ernie-image-base-studio-ghibli-style-lora

Studio Ghibli Style LoRA for ERNIE Image

<Gallery />

Model description

This is a LoRA adapter for baidu/ERNIE-Image tuned toward photo-to-anime Studio Ghibli-style snapshots, casual portraits, everyday interiors, muted earthy colors, warm beige light, and clean cel-shaded outlines.

The repository contains the LoRA adapter only. It does not include or redistribute the base model.

Trigger words

Use:

studio ghibli style

Recommended settings

• LoRA weight: 0.85-0.95
• Steps: 36-45
• Guidance: 3.8-4.2
• Sampler: FlowMatch if available, otherwise UniPC Trailing
• Resolution: 1024 x 1024

Prompting guidance

This LoRA tends to respond best when prompts follow the same rhythm as the training captions:

studio ghibli style, photo-to-anime [portrait/group snapshot/everyday scene] of [main subject and action], [pose or scene clue], [simple setting clue], clean cel-shaded outlines, muted earthy color palette

Keep the opening phrase consistent, describe the subject plainly, then end with good style anchors. Useful ending anchors include front-facing character, rounded face, everyday clothing, simple real-world background, warm natural light, clean cel-shaded outlines, and muted earthy color palette.

Training data

Trained on a curated 255-image subset of filberthamijoyo/AI_Generated_Ghibli. Captions were expanded into concise photo-to-anime descriptions for subject, pose, setting, color, and rendering style, then prefixed with the trigger phrase studio ghibli style.

File

studio_ghibli_style_lora.safetensors

SHA256:

ebfe3db0ea684af60cf6a76a64269acfedc66a546f75e29685fc2790c24bf7f3

License

This model is licensed under the Apache License 2.0.