Prompts and tag mode

The top row of the prompt editor has Prompt, Negative, Quality tags and Tags.

Two ways to write

• Plain text: type the prompt directly.
• Tag mode: press “Tags” to switch. Each tag gets its own chip that you can drag to reorder, reweight, or double-click to edit. Chips are coloured by category (hair, clothing, expression, scene and light…), so you can see at a glance what you’ve written.

Both modes edit the same prompt, so switching back and forth never loses anything.

Finding tags

In either mode, suggestions pop up as you type:

• English autocompletes: typing long lists long hair, long sleeves…
• Chinese works too: typing 雙馬尾 lists twintails. About 45,000 tags have Chinese translations.

Use ↑ ↓ to choose and Enter to insert.

Weights

• Tag mode: each chip has “Weight −0.1” and “Weight +0.1”.
• Plain text: the usual (blue eyes:1.2).

Turning a tag off for now

To see what happens without a tag but keep it around, press the chip’s ⊘ “Hide for now” in tag mode.

• A tag that’s off isn’t generated and is hidden in plain text mode.
• Back in tag mode it stays, dimmed, in its original place; press again to turn it back on.

Whole dynamic syntax groups ([[ ]], { }) can be turned off the same way.

Quality tags and the default negative

Each model has recommended quality tags (such as masterpiece, best quality) and a default negative prompt, added when you generate but not shown in the prompt box. Tags you already wrote aren’t added twice.

• Hover over “Quality tags” to see what’s added and whether it goes at the start or the end.
• Click it to turn it off (remembered); click again to turn it on.
• To change them, open the model in the asset library and edit “Quality tags & default negative”.

Token count

The bottom-left corner of the editor shows the token count, such as 54/77. With SDXL’s “native” encoding, anything past the limit is cut off and the number is highlighted; the default encoding splits long prompts into chunks instead of cutting them.