Import a Custom SDXL Checkpoint or LoRA

Import a custom SDXL checkpoint

Goal

Add a compatible SDXL model to the local image library.

Before you start

The file must be:

• one regular .safetensors file;
• at least 512 MiB;
• a valid SafeTensors container;
• an SDXL architecture;
• stored on a disk with enough space for the file plus import overhead.

.ckpt is not supported by the 1.7.3 importer.

Steps

1. Open Settings → Models → Image Models.
2. Expand Advanced options.
3. Select Import SDXL checkpoint.
4. Choose the .safetensors file.
5. Wait for validation and copying to finish.
6. Select the imported model.
7. Run a test generation before enabling LoRAs.

Expected result

The model appears under Your models and is available as an SDXL base.

If something goes wrong

Errors for a tiny file, malformed/truncated header, no tensors, wrong architecture, checksum mismatch, or insufficient disk space mean the file should not be used. Download a verified SDXL SafeTensors checkpoint from a trusted source.

Import and activate a LoRA

Goal

Apply a compatible SDXL LoRA to an SDXL base model.

Before you start

• Use one .safetensors file of at least 1 MiB.
• Select an SDXL base first.
• FLUX and non-SDXL models cannot use this LoRA path.
• At most three LoRAs can be active on one model.

Steps

1. Select the target SDXL model.
2. Open its LoRA controls.
3. Import the LoRA .safetensors file.
4. Add up to eight trigger words if needed.
5. Enable the LoRA.
6. Start near the default weight 0.8.
7. Adjust between 0.0 and 1.5 in small steps.
8. Run a test render.

Expected result

The LoRA affects renders from the selected SDXL model.

If something goes wrong

• A family marked unknown may be imported but cannot be activated safely.
• Disable another LoRA if three are already active.
• Lower the weight if the image is overcooked or distorted.
• Header metadata cannot guarantee compatibility; the final check is a successful test render.