Fooocus Presets Explained: Default, Anime, Realistic, SAI & LCM

Fooocus presets are saved configurations that select models, styles, and generation settings together. Choose one in Advanced → Settings → Preset, or launch with --preset anime, replacing anime with your chosen preset name.

The official repository includes eight preset files: default, anime, realistic, sai, lcm, lightning, playground_v2.5, and pony_v6. Their configurations are compared below. Sources: preset folder, interface.

What are presets in Fooocus?

A preset is a JSON file in Fooocus/presets/. It can select the base checkpoint, refiner, LoRAs, guidance scale, sampler, scheduler, performance mode, styles, prompts, and image dimensions. Source: configuration.

For example, switching from default to realistic changes the checkpoint and photography LoRA together. It also changes the guidance scale, styles, and starting aspect ratio. Sources: default, realistic.

That makes a preset useful when moving between projects. Save a configuration for portraits, another for illustrations, and compare each from a consistent starting point instead of rebuilding its settings by hand.

For the surrounding workflow, see our Fooocus practical guide and AI image generator comparison.

Presets at a glance

These are the official files' starting configurations. Model names omit the .safetensors extension in this table. Steps describe ordinary text-to-image generation with the selected performance mode and no manual step override.

Step counts come from the performance definitions. LCM and Lightning override their JSON guidance and sampling settings in the generation worker; reading their preset files alone misses those changes.

Presets provided with Fooocus

Default: a general starting point

Default uses JuggernautXL V8 without a refiner. Its active LoRA is sd_xl_offset_example-lora_1.0.safetensors at 0.1. Its styles are Fooocus V2, Fooocus Enhance, and Fooocus Sharp. Source: default preset.

Start here when you want to explore a subject without committing to the photography or anime configuration. Keep the preset fixed while testing your wording, then compare another preset using the same description.

Anime: anime and illustrated characters

Anime selects animaPencilXL_v500.safetensors, with no refiner or active preset LoRA. It loads Fooocus V2, Fooocus Semi Realistic, and Fooocus Masterpiece. Its initial dimensions are portrait-oriented. Source: anime preset.

Choose it for an anime-focused starting point. The checkpoint and selected styles both belong in your comparison notes; changing only the checkpoint does not reproduce the complete preset.

Realistic: photography and portraits

Realistic combines realisticStockPhoto_v20.safetensors with SDXL_FILM_PHOTOGRAPHY_STYLE_V1.safetensors at 0.25. It uses no refiner and selects Fooocus V2, Fooocus Photograph, and Fooocus Negative. Source: realistic preset.

It also supplies a negative prompt covering nonphotographic looks and unwanted markings. Review that field when changing direction: a photography configuration may contain exclusions you do not want for an illustration.

SAI: SDXL base plus refiner

SAI uses sd_xl_base_1.0_0.9vae.safetensors with sd_xl_refiner_1.0_0.9vae.safetensors. The refiner switch is 0.75, and the SDXL offset LoRA weight is 0.5. Source: SAI preset.

Its styles are Fooocus V2 and Fooocus Cinematic. Choose SAI to explore Fooocus's base-and-refiner configuration. Treat the supplied values as Fooocus settings, rather than universal recommendations for every SDXL interface.

LCM and Lightning: shorter generation runs

LCM selects Extreme Speed. At runtime, Fooocus adds sdxl_lcm_lora.safetensors at 1.0, disables the refiner, and uses CFG 1 with the lcm sampler and scheduler. Sources: worker, performance definitions.

Lightning adds sdxl_lightning_4step_lora.safetensors at 1.0, disables the refiner, and uses CFG 1, sampler euler, and scheduler sgm_uniform. Its standard generation uses four steps. Sources: worker, performance definitions.

Both preset files select Fooocus V2, Fooocus Enhance, and Fooocus Sharp. Their JSON files contain CFG 4 and ordinary sampling defaults, but the performance mode replaces those values during generation. Sources: LCM, Lightning.

Use these presets when exploring alternatives quickly. Evaluate the resulting image against your intended use before deciding whether its speed and appearance suit the final output.

Playground and Pony: model-specific configurations

Playground selects only Fooocus V2, no refiner, and no active LoRA. Its sampler is dpmpp_2m, scheduler is edm_playground_v2.5, and inpaint engine is None. Source: Playground preset.

Pony selects Fooocus Pony, no refiner, and no active LoRA. It supplies ponyDiffusionV6XL_vae.safetensors and sets the inpaint engine to None. Choose it when you want its Pony checkpoint. Source: Pony preset.

Default, anime, realistic, SAI, and Pony all specify sampler dpmpp_2m_sde_gpu and scheduler karras. Their different model, LoRA, style, and guidance choices still make them distinct configurations.

How to switch Fooocus presets

Enable Advanced, open Settings, and select Preset. Check the model and styles before generating. The initial entry represents startup settings; it is not another JSON preset. Sources: interface, configuration.

On Windows, the supplied run.bat, run_anime.bat, and run_realistic.bat launch the corresponding configurations. For a Python installation, run this from the Fooocus directory with its environment active:

python entry_with_update.py --preset anime

Replace anime with realistic, sai, lcm, or another preset filename without .json. Browser selection can be hidden with --disable-preset-selection. Source: README.

Missing models may trigger downloads. Fooocus can also reuse a listed older default model already on disk. Add --always-download-new-model when you want the preset's specified model downloaded instead of that fallback. Source: README.

Allow downloads to finish before comparing results. Record the model that actually loads, particularly when an existing installation contains checkpoints named in a preset's previous_default_models list.

How to write a custom preset JSON

Create a new file such as presets/my_portraits.json. Copying a bundled preset is a useful starting point because it keeps the checkpoint, download mappings, and related settings together.

For a minimal example using an already installed default checkpoint, save:

{
  "default_model": "juggernautXL_v8Rundiffusion.safetensors",
  "default_refiner": "None",
  "default_loras": [],
  "default_cfg_scale": 4.0,
  "default_sampler": "dpmpp_2m_sde_gpu",
  "default_scheduler": "karras",
  "default_performance": "Speed",
  "default_overwrite_step": -1,
  "default_styles": ["Fooocus V2", "Fooocus Photograph"],
  "default_aspect_ratio": "896*1152"
}

These keys and values follow the configuration implementation and bundled presets. This example deliberately uses no LoRA; its style combination is a starting configuration you can adapt.

Launch it with:

python entry_with_update.py --preset my_portraits

Use exact checkpoint filenames. For a downloadable custom checkpoint, copy the checkpoint_downloads structure from a bundled preset and supply the actual model URL. LoRA downloads use their own lora_downloads mapping.

The default_loras entries use [enabled, filename, weight]. An empty list selects no preset LoRAs, although accelerated modes can add their own. Sources: configuration, worker.

Keep default_overwrite_step at -1 to use performance-defined steps. Source: generation worker.

At startup, Fooocus reads the default preset, then local configuration, then the requested preset. Specify settings you intend to control instead of depending on inherited values. Source: configuration.

Use double quotes, valid JSON, and no trailing commas. Keep custom files separate from bundled presets, then restart Fooocus. The generated config_modification_tutorial.txt documents additional keys. Source: README.

FAQ

Which Fooocus preset should I choose first?

Try default for general exploration, realistic for photography, and anime for illustrated characters. Choose SAI, Playground, or Pony when you specifically want their supplied models. Compare LCM and Lightning when generation speed matters.

Is a preset the same as a style?

No. A preset can load models and generation settings as well as styles. A selected style is one component of that configuration, as the separate default_styles entries show. Source: default preset.

Why does LCM show different settings in its JSON?

Extreme Speed applies runtime overrides, including the LCM LoRA, CFG 1, and LCM sampling. The preset's stored sampler and guidance values therefore do not describe the final generation configuration. Source: worker.

Can I keep several custom presets?

Yes. Fooocus discovers .json files in its presets directory. Give each configuration a distinct filename and launch it by that name, or select it after restarting the interface. Source: configuration.

Sources

• Fooocus README: launchers, switching, downloads, customization
• Official preset directory
• Default preset
• Anime preset
• Realistic preset
• SAI preset
• LCM preset
• Lightning preset
• Playground preset
• Pony preset
• Configuration loading and preset keys
• Preset selection interface
• Performance modes, LoRAs, and step counts
• Generation worker and runtime overrides

Last reviewed and updated: September 2026.