# Generate images and video with Agent > Ask Nodespell Agent to estimate, prepare, and start a generation after your approval, then review, compare, choose, and revise takes in the Results tab. Source: https://docs.nodespell.com/agent/generate-with-agent/ · Last updated: 2026-11-10 Generating with Agent in Nodespell means asking Agent to run part of your graph for you. Agent estimates the credit cost, prepares the exact run, and asks for your approval. After you approve, Agent starts the run, follows its progress, and collects the finished takes in the **Results** tab. There you can review, compare, choose, and revise them. ## How does a generation with Agent work? A generation with Agent always passes through the same four stages. Only the last two can spend credits, and each of those waits for you. | Stage | What Agent does | Spends credits | Needs your approval | | -------- | -------------------------------------------------------------------------------------------------------------------- | -------------- | ------------------- | | Estimate | Calculates the expected credit cost of running the target nodes | No | No | | Prepare | Checks the run plan, your balance, model access, and a credit cap, then reserves an authorization for that exact run | No | Yes | | Start | Submits the prepared run | Yes | Yes | | Follow | Tracks progress and reports the takes as they arrive | No | No | Agent's own reasoning uses credits for every request, separately from the generation. See [Sessions and models](https://docs.nodespell.com/agent/sessions-and-models/). ## Run a generation with Agent 1. Make sure the mode menu shows **Edit**. In **Ask**, Agent can estimate the cost but can't prepare or start a run. 2. Include the node you want to run, or mention it with `@`. Then describe the run, for example "Run @Hero shot twice and show me both takes." See [Context and references](https://docs.nodespell.com/agent/context-and-references/). 3. Read the estimate in Agent's reply. If you're unsure, ask "What will this cost?" before you approve anything. A prepared run can produce one to four outputs. 4. When the **Needs your approval** card appears for preparing the run, select **Approve**. Select **Deny** to stop here without spending generation credits. 5. When the second **Needs your approval** card appears for starting the run, select **Approve** to submit it. 6. Watch the run card. It moves through **Queued** and **Running** to **Completed**, **Partially completed**, **Failed**, or **Cancelled**. It also lists progress and the takes it produced. If your graph uses nodes whose cost can only be known after they run, the estimate is a lower bound. Agent must then set an explicit credit cap before it can ask for approval. Credit estimates are included in paid plans. On other plans, Agent tells you that an estimate isn't available instead of giving a number. For plans, see [Plans and credits](https://docs.nodespell.com/account/plans-and-credits/). ## Why does Agent need to prepare a run again? A prepared run is tied to the graph exactly as it was when Agent prepared it. Nodespell refuses to start the run and Agent prepares it again when any of these happen: | Situation | What happens | | ---------------------------------------------- | ----------------------------------------------------- | | You or Agent changed the graph after preparing | The run is refused. Agent must prepare and ask again. | | The estimate rose above the approved cap | The run is refused. Agent must prepare and ask again. | | More than five minutes passed since preparing | The authorization expires. Agent must prepare again. | | Your balance or model access changed | Nodespell rejects the run. Nothing is submitted. | To refresh an out-of-date estimate, ask Agent to estimate or prepare the run again on the current graph. ## Review takes in the Results tab When the conversation references media takes, the top of the panel shows two tabs: **Chat** and **Results**, with a count. **Results** opens **Session results**, a gallery of the takes from this conversation. It includes takes from runs Agent started and results Agent inspected. The gallery shows takes from nodes on the current Level. | Control | What it does | | ------------------------ | --------------------------------------------------------------------------------------------------------------- | | **Review** | Opens the take in the full-screen viewer for its media type | | **Open result** | Opens a take that has no inline preview | | **Choose take** | Makes this take the node's selected output, so downstream nodes use it | | **Revise** | Returns to **Chat** with a follow-up started for you: keep the approved work and change only what you type next | | Focus node button | Moves the canvas to the node that produced the take | | **Compare** (video only) | Marks a video for comparison | | **Compare selected** | Appears when two videos are marked, and opens them side by side | The gallery shows the 24 most recent takes. A note tells you how many older results remain in the session history. For take selection on the canvas, see [Takes](https://docs.nodespell.com/graph/takes/). For the viewer, see [Viewers](https://docs.nodespell.com/edit/viewers/). ### What is a usable partial result? A take labeled **Usable partial result** came from a run that finished only part of its work. You can review it and choose it like any other take. Ask Agent to check the run if you need the rest. > **Note:** Stopping Agent does not always stop a submitted run. Completed changes are kept, and work already submitted may still finish and use credits. Ask Agent to cancel the run, which needs your approval, or cancel it from the [Tasks panel](https://docs.nodespell.com/graph/tasks/). ## Frequently asked questions ### Will Agent start a paid generation without asking? No. Agent asks for approval before it prepares a run and again before it starts the run, in every mode. ### How many takes can Agent generate in one run? A prepared run produces between one and four outputs. Ask for the number you want, for example "make three takes." ### Why was my approved run not started? The graph probably changed, the estimate rose above the approved cap, or the five-minute authorization expired. Ask Agent to prepare the run again on the current graph. ### Does Choose take change which result the graph uses? Yes. **Choose take** makes that take the node's selected output, the same as choosing it on the canvas. ## Related - [Modes and approvals](https://docs.nodespell.com/agent/behavior-modes-and-approvals/) — which actions always ask first. - [Run nodes](https://docs.nodespell.com/graph/run-nodes/) — runs, reuse of earlier results, and estimates outside Agent. - [Takes](https://docs.nodespell.com/graph/takes/) — browse and choose results on the node. - [Tasks panel](https://docs.nodespell.com/graph/tasks/) — progress, failures, and cancelling. - [Plans and credits](https://docs.nodespell.com/account/plans-and-credits/) — what pays for generations.