Hugging Face's logo

Datasets:
abide1622
/
lip-biting-V2

Tasks:
Text-to-Image
Size:
n<1K
Tags:
lora-training
facial-expression
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot extract the features (columns) for the split 'train' of the config 'legacy' of the dataset.
Error code:   FeaturesError
Exception:    UnicodeDecodeError
Message:      'utf-8' codec can't decode byte 0x89 in position 0: invalid start byte
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4645, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2866, in _head
                  return next(iter(self.iter(batch_size=n)))
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3073, in iter
                  for key, pa_table in ex_iterable.iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2581, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/text/text.py", line 98, in _generate_tables
                  batch = f.read(self.config.chunksize)
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 838, in read_with_retries
                  out = read(*args, **kwargs)
                File "<frozen codecs>", line 325, in decode
              UnicodeDecodeError: 'utf-8' codec can't decode byte 0x89 in position 0: invalid start byte

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Lip Biting V2

Trigger: lpb1te (same as V1; V2 revisions before 2026-11-09 used lipb1te).

94 image/caption pairs: 45 newer and 32 legacy images rated four stars, plus 17 strong images rated three stars because their bite is stronger.

data/main/, data/legacy/ and data/strong/ each contain PNG/TXT pairs and ImageFolder metadata. masks/legacy/ contains same-filename, full-resolution grayscale mouth loss-mask candidates: white includes the region, black excludes it, gray feathers the boundary. Masks are automatically generated with YuNet and require visual review before training. Do not include the mask directory in ordinary image discovery.

masks/main/ (added 2026-11-08) holds the same kind of mask for every data/main/ image: the YuNet-located blend ellipse the inpaint step actually edited, feathered. All 45 were reviewed on overlay sheets. With masks on both subsets, a trainer can confine learning to the mouth (for example ai-toolkit's inverted_mask_prior) so the LoRA does not learn the subjects' faces.

data/strong/ (added 2026-11-08) holds 17 images from the same V1 source as legacy that were dropped from it for a lip bite stronger than the rest of the set. They are a separate subset so a trainer can raise or lower their exposure to tune effect strength. They share legacy's correlated faces, so more repeats also mean more identity bias; masks/strong/ holds same-filename mouth mask candidates made like masks/legacy/ (review not confirmed).

Legacy images originate from Lip Biting V1. They capture the effect well but have correlated facial appearance. Legacy captions are vision-model-assisted general descriptions; generated batches use requested-attribute captions which may not match every visual detail. Review captions before training. No rating lines or local absolute paths are included in the public captions/manifest. Public PNG metadata is stripped while decoded pixels are preserved.

Training integration

The intended consumer is the expanded Krea2 LoRA workflow built on AI Toolkit. Its public repository still exposes v1 at preparation time; compatibility with the updated ingestion/schema has NOT been verified. This package is data, not a runnable training config. No masks, repeats or loss weights are applied just by downloading it. Configure each subset explicitly. A 3:1 main/legacy repeat ratio is an experiment, not a validated optimum. With this candidate it gives 32/(45*3+32) legacy exposure without strong. Do not use that ratio together with other loss scaling without accounting for both.

Compare newer-only, reduced-exposure legacy, and masked legacy using fixed evaluation prompts/seeds. Masks do not guarantee identity removal. Avoid horizontal flips if captions retain left/right descriptions. No independent validation split is supplied; keep variants of one source/identity together when creating one.

License is CC-BY-NC-4.0, the same declaration as V1. Attribution: abide1622. Read dataset.json for the subset and mask contract. It does not apply training settings.
Downloads last month
201
Total file size:
425 MB

Models trained or fine-tuned on abide1622/lip-biting-V2