AI video that remembers your brand: how Blooper's creative memory works

Most AI video tools start every chat from zero. You explain the brand again, the product rules again, the client’s colour again. The tool did good work last week, and none of it carried over.

Blooper keeps a memory of how you work. Not a transcript of old chats: a small set of cards that say what was settled, who it applies to, and where it came from. You can open every one, change it, pause it or delete it.

Three kinds of memory, one page

Inside a project there is a Team memory page. Everything Blooper keeps about how your team works lives there, as a grid of tiles with pictures, under four tabs:

• Inbox. What Blooper has learned and is waiting for you to accept or dismiss.
• Notes. Things the agent takes into account: a preference, a fact about a client, a quirk of a model. A note suggests. It does not force.
• Rules. Settings your team has decided, like an aspect ratio or a model. A rule binds: Blooper holds the agent’s generations to it.
• References. Images your team keeps on purpose, so anyone can use them in a project.

Each tile has a picture, such as the image or clip a note is about. Click one to open it full size.

You, your team, your client

Every memory says whose it is:

• Personal. Only you. How you like to work, in any project.
• Team. Shared with everyone on the team.
• Blooper. General notes about the tools themselves, such as how a model is best prompted or what a vertical social format needs. They are the same for every team, and you can mute any of them in a project.

On top of that, a memory can apply to the whole team, to one client, to one project, or only to you. So “the product is never cropped” can be a team habit, while “emerald is the house colour” belongs to one client and never leaks into another client’s work.

A note you write yourself starts private. It reaches your teammates only when you share it.

How it learns

Blooper learns from the moments you correct it. Today that means three things:

1. You edit a prompt before approving it. The difference between what Blooper proposed and what you approved is a correction.
2. You redirect a run that is still going. A message typed while the agent is working is a correction too.
3. You ask for a change to what it just made. “Tighter on the bottle, keep the label to camera” after a shot comes back is the most common correction of all.

For each one, Blooper asks a simple question: is this about this one shot, or is it something that should hold next time? Most corrections are about this shot, and nothing is kept. When one is meant to last, Blooper writes it as a short note, private to you, and puts it in your Inbox.

There is a second, quieter source. If you keep asking for the same setting, say 9:16 for one client, in at least three messages across two or more chats within two weeks, Blooper suggests a note like “Usually 9:16 for this client”. That check is plain counting, with no model guessing at what you meant.

Nothing learned goes straight into your work. A suggestion waits in your Inbox until you accept it, or until the same lesson has come from corrections in at least two separate chats. Either way it shows up on the page, where you can see it, change it or delete it.

Notes suggest, rules bind

A note is advice to the agent. It reads the notes that apply before it plans a shot, and weighs them against what you are asking for right now. If you ask for something different today, today wins.

A rule is a decision. A team admin sets it, for the whole team or for one client, and every generation the agent makes is checked against it before it is priced. If the agent picks a value that breaks a rule, that call is refused and the agent tries again, so you are not charged for a shot that was wrong from the start. A rule set inside a project comes first, then the client’s, then the team’s.

Blooper never turns a guess into a rule on its own. A note becomes a rule only when someone on your team makes it one.

Seeing it at work

Under your message in a chat, a small line tells you when memory was in play: “3 team memories in context”, with the picture of each. Click one to open that memory on the Team memory page. If a note is wrong, you fix it right there, and the next chat reads the fixed version.

From a memory’s page you can:

• edit its words in place,
• Pause it, if it is true but not wanted for now,
• Delete it, with a few seconds to undo,
• Share a personal note with the team,
• Make it a rule, if it should bind,
• Use in this project, for a reference image.

Starting with what you already have

You do not have to wait for Blooper to learn. Write a note yourself. A team admin can set rules directly, or import a brand book: a PDF with selectable text, a Word file, or a plain text or Markdown file. Blooper reads it and proposes notes and rules one by one. Nothing is ticked in advance; you confirm each one.

What memory is not

It is not a promise that the agent always gets it right. The agent can still misread a note. That is why every note is visible and editable, and why only rules are enforced.

It is not training. No model is retrained on your work. Memory is stored as your team’s own cards, and a card one team writes never reaches another team.

And learning a lesson does not cost you credits. Blooper pays for that step.

Where it is today

Creative memory is in early access. It runs on Blooper’s own team first, and a team admin switches it on for their team when it reaches them. If you want it on yours, tell us.

The next two posts look at what memory saves: the credits a wrong shot costs, and the time a brief takes when it starts where the last project ended.

Start a project in Blooper, or see pricing.