Presets

Return components configured for common tasks.

Every preset is a function that returns an ordinary component with its instructions and operation set, and passes extra keyword arguments, such as backend, to the component. INSTRUCTIONS holds the instructions of the presets that are text generators, extractors or classifiers.

Examples:

>>> from skai import presets
>>> summarizer = presets.summarizer(max_length=50, backend='dummy')
>>> type(summarizer).__name__, summarizer.max_length
('TextToTextGenerator', 50)

animator(instructions=None, duration=2.0, **kwargs)

Return a generator that turns images into short videos.

Parameters:

Name	Type	Description	Default
instructions	str | None	A description of the motion.	None
duration	float	The length of the videos in seconds.	2.0
**kwargs	Any	Other parameters of ImageToVideoGenerator.	{}

Returns:

Type	Description
ImageToVideoGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def animator(instructions: str | None = None, duration: float = 2.0, **kwargs: Any) -> ImageToVideoGenerator:
    """Return a generator that turns images into short videos.

    Args:
        instructions:
            A description of the motion.

        duration:
            The length of the videos in seconds.

        **kwargs:
            Other parameters of `ImageToVideoGenerator`.

    Returns:
        The configured generator.
    """
    return ImageToVideoGenerator(instructions=instructions, duration=duration, **kwargs)

background_remover(**kwargs)

Return a generator that removes the background of images.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of ImageToImageGenerator.	{}

Returns:

Type	Description
ImageToImageGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def background_remover(**kwargs: Any) -> ImageToImageGenerator:
    """Return a generator that removes the background of images.

    Args:
        **kwargs:
            Other parameters of `ImageToImageGenerator`.

    Returns:
        The configured generator.
    """
    return ImageToImageGenerator(operation='remove_background', **kwargs)

describer(kind='image', **kwargs)

Return a text generator that describes images, audio, videos or multimodal samples.

Parameters:

Name	Type	Description	Default
kind	Kind	The input kind.	'image'
**kwargs	Any	Other parameters of the generator.	{}

Returns:

Type	Description
BaseSupervisedGenerator	The configured {Kind}ToTextGenerator.

Source code in src/skai/presets/_presets.py

def describer(kind: Kind = 'image', **kwargs: Any) -> BaseSupervisedGenerator:
    """Return a text generator that describes images, audio, videos or multimodal samples.

    Args:
        kind:
            The input kind.

        **kwargs:
            Other parameters of the generator.

    Returns:
        The configured `{Kind}ToTextGenerator`.
    """
    return _instructed(kind, INSTRUCTIONS['describer'].format(kind=kind), **kwargs)

diarizer(**kwargs)

Return a segmenter that splits audio into speaker turns.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of AudioSegmenter.	{}

Returns:

Type	Description
AudioSegmenter	The configured segmenter.

Source code in src/skai/presets/_presets.py

def diarizer(**kwargs: Any) -> AudioSegmenter:
    """Return a segmenter that splits audio into speaker turns.

    Args:
        **kwargs:
            Other parameters of `AudioSegmenter`.

    Returns:
        The configured segmenter.
    """
    return AudioSegmenter(by='speaker', **kwargs)

document_parser(**kwargs)

Return a text generator that transcribes page images as Markdown that keeps their structure.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of ImageToTextGenerator.	{}

Returns:

Type	Description
ImageToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def document_parser(**kwargs: Any) -> ImageToTextGenerator:
    """Return a text generator that transcribes page images as Markdown that keeps their structure.

    Args:
        **kwargs:
            Other parameters of `ImageToTextGenerator`.

    Returns:
        The configured generator.
    """
    return ImageToTextGenerator(operation='transcribe', format='markdown', **kwargs)

lip_syncer(**kwargs)

Return a generator that makes the video of every sample speak its audio.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of MultimodalToVideoGenerator.	{}

Returns:

Type	Description
MultimodalToVideoGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def lip_syncer(**kwargs: Any) -> MultimodalToVideoGenerator:
    """Return a generator that makes the `video` of every sample speak its `audio`.

    Args:
        **kwargs:
            Other parameters of `MultimodalToVideoGenerator`.

    Returns:
        The configured generator.
    """
    return MultimodalToVideoGenerator(operation='lip_sync', **kwargs)

moderator(labels=None, kind='multimodal', **kwargs)

Return a classifier that labels samples as safe or unsafe.

Parameters:

Name	Type	Description	Default
labels	list[str] | None	The labels, ['safe', 'unsafe'] when it is None.	None
kind	Kind	The input kind.	'multimodal'
**kwargs	Any	Other parameters of the classifier.	{}

Returns:

Type	Description
BaseClassifier	The configured {Kind}Classifier.

Source code in src/skai/presets/_presets.py

def moderator(labels: list[str] | None = None, kind: Kind = 'multimodal', **kwargs: Any) -> BaseClassifier:
    """Return a classifier that labels samples as safe or unsafe.

    Args:
        labels:
            The labels, `['safe', 'unsafe']` when it is `None`.

        kind:
            The input kind.

        **kwargs:
            Other parameters of the classifier.

    Returns:
        The configured `{Kind}Classifier`.
    """
    instructions = INSTRUCTIONS['moderator'].format(kind='sample' if kind == 'multimodal' else kind)
    return _CLASSIFIERS[kind](labels=labels or ['safe', 'unsafe'], instructions=instructions, **kwargs)

music_generator(duration=10.0, **kwargs)

Return a generator that creates music from descriptions.

Parameters:

Name	Type	Description	Default
duration	float	The length of the music in seconds.	10.0
**kwargs	Any	Other parameters of TextToAudioGenerator.	{}

Returns:

Type	Description
TextToAudioGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def music_generator(duration: float = 10.0, **kwargs: Any) -> TextToAudioGenerator:
    """Return a generator that creates music from descriptions.

    Args:
        duration:
            The length of the music in seconds.

        **kwargs:
            Other parameters of `TextToAudioGenerator`.

    Returns:
        The configured generator.
    """
    return TextToAudioGenerator(operation='music', duration=duration, **kwargs)

object_detector(labels, **kwargs)

Return a segmenter that finds the boxes of objects in images.

Parameters:

Name	Type	Description	Default
labels	list[str]	The kinds of objects to find.	required
**kwargs	Any	Other parameters of ImageSegmenter.	{}

Returns:

Type	Description
ImageSegmenter	The configured segmenter.

Source code in src/skai/presets/_presets.py

def object_detector(labels: list[str], **kwargs: Any) -> ImageSegmenter:
    """Return a segmenter that finds the boxes of objects in images.

    Args:
        labels:
            The kinds of objects to find.

        **kwargs:
            Other parameters of `ImageSegmenter`.

    Returns:
        The configured segmenter.
    """
    return ImageSegmenter(by='object', labels=labels, **kwargs)

ocr(**kwargs)

Return a text generator that transcribes the text of images as plain text.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of ImageToTextGenerator.	{}

Returns:

Type	Description
ImageToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def ocr(**kwargs: Any) -> ImageToTextGenerator:
    """Return a text generator that transcribes the text of images as plain text.

    Args:
        **kwargs:
            Other parameters of `ImageToTextGenerator`.

    Returns:
        The configured generator.
    """
    return ImageToTextGenerator(operation='transcribe', **kwargs)

prompt_writer(target='image', **kwargs)

Return a text generator that writes a generation prompt from every text.

Parameters:

Name	Type	Description	Default
target	Literal['image', 'video', 'music']	The kind of generation model the prompt is for.	'image'
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def prompt_writer(target: Literal['image', 'video', 'music'] = 'image', **kwargs: Any) -> TextToTextGenerator:
    """Return a text generator that writes a generation prompt from every text.

    Args:
        target:
            The kind of generation model the prompt is for.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    return TextToTextGenerator(instructions=INSTRUCTIONS['prompt_writer'].format(target=target), **kwargs)

question_answerer(question, kind='text', **kwargs)

Return a text generator that answers one question about every sample.

Parameters:

Name	Type	Description	Default
question	str	The question.	required
kind	Kind	The input kind.	'text'
**kwargs	Any	Other parameters of the generator.	{}

Returns:

Type	Description
BaseSupervisedGenerator	The configured {Kind}ToTextGenerator.

Source code in src/skai/presets/_presets.py

def question_answerer(question: str, kind: Kind = 'text', **kwargs: Any) -> BaseSupervisedGenerator:
    """Return a text generator that answers one question about every sample.

    Args:
        question:
            The question.

        kind:
            The input kind.

        **kwargs:
            Other parameters of the generator.

    Returns:
        The configured `{Kind}ToTextGenerator`.
    """
    return _instructed(kind, INSTRUCTIONS['question_answerer'].format(kind=kind, question=question), **kwargs)

quiz_writer(n_questions=5, **kwargs)

Return a text generator that writes a multiple-choice quiz from every text.

Parameters:

Name	Type	Description	Default
n_questions	int	The number of questions.	5
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def quiz_writer(n_questions: int = 5, **kwargs: Any) -> TextToTextGenerator:
    """Return a text generator that writes a multiple-choice quiz from every text.

    Args:
        n_questions:
            The number of questions.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    return TextToTextGenerator(instructions=INSTRUCTIONS['quiz_writer'].format(n_questions=n_questions), **kwargs)

rewriter(style='clear and concise', **kwargs)

Return a text generator that rewrites texts in a style.

Parameters:

Name	Type	Description	Default
style	str	The style of the rewritten texts.	'clear and concise'
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def rewriter(style: str = 'clear and concise', **kwargs: Any) -> TextToTextGenerator:
    """Return a text generator that rewrites texts in a style.

    Args:
        style:
            The style of the rewritten texts.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    return TextToTextGenerator(instructions=INSTRUCTIONS['rewriter'].format(style=style), **kwargs)

script_writer(format='video narration', length='one-minute', **kwargs)

Return a text generator that writes a spoken script from every text.

Parameters:

Name	Type	Description	Default
format	str	The kind of script, such as 'video narration' or 'podcast'.	'video narration'
length	str	The length of the script, in words a model understands, such as 'one-minute'.	'one-minute'
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def script_writer(
    format: str = 'video narration',  # noqa: A002 - the name of the kind of script
    length: str = 'one-minute',
    **kwargs: Any,
) -> TextToTextGenerator:
    """Return a text generator that writes a spoken script from every text.

    Args:
        format:
            The kind of script, such as `'video narration'` or `'podcast'`.

        length:
            The length of the script, in words a model understands, such as `'one-minute'`.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    instructions = INSTRUCTIONS['script_writer'].format(format=format, length=length)
    return TextToTextGenerator(instructions=instructions, **kwargs)

separator(stem='vocals', **kwargs)

Return a generator that separates the vocals or the accompaniment of audio.

Parameters:

Name	Type	Description	Default
stem	Literal['vocals', 'accompaniment']	The stem to keep.	'vocals'
**kwargs	Any	Other parameters of AudioToAudioGenerator.	{}

Returns:

Type	Description
AudioToAudioGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def separator(stem: Literal['vocals', 'accompaniment'] = 'vocals', **kwargs: Any) -> AudioToAudioGenerator:
    """Return a generator that separates the vocals or the accompaniment of audio.

    Args:
        stem:
            The stem to keep.

        **kwargs:
            Other parameters of `AudioToAudioGenerator`.

    Returns:
        The configured generator.
    """
    return AudioToAudioGenerator(operation='separate', stem=stem, **kwargs)

shot_detector(**kwargs)

Return a segmenter that splits videos into shots locally.

Parameters:

Name	Type	Description	Default
**kwargs	Any	Other parameters of VideoSegmenter.	{}

Returns:

Type	Description
VideoSegmenter	The configured segmenter.

Source code in src/skai/presets/_presets.py

def shot_detector(**kwargs: Any) -> VideoSegmenter:
    """Return a segmenter that splits videos into shots locally.

    Args:
        **kwargs:
            Other parameters of `VideoSegmenter`.

    Returns:
        The configured segmenter.
    """
    return VideoSegmenter(by='shot', **kwargs)

storyboarder(n_scenes='a few', **kwargs)

Return an extractor that splits every text into scenes with narration and visuals.

Parameters:

Name	Type	Description	Default
n_scenes	int | str	The number of scenes, or words such as 'a few'.	'a few'
**kwargs	Any	Other parameters of TextExtractor.	{}

Returns:

Type	Description
TextExtractor	The configured extractor, whose only field scenes is a list of records with narration and visual.

Source code in src/skai/presets/_presets.py

def storyboarder(n_scenes: int | str = 'a few', **kwargs: Any) -> TextExtractor:
    """Return an extractor that splits every text into scenes with narration and visuals.

    Args:
        n_scenes:
            The number of scenes, or words such as `'a few'`.

        **kwargs:
            Other parameters of `TextExtractor`.

    Returns:
        The configured extractor, whose only field `scenes` is a list of records with `narration` and `visual`.
    """
    fields = {'scenes': [{'narration': 'the words spoken in the scene', 'visual': 'what the viewer sees'}]}
    instructions = INSTRUCTIONS['storyboarder'].format(n_scenes=n_scenes)
    return TextExtractor(fields=fields, instructions=instructions, **kwargs)

summarizer(max_length='auto', **kwargs)

Return a text generator that summarizes texts.

Parameters:

Name	Type	Description	Default
max_length	int | Literal['auto'] | None	The maximum number of words of every summary, or 'auto' to learn it from reference summaries.	'auto'
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def summarizer(max_length: int | Literal['auto'] | None = 'auto', **kwargs: Any) -> TextToTextGenerator:
    """Return a text generator that summarizes texts.

    Args:
        max_length:
            The maximum number of words of every summary, or `'auto'` to learn it from reference summaries.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    return TextToTextGenerator(instructions=INSTRUCTIONS['summarizer'], max_length=max_length, **kwargs)

talking_head(avatar, **kwargs)

Return a generator that makes a portrait speak audio.

Parameters:

Name	Type	Description	Default
avatar	Any	The portrait.	required
**kwargs	Any	Other parameters of AudioToVideoGenerator.	{}

Returns:

Type	Description
AudioToVideoGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def talking_head(avatar: Any, **kwargs: Any) -> AudioToVideoGenerator:
    """Return a generator that makes a portrait speak audio.

    Args:
        avatar:
            The portrait.

        **kwargs:
            Other parameters of `AudioToVideoGenerator`.

    Returns:
        The configured generator.
    """
    return AudioToVideoGenerator(avatar=avatar, **kwargs)

translator(target_language, *, learn_terms=True, **kwargs)

Return a text generator that translates texts and learns a glossary from reference translations.

Parameters:

Name	Type	Description	Default
target_language	str	The language of the translations.	required
learn_terms	bool	Whether fit learns a glossary of consistently rendered terms.	True
**kwargs	Any	Other parameters of TextToTextGenerator.	{}

Returns:

Type	Description
TextToTextGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def translator(target_language: str, *, learn_terms: bool = True, **kwargs: Any) -> TextToTextGenerator:
    """Return a text generator that translates texts and learns a glossary from reference translations.

    Args:
        target_language:
            The language of the translations.

        learn_terms:
            Whether `fit` learns a glossary of consistently rendered terms.

        **kwargs:
            Other parameters of `TextToTextGenerator`.

    Returns:
        The configured generator.
    """
    instructions = INSTRUCTIONS['translator'].format(target_language=target_language)
    return TextToTextGenerator(instructions=instructions, learn_terms=learn_terms, **kwargs)

upscaler(scale=2, **kwargs)

Return a generator that upscales images.

Parameters:

Name	Type	Description	Default
scale	int	The factor of upscaling.	2
**kwargs	Any	Other parameters of ImageToImageGenerator.	{}

Returns:

Type	Description
ImageToImageGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def upscaler(scale: int = 2, **kwargs: Any) -> ImageToImageGenerator:
    """Return a generator that upscales images.

    Args:
        scale:
            The factor of upscaling.

        **kwargs:
            Other parameters of `ImageToImageGenerator`.

    Returns:
        The configured generator.
    """
    return ImageToImageGenerator(operation='upscale', scale=scale, **kwargs)

voice_converter(voice, **kwargs)

Return a generator that says the speech of audio in another voice.

Parameters:

Name	Type	Description	Default
voice	Any	One recording or a list of recordings of the target voice.	required
**kwargs	Any	Other parameters of AudioToAudioGenerator.	{}

Returns:

Type	Description
AudioToAudioGenerator	The configured generator.

Source code in src/skai/presets/_presets.py

def voice_converter(voice: Any, **kwargs: Any) -> AudioToAudioGenerator:
    """Return a generator that says the speech of audio in another voice.

    Args:
        voice:
            One recording or a list of recordings of the target voice.

        **kwargs:
            Other parameters of `AudioToAudioGenerator`.

    Returns:
        The configured generator.
    """
    return AudioToAudioGenerator(operation='convert_voice', voice=voice, **kwargs)