Prompting
The prompt card
A place to write before you know which tool wants it, and a way to reuse the same brief across five cards without retyping.
A prompt card is a piece of writing that lives on the canvas rather than inside one tool. Double-click empty canvas and pick **Prompt** from the ring. The reason it exists is sequencing. You often know what you want before you know which tool is going to make it, and typing that into the first card you happen to open buries it there. ## Dragging it into a tool Drag a prompt card onto any tool's prompt box and its text drops in. The card snaps back to where it was, so the same brief can feed a second tool, and a third. The workflow it is built for: 1. Write the brief once, on the canvas 2. Drag it into **text-to-image**, run 3. Drag the same card into **edit-images** with a reference, run 4. Compare, without having retyped a word Anything typed twice drifts. Two prompts that were meant to be identical and differ by one adjective produce two results you cannot honestly compare. ## Writing one that works Be concrete about the things a model can act on: ```text title="A prompt with something to hold onto" a stacking chair in cast aluminium, raw mill finish, single soft window light from the left, 85mm, plain grey backdrop ``` Subject, material, finish, light, lens, background. Every one of those is checkable in the result, which means a disappointing run tells you which part to change. :::tip[Leave the style words until after the first pass] Adjectives like "cinematic", "stunning" and "beautiful" tend to overwhelm the specific things you asked for, and they are the hardest part to debug. Get the subject and the light right first, then add mood. ::: ## Two things you can put inside a prompt **`@` mentions**, on tools with more than one image slot, address a specific input: `put @image-2's logo on @image-1`. **`{ }` variables** turn one prompt into a batch. Describe a list inside the braces and one press fires one run per item. See [variables and batch runs](/docs/using-plnty/using-a-tool/variables-and-batch-runs/). ## When the words are the problem **prompt-enhancer** takes a rough line and returns a fuller brief, which you then edit. It has a 3D mode for when the target is a mesh rather than a picture. The output is a starting point rather than an answer. Read it, cut the half that is generic, and keep the half that named something you had not thought to specify. :::note[The prompt is stored with the result] Whatever ran is written into the asset's record, so the exact words are recoverable months later from the thing they produced. That is also why the batch substitutes per run: each of twenty results carries its own resolved sentence rather than the instruction that generated the list. :::A prompt card is a piece of writing that lives on the canvas rather than inside one tool. Double-click empty canvas and pick Prompt from the ring.
The reason it exists is sequencing. You often know what you want before you know which tool is going to make it, and typing that into the first card you happen to open buries it there.
Dragging it into a tool
Section titled “Dragging it into a tool”Drag a prompt card onto any tool’s prompt box and its text drops in. The card snaps back to where it was, so the same brief can feed a second tool, and a third.
The workflow it is built for:
- Write the brief once, on the canvas
- Drag it into text-to-image, run
- Drag the same card into edit-images with a reference, run
- Compare, without having retyped a word
Anything typed twice drifts. Two prompts that were meant to be identical and differ by one adjective produce two results you cannot honestly compare.
Writing one that works
Section titled “Writing one that works”Be concrete about the things a model can act on:
a stacking chair in cast aluminium, raw mill finish,single soft window light from the left, 85mm, plain grey backdropSubject, material, finish, light, lens, background. Every one of those is checkable in the result, which means a disappointing run tells you which part to change.
Two things you can put inside a prompt
Section titled “Two things you can put inside a prompt”@ mentions, on tools with more than one image slot, address a specific input:
put @image-2's logo on @image-1.
{ } variables turn one prompt into a batch. Describe a list inside the braces and
one press fires one run per item. See variables and batch
runs.
When the words are the problem
Section titled “When the words are the problem”prompt-enhancer takes a rough line and returns a fuller brief, which you then edit. It has a 3D mode for when the target is a mesh rather than a picture.
The output is a starting point rather than an answer. Read it, cut the half that is generic, and keep the half that named something you had not thought to specify.