Generating images

Plank can create images for you — a cover for a deck, an illustration for a report, a background for a title slide, a concept frame for a pitch. The assistant writes them straight into your workspace as files, so the deck or document it is building can use them immediately.

It is off until you turn it on, because unlike everything else the assistant does, each image has a real, visible price.

Turning it on

Go to Settings → Skills, find Image Generation under Visuals & Media, and install it.

That is the whole switch. The skill is both the permission and the instructions: while it is installed the assistant can generate images and knows how; while it is not, it cannot, and it will tell you so rather than trying. You can disable it again from the same screen at any time, and re-enable it later without losing anything.

It applies to one workspace at a time. Turning it on in your Marketing workspace does not turn it on in Finance. This is deliberate — images are charged to the plan that pays for that workspace, so the decision belongs with the workspace, not with your account.

Asking for an image

Just say what you want in the chat, in as much detail as you can:

Make a cover image for the Q3 deck — a wide, cinematic photo of an empty modern office at dawn, warm light, muted colours, no people and no text.

The more specific you are, the fewer attempts it takes, and each attempt costs money. Some things that reliably help:

• Describe the shot, not just the subject. "Wide, cinematic, low warm light, shallow depth of field" gets you much further than "a nice office photo".
• Say what should not be there. These models like to put words into pictures, and the words are usually misspelled. If the image sits under a headline, ask for "no text".
• Say the colours, so the image sits with your deck instead of fighting it.
• Leave room. For a title background, ask for empty space where the headline goes.
• Be careful with people. Faces come out slightly wrong at small sizes and date quickly. Places, objects, textures and abstractions age better.

Where the images go

Into your workspace, as ordinary files, next to the thing they belong to:

Q3 report/
  index.html
  assets/
    cover.png

You can open them, download them, replace them, or delete them like any other file. The deck builder and the document templates reference them by their position beside the deliverable, so moving a deliverable moves its images with it.

What it costs

Images are charged to the same monthly AI allowance as everything else the assistant does — see AI usage and limits. There is no separate bill and nothing extra to buy.

What is worth knowing is the scale. A page of writing costs a fraction of a cent. One image costs a few cents — roughly a hundred times more. That is not a problem in itself; a dozen images for a deck is a few tens of cents. It is only a problem if it happens without you meaning it to, which is why the feature is off by default and why there are two limits:

• Four images per request, so one instruction cannot turn into ten pictures.
• Fifty images per hour per workspace. Past that, the assistant is stopped and told to wait. If you genuinely need more, do it in batches across the hour.

If your workspace reaches 100% of its monthly allowance, image generation stops along with everything else until the allowance resets or the plan is upgraded.

Ask for the cheaper option when quality does not matter. A background sitting behind a block of text does not need the best model. Say "use the cheap model" or "low quality is fine" and the assistant will — it costs about a fifth as much.

What it will not do

• Edit an existing image. Cropping, resizing, recolouring, adding a logo, combining two pictures — the assistant can already do all of that, for free, without this feature. Ask for it directly.
• Make video. Not supported.
• Reproduce a real person, a real logo, or a copyrighted character. Ask for those and you will get something that is nearly right and legally awkward. Use the real asset.
• Produce text inside the image reliably. Put the words on the slide, over the image, where they will be spelled correctly and can be translated.

Using your own Higgsfield account instead

Some people prefer Higgsfield for cinematic stills and video. You can use it inside Plank, and if you do, it is billed to your own Higgsfield account, not to your Plank AI allowance. Plank is not involved in the payment at all — it does not resell Higgsfield, does not hold a Higgsfield account on your behalf, and cannot see what you spend there.

Ask the assistant to set it up. It will install the Higgsfield CLI in your workspace and then hand the login back to you, because only you can log into your own account.

Where the login is kept matters. Ask the assistant to store the Higgsfield credential in your private home directory inside the workspace — never in the shared workspace folder. Anything in the shared folder is readable by every member of that workspace and by the assistant on their behalf; a credential there is a credential you have given to everybody. Your home directory is yours alone. If you are ever unsure where something landed, ask the assistant to show you the path before you log in.

Two consequences of it being your account, both worth knowing up front:

• It does not follow you. The login lives in that one workspace, for you. A colleague in the same workspace has to connect their own account.
• Plank cannot help with it. Rate limits, failed generations, billing questions and outages are between you and Higgsfield.

Everything on this page above this section — the allowance, the caps, the per-request limit — applies to Plank's built-in image generation only. The Higgsfield path has its own limits, set by your Higgsfield plan.