage allowance for photos and videos; when it's full you'll be asked to remove some.
- Public by address. Uploaded files are site assets, served at an unguessable address. Don't upload anything you wouldn't put on a public website.
- Video plays only in apps. Pages laid out by the design engine can't show video; link to YouTube or Vimeo from them instead.
- No editing. There's no cropping, resizing or compression; prepare files before uploading, and prefer JPEG or WebP for photos.
- Rights are yours to check. Only upload pictures you own or have permission to use. See finding free images legally.
Questions people ask
Can the AI see my photos?
Yes. Photos are sent to the model as images, so it can describe and place them. Videos are described to it by name and type only.
Why was my file refused?
It was over the size limit, or its contents weren't one of the accepted formats, whatever its name said. Convert it to JPEG, PNG, WebP or GIF for photos, or MP4, MOV or WebM for video.
Do I need to write alt text for my photos?
Every image needs a text alternative to pass the security and accessibility checks. Check what the build wrote describes your photo accurately.
Can I reuse a photo in a later change?
Add it again with that change. Your earlier upload stays in your storage until you remove it.