AI Image Generation

When a page needs a feature image and you do not have one, you can describe the picture instead of finding it. Leed sends your description to an image model together with your workspace’s standing visual guidance, returns three candidates, and lets you push the one you like in a direction until it is right. Saving it uploads it as an ordinary image asset and sets it as the page’s feature image in the same step.

Where it lives — and it is only in one place

Open a page in the editor, open the right rail’s Page Settings, and look at the Page Details section. Beside the Feature Image label there is a small sparkle button. That button is the entire feature. There is no Generate control in the Assets workspace, no way to generate an image into your library on its own, and no way to regenerate an image you have already saved.

Describing what you want

The Generate Feature Image dialog has three inputs, and only the last is required.

Company Theme is your workspace’s standing description of how your pictures should look — the same text every generation starts from, so you never have to retype your brand. It is prefilled from the AI Image Theme Guide field in your workspace’s shared content settings, and it is read-only until you flip the Edit switch beside it. Edits you make here apply to this generation only; they are never written back to the workspace setting. To change it for everyone and for good, edit it in Default Content Configuration.

Palette (optional) is a collapsible list of four fixed swatches. Tick any of them and their hex values are named in the instruction as a color scheme. There is no color picker and no way to add a fifth.

Prompt is the description itself, and the Generate button stays disabled until you type something into it. Describe the picture, not your brand — the theme is already carrying that.

The three are concatenated into a single instruction before anything is sent:

<Company Theme> <your prompt> Color scheme: #7C3AED, #0F766E

The color clause is appended only when you tick at least one swatch. This is why repeating brand details in the prompt is wasted effort, and why a very long Company Theme leaves less room for your own description to matter.

Three candidates, generated at once

Pressing Generate opens the result view and immediately runs three independent generations from that one instruction. They fill in as they finish, each slot showing Generating... until its image arrives and Error generating image if that one call fails — one slot failing does not stop the other two.

Click any candidate to select it. The selected image enlarges on the left, the other two stay in a column beside it, and clicking a different one switches your selection at any time. Nothing is saved by selecting.

Every generated image comes back at 1024 × 576 — a 16:9 frame. There are no size, aspect-ratio or quality controls, and the model is not a choice you make.

Refining the one you picked

With a candidate selected, four controls appear beneath it. They all operate on the selected slot alone.

The adjustment box is the part people miss: typing in it does nothing until you press More this, Less that or Try again. It is cleared when you select a different candidate, so an adjustment never leaks onto an image you were not looking at.

Here is the whole run, with both regeneration paths on it:

flowchart TD
  T["AI Image Theme Guide<br/>(Company Theme)"] --> I
  P["Your prompt<br/>(required)"] --> I
  C["Palette swatches<br/>(optional)"] --> I
  I["One instruction"] --> G1["Candidate 1"]
  I --> G2["Candidate 2"]
  I --> G3["Candidate 3"]
  G1 --> SEL
  G2 --> SEL["Click one to select it"]
  G3 --> SEL
  SEL --> R{"Which button?"}
  R -->|"More this / Less that / Try again"| ONE["Regenerate the selected slot only"]
  ONE --> SEL
  R -->|"Generate, after editing the prompt"| I
  R -->|"Save Image"| FN["Set File Name"]
  FN --> A["New image asset in your library"]
  A --> F["Set as this page's feature image"]

Saving it

Save Image opens a small Set File Name dialog holding a default of ai_generated_image.png. Change it or accept it, press Save, and two things happen in order.

First, the image is uploaded as an ordinary image asset — the same upload path as any file you drag in, so it appears in your library, can be labeled and approved, and can be placed on other pages. Its Name and Filename are both the file name you just typed, and its Alt Text is set to the literal string Generated image.

Second, that asset is set as the page’s feature image, with the Feature Image Alt Text taken from the asset’s name — which means the alt text on the page defaults to your file name, not to a description of the picture.

Because the result is a normal asset, everything on Asset Details and Usage applies to it — labels, marketing approval, the list of pages it is used on, and the rule that stops you deleting it while a page still references it.

What image generation does not do

• It does not generate into your library. The only way to keep an image is to save it as the current page’s feature image; you can reuse it elsewhere afterwards, but the first destination is fixed.
• It does not regenerate a saved asset. Once saved, an image is a file. To try again, open the dialog again and generate a fresh set.
• It does not remember anything. There is no history of past prompts, no seed you can reuse, and closing the dialog discards the three candidates permanently.
• It does not read your page. Leed’s API can draft an image prompt from a page’s own content, but no button in the CMS calls it — what the model sees is the theme, your prompt and your swatches, and nothing else.
• It does not use a page-type-level theme. The dialog seeds Company Theme from the workspace setting only.

This is one of several AI helpers built into the editor — the others write summaries, keywords, alt text and draft copy, and none of them is plan-gated either; AI in the Editor indexes them. For the wider picture of where AI appears across the whole product, including the parts that run without you asking, see AI in Leed.