Skip to main content

Images & attachments

You can hand Berrybell files as well as words. Attach a screenshot of a design you like, a PDF of your copy, or your logo — and generate new images for the app when you don't have one.

Attaching files to a message​

Use the paperclip in the chat composer (its tooltip reads Attach files or images), or drag files onto the composer. You can attach up to five files per message; files beyond the fifth are left off.

Supported types are images (JPEG, PNG, WebP, GIF), up to 5 MB each, and documents (PDF, plain text, CSV, Markdown, JSON), up to 10 MB each. The file picker only lets you choose these types. Dragged image files in other formats, such as SVG, are accepted too, but Berrybell may not be able to describe them. A file that's too large or of the wrong type is refused, and the reason — for example Attachment upload failed: File exceeds 5 MB limit — appears in the Logs tab under Raw → App logs (for your current Studio session only), not in the chat. The log line doesn't name the file. Uploading stops at the first refused file, so any files picked or dropped after it in the same go aren't added either. Check the chips above the composer and add the rest again.

Attaching files needs the Editor role or above (see Team & roles). Inside a project you can send files with no text at all.

You can also attach files on the What do you want to build? screen, before the project exists. There you must also describe what you want: files alone can't start a project, and sending without a description shows Describe what you want to build — attachments alone cannot start a project. The files are uploaded once the project is created. A file that fails to upload, for example one over the size limit, is left out of the first message without any separate notice. Check the attachments shown on that message and attach anything missing again from inside the project.

Removing an attachment before you send deletes the stored copy. That includes an image you attached with Attach to message in the Generate image panel: removing it deletes that generated image, although it still counts toward the project's limit of 50 generated images. After you send, attachments stay with the message.

What Berrybell does with them​

When you send a message with attachments:

  • The AI gets the file itself. It opens the file directly — for an image, that's the picture itself, so it can build from a screenshot or a mockup. A written description of an image, or the text of a PDF or text file (up to about 12,000 characters), is also made for quick reference, but the whole file stays available to the AI.
  • It's kept apart from your app's code. In a new project, attachments aren't saved as part of a version, and GitHub sync doesn't push them to your repository, because the project's .gitignore excludes the .attachments/ folder. An older project may not have that line; if you sync to GitHub, ask the AI to add .attachments/ to .gitignore.
  • It may also be copied into your site. Berrybell copies the attached files into the site's public folder, under their filenames, when your message mentions a site asset (a logo, favicon, banner, avatar, hero image or background image), or refers to the attachment together with a word like use, replace, add or make. For example, "make it look like the attached screenshot" counts. Spaces and unusual characters in the name become underscores, and a file of the same name already there is replaced. A copy there is part of your app: it's saved in versions, pushed by GitHub sync and published with the site. If you attached something private just for reference, ask the AI to remove the copy from the public folder. Otherwise, the AI copies a file there itself if your request needs it.
What this costs

Uploading is free, and PDF and text extraction is free. Describing an attached image is billed as AI usage — see Plans & credits.

Re-attaching an image that Berrybell generated for you doesn't re-run the description step and isn't billed again.

Generating images​

The image button in a project's composer (tooltip Generate image with AI) opens the Generate image panel. It isn't on the new-project screen, so create the project first. The panel has:

  • A prompt describing the image you want.
  • A Style preset — Hero, Illustration, Photo, Icon or Logo.
  • A Size — Square, Landscape or Portrait.
  • A Quality setting — Low, Medium or High — and a Model tier: Fast or Best.

An image is charged what it actually cost to make plus Berrybell's margin, like any other AI request. There's no flat price per image. The generate button shows an estimate before you run it: at the defaults (Best, Medium, Square) it reads Generate (~0.34 credits). Estimates run from about 0.02 credits (Fast, Low, Square) to about 2 credits (Best, High, Landscape or Portrait), and Landscape and Portrait cost about 1.5× as much as Square. The final charge can differ slightly from the estimate.

To start any generation, the workspace needs about 2 AI credits available. That's a minimum balance, not a price: you're charged only what the image actually cost. A generation that fails, or that's rejected by content moderation, isn't charged.

Once an image is generated you can:

  • Add to project — copy it into the site under /generated/.
  • Use as hero — copy it into the site and put a suggested request in the chat box asking the AI to use it as the hero background. Nothing on the page changes until you send that message, which runs as a normal build.
  • Use as OG — make it the social preview (OG) image. This one also updates the site's metadata, so it's what shows up when the link is shared. See SEO & discoverability.
  • Attach to message — attach it to your next message.
  • Regenerate — run the prompt again. This is a new generation, charged like any other.
  • Copy path — once the image is in the site, copy its path so you can refer to it in a request.
  • Generate new image — clear the panel and start on a new prompt.

Both Add to project and Use as hero fill the chat box with that hero suggestion and replace anything you'd typed there. If you only wanted the file, clear it.

The Recent grid at the bottom of the panel shows the project's last 12 generations; pick one to bring it back up. In a shared project, images other people generated appear in the grid but without a preview, and can't be attached to your messages.

Inserting an image into the project needs the preview to be running. Otherwise you'll see "Open the preview to add images to the project." (or, for Use as OG, "Open the preview to promote images to the project."). Open the project and let the live preview start first.

To set a favicon, use the icon picker in Site info, in the Publish popover's ⋮ menu, rather than the image generation panel.

Limits​

  • Each person can start 10 image generations per hour. That includes the images the AI makes during your builds, and attempts that fail.
  • A project can hold 50 generated images. After that, every new generation — yours or the AI's — is refused with This project has reached the maximum of 50 stored generations. Delete old ones first. The Studio doesn't have a way to delete generations yet, whatever the message says, so contact support if you reach it.
  • Prompts are capped at 4,000 characters and screened before they run.

The AI generates images while building, too​

This is the part worth reading twice.

Rather than faking a gradient or linking to an external image that might break, the AI can create real images itself during a build and drop them into your site — up to four images per message, without you pressing anything. A failed attempt counts toward the four. Each image is saved under /generated/ in your site and shows up in the panel's Recent grid; while one is being made, the activity reads Generating an image: ….

These images are billed like the ones you generate yourself, always at the Best tier. Each one is charged as it's made, so you pay for it even if the run later fails or you cancel it. They're listed under Image generation in your usage breakdown, not in the run's own charge or the chat's running total. The same hourly and per-project limits apply. When a limit stops an image, or the workspace has less than about 2 credits available, the AI builds without it and tells you so.

If you're watching your credit usage, this is the most surprising line item in the product. You can tell the AI not to: "don't generate images, use plain colour blocks" works fine as an instruction.

Next steps​