Coming soon on 8frame. AI agents on the board, and the approval card that will stand in front of every step that spends credits, are not released yet. This page covers why approval matters, how to keep AI generation spend under control on 8frame today, and how the approval step will work once agents ship. No date is announced.
An agent that can only read and write text is cheap to get wrong: you delete the answer. An agent that can start image and video generations is different. Each run costs money, takes time, and leaves output on a shared board that colleagues and clients will see. The safest rule is also the simplest: the agent proposes, a person approves, and only then does anything get generated.
Why "ask before you spend" is the right default
Security people have a name for an agent that does more than it should: excessive agency. The OWASP Top 10 for LLM Applications lists it as LLM06:2025 and traces it to three causes: "excessive functionality", "excessive permissions" and "excessive autonomy". One of its mitigations is to "require a human to approve high-impact actions before they are taken."
For a creative team, "high-impact" mostly means "costs credits" and "puts work in front of other people". A good approval step answers four questions before anyone says yes:
- What will it make? In plain words, not a tool call.
- With what? Which model, how many outputs, from which inputs.
- Where will it go? Next to the original, in a new area, replacing nothing.
- What will it cost? The credit total, before the run.
If any of the four is missing, the approver is guessing.
How it works on 8frame today
There are no agents on the board yet, so today every generation on 8frame is started by a person pressing launch on a node. That already gives you a natural approval point, and a few facts help you manage spend:
- Credits are per person. Each generation is charged to the user who runs it, on their own plan. Credits are not pooled across a team.
- Prices are fixed per setting. Nano Banana 2, the Nano Banana node's default, is 11 credits per image at 1K; Nano Banana Pro is 21 at 1K or 2K and 42 at 4K. The Veo tool opens on Veo 3.1 Fast, 168 credits for 8 seconds with sound.
- Failed generations are refunded automatically.
- A credit is about $0.01 at pack rate. Starter is $19 a month for 1,000 credits, Creator $49 for 3,000.
A manual approval routine you can run now
- Agree a budget per stage. For example: directions in stills only, then motion once one still is approved.
- Price the step before you run it. Four camera angles of an approved frame on Nano Banana Pro is 4 x 21 = 84 credits. One 8-second Veo 3.1 Fast clip of the winner is 168. Write the number in a comment next to the node.
- Get the yes in writing. A comment on the board from the person who owns the budget ("Approved, go ahead") is your record. There is no approval button or node lock on the canvas today.
- Run it, and keep the result next to its source. The result stays on the board with the node and prompt that made it, so the next reviewer sees both.
This is the same discipline as a client approval workflow for AI creative, applied inside the team.
What is coming: the approval card
When agents arrive on the board, the routine above will be built in.
Every spending step will wait for a card. When an agent wants to generate, it will post an approval card in the board chat: a short sentence about what it will do, a few lines on how (for example the model and the number of outputs, and where they will be placed), and two buttons, Approve and Not now. Nothing will be generated before someone presses Approve. While a card is waiting, the agent's node on the board will show that a plan needs review.
Reading and proposing will not need a card. An agent that reads a brief, comments on a node or proposes a layout spends nothing, so it will be able to do that straight away. Proposed nodes will appear on the board inside the agent's own frame, marked as the agent's work.
Multi-step plans will show the whole picture first. For a larger job, a Producer-style agent is planned to show a plan with who does what and the total before any of it runs, and each spending step will still have its own card.
Roles will decide who can approve. The plan is that editors on a board can launch agents and approve their cards, commenters can ask an agent a question, and viewers can read the thread.
No unattended spending. In the first version an agent runs when a person launches or mentions it. There are no scheduled or background runs that could spend credits while nobody is looking.
For how agents fit on the board more broadly, see AI agents as teammates on a creative canvas. For the basics, what an AI agent is.
What to decide as a team before agents arrive
Even before the feature ships, these are worth agreeing:
- Who approves spend on each board. Usually the person who owns the budget for that piece of work, not whoever happens to be online.
- A per-stage ceiling. For example, no more than a set number of stills per direction before a review.
- What "approved" means. An approval of a step is not an approval of the final asset. Keep the creative sign-off separate, as in any creative review.
- Where the record lives. Today, comments on the board. With agents, the approval card and the agent's thread will hold it.
FAQ
Can an 8frame agent generate images without my approval?
Agents are not released yet. When they ship, every step that spends credits is designed to wait for someone to press Approve on an approval card.
What will the approval card show?
What the agent will make, how (such as the model and number of outputs), where the result will go, and two buttons: Approve and Not now. Credit costs for agent steps will be shown when the feature launches; we are not publishing them in advance.
Who pays for a generation on 8frame today?
The person who runs it. Credits are charged to that user's own plan and are not pooled across a team. Failed generations are refunded automatically.
Is there an approval button for people today?
No. There is no approval marker or node lock on the canvas. Record approvals as comments on the board.
Will agents run overnight or on a schedule?
Not in the first version. An agent works when a person launches it or mentions it in the board chat.
Sources
- OWASP Top 10 for LLM Applications, LLM06:2025 Excessive Agency (root causes "excessive functionality", "excessive permissions", "excessive autonomy"; mitigation "require a human to approve high-impact actions before they are taken"): genai.owasp.org/llmrisk/llm062025-excessive-agency, checked 2026-10-21
- 8frame prices, defaults and limits: the 8frame app, checked 2026-10-21
- Upcoming 8frame features: the 8frame product roadmap, 2026-10-21
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.