dataset_type string | contains_images bool | contains_captions bool | contains_identifiable_people bool | contains_model_weights bool | workflow dict | expected_private_data_schema dict | compliance_checklist list |
|---|---|---|---|---|---|---|---|
metadata-only | false | false | false | false | {
"base_model_family": "SDXL",
"task": "personalized LoRA fine-tuning (lookalike-style)",
"execution_env": "Kaggle GPU",
"privacy_mode": "no biometric or personal media published"
} | {
"image_file": "<private_path>/<image_name>.jpg",
"caption_file": "<private_path>/<image_name>.txt",
"caption_style": "identity-preserving descriptive caption",
"split": [
"train",
"val"
]
} | [
"Provenance documented for each source image",
"Rights to train and redistribute are explicitly verified",
"Consent/model release exists for identifiable people",
"No minors or sensitive biometric contexts without explicit authorization",
"No publication of private prompts/captions tied to real identities"
... |
LoRA Lookalike Dataset Metadata (No Images)
This repository is a metadata-only reproducibility companion for personal SDXL LoRA workflow experiments.
It intentionally includes no raw images, no captions, and no trained weights.
Included
- Data governance and provenance policy for safe publication
- Expected dataset structure and caption format specification
- Checklist for consent, likeness rights, and licensing verification
Excluded
- Personal photos
- Third-party headshots (including royalty-free stock photos)
- Captions tied to identifiable individuals
- Generated portraits/lookalikes
- LoRA checkpoints and model weights
Why metadata-only
Even when images are royalty-free, likeness/publicity rights and model-training permissions can be ambiguous. To minimize legal/privacy risk, this repo publishes process metadata only.
Intended use
- Demonstrate a privacy-first publication pattern
- Share reproducible workflow standards without exposing biometric data
- Support academic or portfolio discussion of safe AI data practices
Safety and legal notes
Before publishing any face dataset for lookalike generation, verify all of the following:
- Clear redistribution rights
- Explicit permission for AI training and derivative generation
- Consent/model release for identifiable people
- Compliance with local publicity and biometric/privacy laws
If any point is unclear, do not publish the image data.
Suggested citation (workflow)
If you reference this repository, cite it as workflow documentation by Mihai Chindris.
- Downloads last month
- 21