LoRA and embeddings

Adding a LoRA

1. Click the “LoRA” slot on the left of the generate panel to open the LoRA tab of the asset library.
2. Click a LoRA to add it (a <lora:name:weight> tag appears in the prompt); click again to remove it.
3. Drag the “Weight” slider right on the card.

Writing <lora:name:0.8> in the prompt yourself works the same way.

The library filter “Only show ones that work with SDXL” (following the current main model) hides LoRAs that don’t fit.

When the architecture doesn’t match

A LoRA is trained for one architecture: SDXL LoRAs don’t work on Anima, and FLUX.1 LoRAs don’t work on SDXL.

• Mismatched LoRAs are marked “Incompatible” on their library card and on their tag in the prompt.
• You can still generate; that LoRA is skipped and the log says so.

Trigger words and notes

Click a LoRA card in the asset library for its details:

• Trigger words: read from the LoRA file. If they’re wrong, press “Edit” and enter one per line; after editing, “Use the file’s default” restores the default (the trigger words recorded in the file).
• Notes: your own notes, such as good weights or which model it pairs with.
• Change cover: by default it’s an image from your library that used this LoRA; you can upload one instead.
• “See the N images that used it” jumps to the library, filtered to images that used it.

LoRAs trained in CoCo get their trigger words recorded automatically.

Embeddings

Put embeddings (textual inversion) in the embeddings folder, then write the file name in the prompt or negative to apply it, for example lazyhand in the negative.

The Embeddings tab of the asset library has “Add to prompt” and “Add to negative”. Embeddings that don’t match the current model’s architecture aren’t applied, and you’ll get a notice when generating.

If a LoRA fails to load

• LyCORIS (LoHa, LoKr) isn’t supported yet.
• If diffusers fails to load an ordinary kohya LoRA, CoCo automatically merges it another way and notes why in the log.