Getting models

CoCo doesn’t ship with a generation model. There are three ways to get one, and you can combine them.

Read an existing model folder

If you’ve already downloaded models with another program, there’s no need to copy them — CoCo can read the original folder directly:

1. Open ⚙ Settings → Resource folders.
2. Under “Main model folder”, choose the folder that holds all your models.
3. Its subfolders are sorted by name automatically (for example Lora, VAE, checkpoints); common folder names are all recognised. If a category is wrong, change it in the table.

To add individual folders only, use “Extra folders” on the same page, by type.

Put models in CoCo’s model folder

data\models\ inside the CoCo folder is divided into subfolders by purpose. CoCo always reads them, whether or not a main model folder is set.

For generating

Folder	Purpose
checkpoints	Main models (SDXL, SD 1.5: one file is a complete model)
diffusion_models	Split main models (Anima, Z-Image, Krea 2, Qwen Image, FLUX)
loras	LoRAs
vae	VAEs
text_encoders	Text encoders
embeddings	Embeddings
controlnet	Reference image (ControlNet) models, also used by “Smoother seams” in Refine

For refining

Folder	Purpose
upscale_models	Upscale models for Upscale (the ESRGAN family, for example 4x-UltraSharp)
adetailer	Face detection models for Face fix (face_yolov8*.pt)
rembg	Background removal models (IS-Net and BiRefNet .safetensors)

For other features

Folder	Purpose
explore	The model used by Explore (finding similar images)
tagger	The WD14 model for tagging training images
joycaption	JoyCaption (GGUF) for writing descriptions of training images

Models downloaded inside CoCo (ControlNet, background removal, Explore and so on) are placed in the right folder automatically.

“CoCo’s model folder” at the top of ⚙ Settings → Resource folders lists the purpose and file count of each folder; “Open” opens it.

Download the trial model

If you have no models at all, start with the trial model:

1. Open ⚙ Settings → Resource folders → Download models.
2. Under “Trial model: Anima Aesthetic v1.1”, press “Download”. It’s an anime-style model that runs on 8 GB cards, about 5 GB.
3. When it finishes, choose anima-aesthetic-v1.1 under Models in the asset library.

Downloads only start when you press the button, come only from the model’s official page, and are verified when complete. If a download is interrupted, press “Download” again to resume where it stopped.

Split models need a VAE and text encoder

For architectures such as Anima, Z-Image, Krea 2, Qwen Image and FLUX, the main model, VAE and text encoder are separate files.

• If no VAE or text encoder is selected, CoCo looks for matching files in the asset library automatically.
• If none are found, download them under “VAE and text encoders” in ⚙ Settings → Resource folders → Download models.

The files each architecture needs are listed in the Model guide.