Why EB-1 Shows You the Price Before Anything Is Made
An AI agent that can spend your money needs a simple rule you can rely on. For EB-1 that rule is: nothing is made or charged until you have seen the exact price and tapped Approve.
That sounds obvious, but it shaped a lot of how EB-1 works under the hood. The price you see is worked out on our server from the exact request that will run, not from a rough estimate. An approval runs once, never twice. Chatting is counted separately from making things, so brainstorming never quietly uses up your credits. And when something fails on our side, you get the charge back.
This post explains why we built it that way, and what that means when you use EB-1.
What happens between "make this" and the result?
When you ask EB-1 to make something, four things happen in order:
- EB-1 drafts a proposal: one image or one video, with the model, format, length and the prompt it wrote for you.
- Our server rebuilds and prices it. It does not trust the draft as-is. It keeps only settings the chosen model really supports, then prices the cleaned request with the same pricing the rest of Ecombos uses.
- You see an approval card with what will be made and its exact credit price.
- You approve, and it runs once. The credits are charged when the job starts.
Why does the server set the price, not the chat?
Because a price is only honest if it is calculated from what will actually run.
Early on, we found a real problem: a request could ask for far more than its price covered, for example a long, high-resolution video quoted as if it were a short one. The fix was to stop letting any request travel straight from the chat to a model. Today, every proposal and every approval is rebuilt from an allowlist of what each model supports, and priced from the cleaned settings. Ask for more seconds or a higher resolution, and the price goes up accordingly, before you approve.
What if I change the length after seeing the price?
You can adjust the length or the opening frame on an approval card. When you do, the new version is re-priced on the server and saved as a new revision of the same card. Approval stays blocked until that revision is saved, so you can never approve a price that no longer matches the request.
Why can an approval only run once?
Networks retry, buttons get double-tapped, and browsers reload. If an approval could run twice, any of those could charge you twice. Each proposal has its own id, and the server runs each one at most once, however many times the approval arrives. Approvals are also tied to your account, so one person's approval cannot run on someone else's credits.
Why are chat messages separate from credits?
Thinking should feel free to use, and you should never have to ration questions to protect your credits. Asking EB-1 for hooks, scripts, captions, three creative directions or a full series plan does not spend credits. Instead, each message you send uses one of the EB-1 messages included in your plan each month.
When your monthly messages run out, you can keep chatting: 1 credit buys a block of 5 more messages. Approving a generation never counts as a message.
We count a message before EB-1 starts working on it, so the count cannot be skipped, and we give it back if you never got a proper reply: if the turn hit a hiccup, was stopped by a safety check, or errored on our side.
| You do this | What it uses |
|---|---|
| Send EB-1 a message (ideas, scripts, plans, questions) | One EB-1 message from your monthly allowance |
| Run out of monthly messages | 1 credit for a block of 5 more messages |
| Approve an image or video | Its credit price, shown on the card; never a message |
| A reply fails or is stopped by a safety check | The message is given back |
| A generation fails and our systems confirm it | Its credits are returned automatically |
What happens to credits when a generation fails?
If a job fails at the provider, or our checks confirm it never produced a result, the credits for it are returned automatically. We are careful about the reverse too: once a provider has accepted and delivered a job, credits are not returned for results you simply do not like, because that work was done and paid for on our side.
EB-1 also checks finished images against your brief and the content rules. If a finished image breaks the content rules, it is removed and its charge is refunded.
Why not just charge a flat price per request?
A flat price is simple, but it is unfair in both directions. A 5-second, 480p draft and a 30-second, 1080p hero video cost very different amounts to make. With a flat price, small requests subsidise big ones, or big ones are quietly limited. Pricing the exact request means you pay for what you asked for, and you see it before you agree.
What this means when you use EB-1
- Brainstorm as much as you like; it uses messages, not credits.
- Look at the approval card: it states the model, format, length and price.
- Change the length on the card if the story needs more or less time; the price updates before you approve.
- Approve once; it runs once.
- If something fails on our side, you are not charged for it.
Open [EB-1](/agents/eb-1) and ask for anything. You will always see the price first.
Questions people also ask
Does EB-1 ever charge without asking?
No. Images and videos are only made after you approve a card that shows the exact price.
Do messages cost credits?
Messages come from your plan's monthly EB-1 allowance. Once that is used, 1 credit buys 5 more messages. Approving a generation is never a message.
Why did the price change when I changed the length?
Because longer or higher-resolution videos cost more to make. The server re-prices the exact request before you can approve it.
What if I tap Approve twice?
Each proposal runs at most once, so a double tap or a retry cannot charge you twice.
Will I get my credits back if a render fails?
Yes. When our systems confirm a generation failed, its credits are returned automatically.
Can I see what model EB-1 chose?
Yes. The approval card shows what will be made, and you can ask EB-1 to use a different connected model; the card then shows that model's price.