> For the complete documentation index, see [llms.txt](https://help.intangible.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.intangible.ai/learn-more/faq/troubleshooting-the-agent.md).

# Troubleshooting Intangible agents

What to do when Intangible agents say they are done before the work is on the canvas, change the wrong scene, skip your questions, stall on a step or spend more credits than you expected.

Most problems in the chat come from one of three places: the status line and the work on the canvas disagree, the agents picked a different card from the one you meant, or a run is waiting on you. Start by fixing the problem where it began.

## Fix it where it starts

A wrong result is easier to fix at its source than render by render. Work out where the problem began, then fix it there.

* **The idea is wrong** (the wrong product, audience or location, a shot you no longer need): edit [the brief](https://help.intangible.ai/canvas/the-brief.md#change-the-idea-after-you-confirm). Editing it rebuilds nothing on its own, so then ask in the chat panel to update what depends on it, for example "I changed the location in the brief to a rooftop. Update the scene and the shots to match."
* **Something is wrong in every shot** (an object's size, a missing prop, the wrong reference on a subject): fix the scene. Change it by hand in [Build](https://help.intangible.ai/build/build.md), or ask in the chat panel. Renders you make next follow the scene; renders you already made stay as they are.
* **One output is wrong** (a stray object in one frame, a colour in one render): mark it up with [Edit](https://help.intangible.ai/canvas/edit-media.md) and generate it again. The edited copy lands next to the original, which stays as it was.

```mermaid
flowchart TD
  P[A result is wrong] --> Q{Where did it start?}
  Q -- the idea --> B[Edit the brief, then ask the agents to update]
  Q -- every shot --> S[Change the scene in Build, or ask in the chat]
  Q -- one output --> E[Edit that card and generate again]
```

## The status says done, but the canvas is missing pieces

**Symptom**: the status says the run is done, yet the models, scene or cameras you expected are missing from the canvas.

**Cause**: the status line reads "All done! How does it look?" whenever the agents' turn ends, whether or not every step ran, and renders they started can still be generating on the canvas. In one test run the status said done while the step list was still on its first step, and no models or cameras had been built.

**Fix**: check two things before you move on.

1. **The step list.** It sits under the status; unfold it if you hid it. A step not marked done is still queued or running.
2. **The canvas.** Look for the cards each step should have made: reference sheets in the subject groups, the scene card, a shot card for each shot.

If steps are missing, ask the agents to continue in plain words, for example "Continue with the next step: build the plaza, cafe and robot in Scene 1." See [Status and steps](https://help.intangible.ai/chat-agents-intangible/status-and-steps.md).

## A step stays queued

**Symptom**: the step list stops on a step and nothing new lands on the canvas.

**Cause**: usually the agents are waiting on you. A status of "waiting for your answer" means a choice card in the chat panel needs a click. A step such as "Frame the five shots for your approval" is a point where the agents stop on purpose. After a build they also stop and ask what to do next, with render options on a choice card; nothing is rendered until you pick one.

**Fix**: scroll the chat panel to the latest choice card and click an option, or reply to the step that is waiting on you. The **Running** row with a line of flavour text ("Herding pixels into position.") only shows the run is alive; it does not mean a step is moving.

## A run stops

**Symptom**: the status has not changed for a long time, and no step is spinning, or the status went back to "ready when you are" before the work was done.

**Fix**: read the last line in the thread first.

* **"That didn't go through. Anything already on the canvas stays."** The request failed. Click **Try again** under it to send the same message once more.
* **"Connection lost before I finished. Send "continue" and I'll pick up where I stopped."** The connection dropped, from a network drop, a refresh or a closed tab. Send `continue`; the agents do only the work that is left.

Otherwise, ask the agents where they got to and to carry on, for example "What is left on the step list? Continue from there." If a result came back wrong, undo it and send a more specific request. See [Tips for directing agents](https://help.intangible.ai/chat-agents-intangible/limits.md).

A job that fails is refunded automatically.

## Something is missing from the canvas

**Symptom**: the agents said they made something, but you cannot find the card.

**Fix**:

* **Zoom out** (scroll, or pinch on a trackpad) to see the whole project.
* **Look in the right group.** Reference sheets land in their subject's group, shots in their scene's group, renders in **Visualizations**, and images you attached in the chat panel in **Sources**, until the agents sort them. See [Groups and cards](https://help.intangible.ai/canvas/groups-and-cards.md).
* **Check the step list** in case the step that makes it has not run yet (see above).
* **The view does not follow new cards.** The canvas does not move to show a new render or anything the agents make. It only scrolls to a card you start with Generate on the canvas itself.
* **A model you built in Build** gets a card only once it has a reference image attached, or you turn on **Promote to Canvas** in its More options. See [What Intangible agents build](https://help.intangible.ai/canvas/what-intangible-agents-build.md).

## The agents built without asking

**Symptom**: the project opened and the agents started building without asking you any questions or waiting for you to confirm a brief.

**Cause**: you chose **Skip** ("Skip all approvals") on Home. Skip skips the guiding questions and the approvals: the agents build straight from your prompt, and they may start a render without asking. A brief can still land on the canvas, because the agents write it as they build.

**Fix**: to plan first next time, choose **Guide me** on Home before you click **Create project**. On this project, tell the agents what to change in the chat panel, naming the card. See [Home](https://help.intangible.ai/canvas/home.md) and [The brief](https://help.intangible.ai/canvas/the-brief.md).

## The brief is wrong

**Symptom**: the brief has the wrong subject, tone or shot count.

**Fix**: do not confirm it yet. On the choice card, pick the option to edit it first, then click a section's text to edit it (Cmd/Ctrl+Enter or click away to save). Or tell the agents what to change in the chat panel. Read anything marked *(assumed)* first: those are the agents' guesses. Only editors and owners of the project can edit the brief. See [The brief](https://help.intangible.ai/canvas/the-brief.md).

## The agents changed the wrong scene or card

**Symptom**: you asked for a change and it landed on a different scene, shot or render.

**Cause**: the agents work from the scene and mode you are in, and from the names in your message. A loose name ("the keyframe", "the car") can match more than one card.

**Fix**: undo the change, then ask again with the scene and the card named the way the canvas names them: "In scene 01, make shot 02's keyframe dusk." See [Chat with Intangible](https://help.intangible.ai/chat-agents-intangible/chat-with-intangible.md#point-at-what-you-mean).

## I can't attach a canvas card to my message

**Symptom**: you want to point the agents at a card, but there is no way to attach it, and selecting it on the canvas does not tell the chat panel about it.

**Cause**: canvas cards cannot be attached to a chat message. The chat panel **+** attaches images only.

**Fix**: name the card in the message, for example "shot 02's keyframe" or "the crater reference". For several shots, name them all in one message: "Render shots 02, 03 and 05 as video, each from its keyframe and scene."

## A video costs more than expected

**Symptom**: a video the agents started used more credits than you planned.

**Cause**: when you set up a large video in the Visualize panel, a notice above **Generate** shows its credit count before you click. A video the agents start goes straight to the job, so you never see that notice. Before they start, the agents quote the credits and wait for you to agree; the panel's notice is the only check that does not depend on them.

**Fix**: read the credits the agents quote before you agree, or ask for them, for example "What would shot 02 cost as a 5-second video?" Or render the video yourself from Visualize, where the notice shows the credits. See [Keyframes to video](https://help.intangible.ai/canvas/keyframes-to-video.md#credits) and [Credits and tokens](https://help.intangible.ai/learn-more/concepts/credits-and-tokens.md).

## The video does not move the way the scene does

**Symptom**: a video from the agents ignores the camera move or the motion in your scene, or comes back at a different length from the shot.

**Cause**: a video made from a render animates that still image, and the model invents the motion; it does not follow the scene's animation. The agents follow the scene's animation when the shot has one, and work from the render when it doesn't, so a shot whose move was never keyframed gets invented motion.

**Fix**: check the shot has its camera move or animation on the timeline, then ask for it by name: "Follow the scene's animation for shot 03." Then open the shot in Visualize, where the agents write the settings they used, and change the model, mode or duration yourself before rendering again. See [Generate video](https://help.intangible.ai/visualize/generate-video.md).

## The agents can't find my imported model

**Symptom**: "drop in the green Bronco" does nothing, or places a library asset instead.

**Cause**: the agents match imported models by their names and descriptions. A model named "object-12" does not answer to "the green Bronco".

**Fix**: rename and describe the model in [Object details](https://help.intangible.ai/build/object-details.md), then ask again. See [Tips for directing agents](https://help.intangible.ai/chat-agents-intangible/limits.md).

{% hint style="info" %}
**Still stuck, or think it's a bug?** Report it from inside Studio: click the help icon (?) in the top-right toolbar and pick **Report an Issue**. Describe it, click **Send report** and share the Studio tab for a screenshot. See [Report a bug or get help](https://help.intangible.ai/get-started/help-and-feedback.md).
{% endhint %}

## Related

* [Status and steps](https://help.intangible.ai/chat-agents-intangible/status-and-steps.md)
* [Tips for directing agents](https://help.intangible.ai/chat-agents-intangible/limits.md)
* [Known limitations](https://help.intangible.ai/learn-more/faq/known-limitations.md)
* [Common questions](https://help.intangible.ai/learn-more/faq/common-questions.md)
* [Report a bug or get help](https://help.intangible.ai/get-started/help-and-feedback.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://help.intangible.ai/learn-more/faq/troubleshooting-the-agent.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
