ArtificialGuyBR

Home / Blog / Training LoRAs

Studio Ghibli LoRA for Stable Diffusion 2.1: Freedom Redmond Style Transfer

Apply Studio Ghibli's iconic animation style to SD 2.1 with this LoRA — trigger words, usage guide, and example generations.

By ArtificialGuyBR Published Updated 1 sources cited Training LoRAs

A Low-Rank Adaptation (LoRA) that brings Studio Ghibli's signature animation aesthetic to Stable Diffusion 2.1 base model. Two trigger words activate the style: "Studio Ghibli" for the general look and "StdGBRedmAF" as the trained instance token.

What This Model Does

This LoRA modifies Stable Diffusion 2.1 (specifically the Freedom Redmond variant) to generate images with:

  • Soft, watercolor-like color palettes
  • Expressive character faces with large eyes
  • Whimsical fantasy creatures and environments
  • Hand-drawn animation texture
  • The distinctive "Ghibli glow" in lighting

Happy child portrait in Studio Ghibli style Portrait of a happy child — trigger: "Portrait of a happy child, StdGBRedmAF, Studio Ghibli"

How It Was Trained

  • Base model: stabilityai/stable-diffusion-2-1-base (Freedom Redmond variant)
  • Training type: LoRA (Low-Rank Adaptation)
  • Instance prompt: Studio Ghibli
  • Trigger token: StdGBRedmAF
  • Format: Safetensors
  • License: Bespoke LoRA trained license — commercial use permitted with rental restrictions, derivatives allowed, attribution required

Usage with Diffusers

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained(
    'stabilityai/stable-diffusion-2-1-base',
    torch_dtype=torch.float16
).to('cuda')

pipeline.load_lora_weights(
    'artificialguybr/studioghibli-redmond-2-1v-studio-ghibli-lora-for-freedom-redmond-sd-2-1',
    weight_name='StudioGhibliRedmond21V-FreedomRedmond-StudioGhibli-StdGBRedmAF.safetensors'
)

image = pipeline('A pirate boy, cute, StdGBRedmAF, Studio Ghibli').images[0]

For ComfyUI/A1111: place the .safetensors file in your LoRA folder and add <lora:StudioGhibliRedmond21V-FreedomRedmond-StudioGhibli-StdGBRedmAF:1.0> to your prompt (adjust weight 0.7–1.2 as needed).

Ghost horror scene in Studio Ghibli style Atmospheric horror scene — trigger: "Ghost, horror, terror, StdGBRedmAF, Studio Ghibli"

Example Prompts & Results

Prompt | Style Notes

|---|---

Portrait of a happy child, StdGBRedmAF, Studio Ghibli | Warm lighting, expressive face, soft backgrounds A cat wearing sunglasses, desert background, StdGBRedmAF, Studio Ghibli | Whimsical character, environmental storytelling A pink fluffy monster, StdGBRedmAF, Studio Ghibli | Fantasy creature design, pastel palette A pirate boy, cute, StdGBRedmAF, Studio Ghibli | Adventure theme, detailed costume, dynamic pose Pirate boy character in Studio Ghibli style Pirate boy character — trigger: "A pirate boy, cute, StdGBRedmAF, Studio Ghibli"

  • LoRA weight: 0.8–1.0 (start at 1.0, reduce if style overpowers composition)
  • Sampler: DPM++ 2M Karras or Euler a
  • Steps: 20–30
  • CFG scale: 7–9
  • Resolution: 512×512 or 768×768 (SD 2.1 native)
  • VAE: Use SD 2.1 VAE for best color fidelity

Common Errors and Fixes

Error | Cause | Fix | Source

|---|---|---|---

Style not appearing | Missing trigger tokens | Include both "Studio Ghibli" and "StdGBRedmAF" in prompt | Model card Over-saturated colors | LoRA weight too high | Reduce weight to 0.7–0.8 | Community testing Blurry outputs | Wrong base model | Ensure using stabilityai/stable-diffusion-2-1-base | Model card License violation | Commercial rental use | Check bespoke license terms before renting generated assets | License link

FAQ

Q: Does this work with SDXL or SD 1.5? A: No — trained specifically for Stable Diffusion 2.1 base model.

Q: Can I merge this LoRA into the model? A: Yes, diffusers supports fuse_lora() for permanent merging.

Q: What's the Freedom Redmond variant? A: A community fine-tune of SD 2.1 with adjusted color space and prompt adherence; this LoRA was trained on that variant.

Q: Are the sample images cherry-picked? A: The model card shows 7 widget examples; the three displayed here are representative of the consistent style transfer.

Sources

  • HuggingFace model card: https://huggingface.co/artificialguybr/studioghibli-redmond-2-1v-studio-ghibli-lora-for-freedom-redmond-sd-2-1
  • License terms: https://multimodal.art/civitai-licenses?allowNoCredit=True&allowCommercialUse=Rent&allowDerivatives=True&allowDifferentLicense=False
  • Diffusers LoRA loading docs: https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading_adapters
  • ArtificialGuyBR model portfolio: https://artificialguy.com/

Model by ArtificialGuyBR. Enriched for the ArtificialGuyBR portfolio.