Configure Fooocus: Model Paths and Defaults via config.txt

On the very first launch, Fooocus writes a config.txt file to the root of the Fooocus directory. This file is standard JSON and controls everything from where Fooocus looks for model files to the default sampler, styles, and prompt that appear when the UI opens. Editing config.txt is the primary way to customize Fooocus without touching any source code.

A companion file, config_modification_tutorial.txt, is regenerated on every launch. It lists every recognized key with its current value and examples. Use it as a live reference — but make your actual edits in config.txt, not the tutorial file.

Editing and Resetting

Open config.txt

Navigate to your Fooocus installation folder and open config.txt with any plain-text editor. The file is UTF-8 encoded JSON.

Edit values

Change the values you need, following the JSON format rules described below. Save the file when done.

Restart Fooocus

Fooocus reads config.txt at startup. Restart the application for changes to take effect.

Resetting to defaults

If something breaks, simply delete config.txt. Fooocus will regenerate it with default values on the next launch.

Path Settings

These keys tell Fooocus where to find (or place) its various model assets. All values must be absolute paths.

On Windows, use \\\\ (double backslash) as the path separator inside JSON strings, not a single \. For example: "D:\\\\Fooocus\\\\models\\\\checkpoints".

Key	Default Location	Description
`path_checkpoints`	`../models/checkpoints/`	Directory (or list of directories) for SDXL checkpoint files
`path_loras`	`../models/loras/`	Directory (or list of directories) for LoRA files
`path_embeddings`	`../models/embeddings/`	Textual inversion embedding files
`path_vae_approx`	`../models/vae_approx/`	Approximate VAE models used for fast preview
`path_vae`	`../models/vae/`	Full VAE models
`path_upscale_models`	`../models/upscale_models/`	Upscale model files
`path_inpaint`	`../models/inpaint/`	Inpaint head and patch files
`path_controlnet`	`../models/controlnet/`	ControlNet and IP-Adapter models
`path_clip_vision`	`../models/clip_vision/`	CLIP vision encoder models
`path_fooocus_expansion`	`../models/prompt_expansion/fooocus_expansion`	GPT-2 prompt expansion model
`path_outputs`	`../outputs/`	Directory where generated images are saved
`path_wildcards`	`../wildcards/`	Wildcard `.txt` files for inline prompt substitution

Both path_checkpoints and path_loras accept a list of directories, allowing you to point Fooocus at multiple locations simultaneously:

{
    "path_checkpoints": ["D:\\\\MyModels\\\\checkpoints", "E:\\\\SharedModels"],
    "path_loras": ["D:\\\\MyModels\\\\loras"]
}

Default Model Settings

Key	Type	Default	Description
`default_model`	string	`"juggernautXL_v8Rundiffusion.safetensors"`	Base checkpoint loaded at startup
`default_refiner`	string	`"None"`	Refiner checkpoint (`"None"` to disable)
`default_refiner_switch`	float (0–1)	`0.5`	Sampling step at which to swap to the refiner
`previous_default_models`	list of strings	`[]`	Fallback model names checked when the primary model file is not found
`default_loras`	list of `[bool, string, float]`	5 disabled LoRAs	Active LoRA slots. Each entry is `[enabled, filename, weight]`
`default_loras_min_weight`	float (-10–10)	`-2`	Minimum allowed LoRA weight in the UI
`default_loras_max_weight`	float (-10–10)	`2`	Maximum allowed LoRA weight in the UI
`default_max_lora_number`	int (≥ 1)	`5`	Number of LoRA slots shown in the UI
`default_vae`	string	`"Default (model)"`	VAE override; `"Default (model)"` uses the VAE baked into the checkpoint

Generation Defaults

These keys set the initial values of controls in the main generation panel.

Key	Type	Default	Description
`default_performance`	string	`"Speed"`	Startup performance mode (`"Quality"`, `"Speed"`, `"Extreme Speed"`, `"Lightning"`, `"Hyper-SD"`)
`default_sampler`	string	`"dpmpp_2m_sde_gpu"`	Default sampler
`default_scheduler`	string	`"karras"`	Default noise scheduler
`default_cfg_scale`	float	`7.0`	Guidance scale (CFG)
`default_sample_sharpness`	float	`2.0`	SAG-based sharpness enhancement weight
`default_aspect_ratio`	string	`"1152*896"`	Default output resolution in `width*height` format
`default_image_number`	int (1–`default_max_image_number`)	`2`	Number of images generated per run
`default_max_image_number`	int (≥ 1)	`32`	Maximum allowed image count in the UI
`default_prompt`	string	`""`	Pre-filled positive prompt
`default_prompt_negative`	string	`""`	Pre-filled negative prompt
`default_styles`	list of strings	`["Fooocus V2", "Fooocus Enhance", "Fooocus Sharp"]`	Style presets enabled at startup
`default_overwrite_step`	int	`-1`	Override sampling steps regardless of performance mode (`-1` = follow mode)
`default_overwrite_switch`	int	`-1`	Override the refiner switch step (`-1` = follow preset)
`default_clip_skip`	int (1–12)	`2`	CLIP skip depth
`default_cfg_tsnr`	float	`7.0`	CFG TSNR correction threshold (applied when CFG > 10)
`default_output_format`	string	`"png"`	Output format: `"png"`, `"jpeg"`, or `"webp"`

UI Defaults

Key	Type	Default	Description
`default_advanced_checkbox`	bool	`false`	Whether the Advanced panel is open on launch
`default_image_prompt_checkbox`	bool	`false`	Whether the Image Prompt panel is open on launch
`default_enhance_checkbox`	bool	`false`	Whether the Enhance panel is open on launch
`default_developer_debug_mode_checkbox`	bool	`false`	Whether Developer Debug Mode is active on launch
`default_save_metadata_to_images`	bool	`false`	Embed generation metadata into output images
`default_metadata_scheme`	string	`"fooocus"`	Metadata format: `"fooocus"` (JSON) or `"a1111"` (plain text)
`metadata_created_by`	string	`""`	Optional author name embedded in image metadata
`default_black_out_nsfw`	bool	`false`	Automatically black out images flagged by the safety checker
`temp_path_cleanup_on_launch`	bool	`true`	Delete the temp folder at startup

Download Settings

Fooocus can automatically download missing model files when a preset requires them. Specify files as a JSON object mapping filename → URL.

Key	Type	Description
`checkpoint_downloads`	object	Checkpoint files to download if absent
`lora_downloads`	object	LoRA files to download if absent
`embeddings_downloads`	object	Embedding files to download if absent
`vae_downloads`	object	VAE files to download if absent

Full Example

The following is a representative config.txt that overrides model paths to custom drives, changes the default model and sampler, and sets a negative prompt:

{
    "path_checkpoints": "D:\\\\Fooocus\\\\models\\\\checkpoints",
    "path_loras": "D:\\\\Fooocus\\\\models\\\\loras",
    "path_embeddings": "D:\\\\Fooocus\\\\models\\\\embeddings",
    "path_vae_approx": "D:\\\\Fooocus\\\\models\\\\vae_approx",
    "path_upscale_models": "D:\\\\Fooocus\\\\models\\\\upscale_models",
    "path_inpaint": "D:\\\\Fooocus\\\\models\\\\inpaint",
    "path_controlnet": "D:\\\\Fooocus\\\\models\\\\controlnet",
    "path_clip_vision": "D:\\\\Fooocus\\\\models\\\\clip_vision",
    "path_fooocus_expansion": "D:\\\\Fooocus\\\\models\\\\prompt_expansion\\\\fooocus_expansion",
    "path_outputs": "D:\\\\Fooocus\\\\outputs",
    "default_model": "realisticStockPhoto_v20.safetensors",
    "default_refiner": "None",
    "default_loras": [[true, "lora_filename_1.safetensors", 0.5], [true, "lora_filename_2.safetensors", 0.5]],
    "default_cfg_scale": 3.0,
    "default_sampler": "dpmpp_2m",
    "default_scheduler": "karras",
    "default_prompt_negative": "low quality",
    "default_prompt": "",
    "default_styles": [
        "Fooocus V2",
        "Fooocus Photograph",
        "Fooocus Negative"
    ]
}

Environment Variable Overrides

Every config key can also be set as an environment variable. Environment variables take priority over values in config.txt. This is useful for containerized or scripted deployments where you want to avoid editing files.

# Override the output path at launch time
path_outputs=/mnt/storage/fooocus-outputs python entry_with_update.py

Fooocus will print Environment: key = value to the console for every key it picks up from the environment.

If you are running Fooocus in Docker or a cloud environment, environment variable overrides are the cleanest way to inject per-instance configuration without baking paths into a shared config.txt.

Get Started

Installation Guides

Core Features

Advanced Usage

Deployment & Operations

Configure Fooocus: Model Paths and Defaults via config.txt

Editing and Resetting

Path Settings

Default Model Settings

Generation Defaults

UI Defaults

Download Settings

Full Example

Environment Variable Overrides

Build docs developers (and LLMs) love

Get Started

Installation Guides

Core Features

Advanced Usage

Deployment & Operations

Documentation Index

​Editing and Resetting

​Path Settings

​Default Model Settings

​Generation Defaults

​UI Defaults

​Download Settings

​Full Example

​Environment Variable Overrides

Build docs developers (and LLMs) love

Editing and Resetting

Path Settings

Default Model Settings

Generation Defaults

UI Defaults

Download Settings

Full Example

Environment Variable Overrides