ArtificialGuyBR

Home / Blog / Training LoRAs

studio-ghibli-lora-prompt-guide

studio-ghibli-lora-prompt-guide

0 sources cited Training LoRAs

On this page

  1. TL;DR
  2. What the Model Does
  3. Base Model and License
  4. Trigger Words and Prompt Format
  5. How to Use It
  6. Recommended Settings
  7. Common Errors and Fixes
  8. FAQ
  9. Sources

Studio Ghibli LoRA for Stable Diffusion 1.5: Prompt Guide and Settings

TL;DR

What the Model Does

StudioGhibli.Redmond 1.5V is a LoRA adapter trained to impart the hand-drawn, warm, and emotionally expressive aesthetic of Studio Ghibli films onto Stable Diffusion generations. It targets the runwayml/stable-diffusion-v1-5 base model and is explicitly designed to pair with the LiberteRedmond SD 1.5 checkpoint. When applied, it tends to soften edges, favor warm daylight lighting, desaturate colors slightly toward a pastel palette, and produce gentle facial expressions with softer skin tones — hallmarks of Ghibli character art.

Base Model and License

Base model: runwayml/stable-diffusion-v1-5. Weights file: StudioGhibliRedmond-15V-LiberteRedmond-StdGBRedmAF-StudioGhibli.safetensors (Safetensors format). The model is distributed under a custom license designated as bespoke-lora-trained-license (tagged as other). Review the license link on the Hugging Face model card before using it in commercial products.

Trigger Words and Prompt Format

Two trigger words activate the style: the token StdGBRedmAF and the phrase Studio Ghibli. The author's instance prompt is Studio Ghibli. Recommended placement is at the end of the subject description:

A cute cat on top a table with food, StdGBRedmAF, Studio Ghibli
A fluffy pink monster, StdGBRedmAF, Studio Ghibli
Portrait of a happy child, StdGBRedmAF, Studio Ghibli
A black boy, StdGBRedmAF, Studio Ghibli

How to Use It

Diffusers (Python)

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained(
    'runwayml/stable-diffusion-v1-5',
    torch_dtype=torch.float16,
).to('cuda')

pipeline.load_lora_weights(
    'artificialguybr/studioghibli-redmond-1-5v-studio-ghibli-lora-for-liberteredmond-sd-1-5',
    weight_name='StudioGhibliRedmond-15V-LiberteRedmond-StdGBRedmAF-StudioGhibli.safetensors',
)

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

ComfyUI

Load a LoraLoader node, set the model to runwayml/stable-diffusion-v1-5 (or LiberteRedmond SD 1.5), and provide the LoRA file. In a CLIPTextEncode pass or conditioning prompt, include StdGBRedmAF, Studio Ghibli.

AUTOMATIC1111 (WebUI)

Place the .safetensors file in models/Lora/, select it from the LoRA dropdown, and append StdGBRedmAF, Studio Ghibli to your prompt. Start with a strength of 0.8.

Parameter | Value

|---|:---:

Base model | runwayml/stable-diffusion-v1-5 / LiberteRedmond SD 1.5 Steps | 20–30 CFG scale | 6–8 (7 default) LoRA weight | 0.6–1.0 Sampler | Euler a or DPM++ 2M Karras Triggers | StdGBRedmAF, Studio Ghibli Higher LoRA weights (above 1.0) may over-apply the style and flatten fine detail. If the output looks too soft or muddy, reduce the weight to 0.6 or add detail descriptors such as soft lighting, pastel colors, film grain.

Common Errors and Fixes

Error | Cause | Fix | Source

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

Style not applied | Missing StdGBRedmAF token | Add both StdGBRedmAF and Studio Ghibli to the prompt | Model card Over-flattened image | LoRA weight too high | Reduce to 0.6–0.8 | Author recommendations Wrong aesthetic, noisy output | Using a non-SD 1.5 base | Use runwayml/stable-diffusion-v1-5 or LiberteRedmond SD 1.5 | Model card File not detected | Wrong weight name | Use exact file StudioGhibliRedmond-15V-LiberteRedmond-StdGBRedmAF-StudioGhibli.safetensors | Diffusers snippet

FAQ

Do I need to merge with LiberteRedmond SD 1.5? No, but the author designed this LoRA to complement LiberteRedmond SD 1.5. It still works on vanilla SD 1.5.

What resolution should I use? The base SD 1.5 default (512×512) is recommended. For portraits, 512×768 works and avoids the well-known SD hands issue.

Is it safe for commercial use? The license is custom (other). Confirm the license link on the model card before commercial deployment.

Sources