Let AI Write Your Step Descriptions

This needs an AI provider connected, and it needs a model that can read images. A text-only model cannot see a screenshot, and Folge stops and tells you rather than filling the guide with guesses. gpt-4o and claude-sonnet-4-5 both work. See Set Up AI in Folge.

1. Select the step.
2. Click the wand button in the description toolbar.Folge sends that step image, its title and the titles around it, then writes the result into the description.

The first time, Folge asks you to confirm that the screenshot is about to be sent to your provider. It remembers your answer.

The wand in the editor header does every step in the guide.

The first time you open it, Folge explains what it is about to do with your screenshots before it does anything.

After that it goes straight to the choice that matters.

• Only steps without a description: Fills the blanks and leaves everything you have written alone. This is the one to re-run after adding steps to a guide you already described, because it skips the steps it has done.
• All steps, replacing existing descriptions: Rewrites everything, including your own text.

Both counts only include steps that have a screenshot. Steps without one are skipped, and the dialog says how many. Under the buttons is an estimate of what the run will cost: a price on OpenRouter, a token count elsewhere. Press Generate, and Stop if you change your mind partway.

The AI sees the step image, the step title, the guide's own title and description, and the titles of the steps before and after. Everything that makes those better makes the descriptions better.

• Give the guide a real title and description: Folge passes them along. "Invite a teammate to Northwind Ops" tells the AI far more than "Untitled guide".
• Capture with Automated Annotations on: Folge then records which element you clicked, so the AI knows the button was called Save rather than having to read it off the pixels. See Capture a guide automatically.
• Fix the titles first: The title goes to the AI with the picture. Correct a wrong one before you generate, or the description gets built on it.
• Say how you want it written: Whatever you put in Custom instructions in AI settings is added to every request Folge makes, so it steers all of the descriptions at once. See below.

Folge sends its own prompt with every request. Whatever you type into Custom instructions, in AI settings, is added to that prompt, so it applies to every description in a run without you repeating yourself.

• Another language: Always use Ukrainian language, and the descriptions come back in Ukrainian whatever language the screenshots are in.
• A house style: Write in the imperative. One sentence per step. No exclamation marks.
• Your own vocabulary: Call it the navigation panel, never the sidebar.
• A length: Keep every description under 25 words.

It is stored with your AI settings rather than with the guide, so it stays in force for every guide until you change it. Set it before you press Generate: changing it afterwards does not rewrite what has already been written.

Folge sends the step image as it looks in the editor. Annotations you drew are on it, and so is anything you hid: a blurred or redacted area reaches the provider blurred, because it is the same picture.

Nothing is sent until you press the button, and the request goes from your computer to the provider you connected, billed to your account. If the screenshots show systems that must not leave the building, connect a local model instead: Set Up AI in Folge covers it.

The AI describes what it can see. It cannot know that the second field is optional, that the approval takes two days, or that your team calls it something else. Those are the sentences worth adding by hand once the draft is there.

Read every generated description before the guide goes out.