Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: TypeError
Message: Couldn't cast array of type string to null
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2303, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1852, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2059, in cast_array_to_feature
_c(array.field(name) if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1854, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2109, in cast_array_to_feature
casted_array_values = _c(array.values, feature.feature)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1854, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2143, in cast_array_to_feature
return array_cast(
array,
...<2 lines>...
allow_decimal_to_str=allow_decimal_to_str,
)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1854, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2005, in array_cast
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
TypeError: Couldn't cast array of type string to null
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
annotation_provenance dict | audio_timeline string | channel string | has_independent_audio bool | has_video_audio_track bool | id string | images list | license string | media_sha256 unknown | messages list | privacy_review dict | redistribution_allowed bool | split string | target_ir string | videos list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | [AUDIO_TIMELINE] block.
Offline audio evidence (authoritative observations; preserve every source label):
[AUDIO_TIMELINE]
source_label: <Audio 1>
source_type: independent_audio
visual_reference: <Audio 1>
duration_seconds: 5.000
language: English
transcript_verbatim: "Great Savior, why you're the one who saved Etharo... | image_audio | true | false | train-00001-44c65e69cff6b334 | [
"media/images/image-0597cdf369350f896bc9.jpg"
] | cc0-1.0 | {
"media/images/image-0597cdf369350f896bc9.jpg": "e0ab77ed45b3cc2abdce85848952a2ae4811bacfd5858412520624426b225d12"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the massive treant in <Picture 1>, featuring a monstrous face formed by intertwining bark and thick roots, a wide-open mouth with jagged wooden teeth, green mossy accents, and large spreading branches.
<Picture 1> is the first frame of [Shot 1], showing <Subject 1> standing at the ed... | [] |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | [AUDIO_TIMELINE]
source_label: <Audio 1>
source_type: embedded_video_audio
visual_reference: <Video 1>
duration_seconds: 5.086
language: Chinese
transcript_verbatim: "警告会，你有犯罪记录，往上写，开始守罪。糟了，赶紧把零度盔戴上，全部交你了。"
aligned_transcript:
00:00.320-00:03.840 speaker=unknown transcript="警告会你有犯罪记录往上写开始守罪糟了赶"
00:03.920-00:05.040 spea... | video_only_single | true | true | train-00002-f631dbe673b72816 | [] | cc0-1.0 | {
"media/videos/video-f1114034a706fa429e61.mp4": "6296991e776364afb5d47962a04af28f176924bebe5181f56746871b80528b50"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the man with short black hair, wearing glasses and a black polo shirt in <Video 1>.
<Subject 2> is the woman with a black ponytail, wearing a black blazer over a white shirt, holding a small black device with a faintly glowing blue ring in <Video 1>.
<Video 1> is the continuation sou... | [
"media/videos/video-f1114034a706fa429e61.mp4"
] |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | image_only_low | false | false | train-00003-0c519774864ae265 | [
"media/images/image-c5833effd2d5dadade15.jpg"
] | cc0-1.0 | {
"media/images/image-c5833effd2d5dadade15.jpg": "a82a5dc1e98f59eb257f89d38aa4f64dca3433b0891b9a5b1049c6fb359d4ec3"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the pair of grey-purple Mary Jane shoes in <Picture 1>. They feature a matte grey-purple surface covered with a regular pattern of white polka dots, a rounded toe, and a delicate silver-toned metal buckle ornament attached to the front vamp.
summary:
[reference generation] The targe... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | text_only | false | false | train-00004-a0453df28009e8bd | [] | cc0-1.0 | {} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | integrated_multimodal_description: [Shot 1] Cinematic, medium shot, the camera pushes in slightly. In a brightly lit corporate conference room with an oval table, an adult male on-screen tech director with a slender build, short combed hair, and a smug smile, wearing a tailored navy suit (S1), stands by a projector scr... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | [AUDIO_TIMELINE] block.
Offline audio evidence (authoritative observations; preserve every source label):
[AUDIO_TIMELINE]
source_label: <Audio 1>
source_type: independent_audio
visual_reference: <Audio 1>
duration_seconds: 5.062
language: English
transcript_verbatim: "A B C D E F G H I G K L L N."
aligned_transcript:... | image_audio | true | false | train-00006-e3f6d45e7ec76432 | [
"media/images/image-633471cf378393a0e1a7.jpg",
"media/images/image-fd5c27c570eb73c30d5e.jpg",
"media/images/image-c5c1bbd2b3ee28cd86f5.jpg"
] | cc0-1.0 | {
"media/images/image-633471cf378393a0e1a7.jpg": "c5c06ecb37a1fbb5f3479eebd5c662d12237d0884fb9093ce0bc67101b52d3ab",
"media/images/image-c5c1bbd2b3ee28cd86f5.jpg": "727bfab634dac89c16f271cee7c375ea2da90011ef505a08b14fc5ef1aae5f8c",
"media/images/image-fd5c27c570eb73c30d5e.jpg": "86a41924deeb4bc593fcd7e036e729ac09... | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the man in the four-view character sheet <Picture 1> and the man on the right in <Picture 3>, an American man with short brown hair, wearing a loosely buttoned white shirt with vertical texture and matching loose white pants.
<Subject 2> is the woman in the three-view character sheet... | [] |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | [AUDIO_TIMELINE]
source_label: <Audio 1>
source_type: embedded_video_audio
visual_reference: <Video 1>
duration_seconds: 5.086
language: unknown
transcript_verbatim: ""
aligned_transcript:
N/A
soundscape_caption: "The audio clip begins with a gentle, solo piano performance in a modern, minimalist style. The piano, reco... | video_only_single | true | true | train-00007-684452a9fd4b19a7 | [] | cc0-1.0 | {
"media/videos/video-bfdb5a31248c7cd34674.mp4": "e057406e213b55f388c0ac127c6969f6de2d793c5e6d0b6f3c2bc79a29ca974f"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the man with dark curly hair and a beard, wearing a light blue shirt in <Video 1>.
<Subject 2> is the woman with long dark wavy hair seen on the smartphone screen in <Video 1>.
<Video 1> is the source video for the edit.
<Audio 1> is directly used as the target video's complete audio... | [
"media/videos/video-bfdb5a31248c7cd34674.mp4"
] |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | image_only_low | false | false | train-00008-f420c0553bfeddd9 | [
"media/images/image-6ce74a91338957220fee.jpg"
] | cc0-1.0 | {
"media/images/image-6ce74a91338957220fee.jpg": "73b42fa597a202b469ae569eba7cdce3290c0d109f0851f415277a7d7b1fa1e0"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Picture 1> is the first frame of [Shot 1], showing a massive, jagged lightning bolt striking a dark, ancient stone platform beneath heavily bruised, churning storm clouds, with a central figure in light robes standing between two chained pillars, surrounded by a scattered crowd.
<Subject 1> is the... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | text_only | false | false | train-00009-bb5cb1b36165f7b4 | [] | cc0-1.0 | {} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | integrated_multimodal_description: [Shot 1] Cinematic, cold and desaturated flashback sequence, medium shot, the camera pushes in on a pitching wooden boat deck at night, framed vertically with heavy rain and crashing waves dominating the scene. A young on-screen male child (7 years old, slight build, wearing a soaked ... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | image_only_low | false | false | train-00012-4bbe43dc3a3f95fe | [
"media/images/image-ae4d1f3dd168e36b4785.jpg",
"media/images/image-8819f6ec477d3e47b639.jpg"
] | cc0-1.0 | {
"media/images/image-8819f6ec477d3e47b639.jpg": "632a4f7da4d57ca22816542b601529cf06de0ee5fc448514eeb2ab7cc208b100",
"media/images/image-ae4d1f3dd168e36b4785.jpg": "3718bece7d6434a085f6c508d91a6ab8df2a1d8d1da708c0b92f201cf0730603"
} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the young woman in <Picture 1>, a three-view character sheet, defined by her dark brown long wavy hair, red lipstick, long gold dangling earrings, and a glossy cream-colored satin outfit consisting of a long-sleeve boat-neck blouse and matching high-waisted pleated wide-leg trousers ... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | text_only | false | false | train-00013-25e77d8f02df6e03 | [] | cc0-1.0 | {} | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | integrated_multimodal_description: [Shot 1] Cinematic, medium close-up, holding a static shot. Inside a slightly dusty, warmly lit wholesale store, Lao Wang (S1), a 50-year-old man with a heavy build, a sweaty face with deep wrinkles, and short, thinning black hair, wearing a loose, faded blue cotton polo shirt, sits b... | [] | |
{
"redistribution_review": "Release owner confirmed on 2026-08-11 that the source corpus, generated annotations, media, and audio are open data and authorized their broadest public release.",
"source": "MiniMax /v2/h3_context_ir"
} | image_only_high | false | false | train-00014-c150c10f37c8c85f | [
"media/images/image-86b77e5602f583f4854d.jpg",
"media/images/image-e9ded38e6b6818d1344d.jpg",
"media/images/image-3521028cf1a720cc26b2.jpg",
"media/images/image-dd9c9106bcf801b1d2bb.jpg"
] | cc0-1.0 | {
"media/images/image-3521028cf1a720cc26b2.jpg": "da21f0f2443ca3f36a2a50adc994f3e6333b77f31e9e9c64873307c4836b782b",
"media/images/image-86b77e5602f583f4854d.jpg": "d12d0a0e0c8512233c9911a1e6ffb1ad2a4e0cf092a0cbb9d447fa67ff82be46",
"media/images/image-dd9c9106bcf801b1d2bb.jpg": "a8fd2352eb44c6fe6120ff0ab0d73a89cf... | [
{
"content": "You convert a user's video-generation request and ordered media attachments into a MiniMax-H3 Context-IR prompt. Return only the finished Context-IR. Do not add Markdown, code fences, analysis, a preface, or commentary. Keep the complete output at or below 7000 Unicode characters.\n\nFollow the au... | {
"categories": [],
"decision": "publish",
"model": "gpt-5.6-sol",
"reasoning_effort": "xhigh"
} | true | train | subject_definitions:
<Subject 1> is the young man from the three-view character sheet in <Picture 3>, wearing a dark traditional outfit with an embroidered red-and-white geometric front panel, a multi-colored wrapped headdress with a tassel, and a red waist sash. He has a serious expression with slight scratches on his... | [] |
End of preview.
H3-IR
H3-IR contains privacy-reviewed prompt/Context-IR pairs for training H3 prompt enhancers. The public export is fail-closed: a row is included only when its text, annotation, and every referenced media asset pass both privacy and redistribution-rights gates.
Splits
| Split | Rows |
|---|---|
| train | 1110 |
| validation | 81 |
| total | 1191 |
Privacy Review
All source rows and unique visual assets were reviewed with gpt-5.6-sol at
reasoning_effort=xhigh, followed by deterministic redaction and a second
local sensitive-data scan. Text decisions were {"drop": 72, "publish": 1992, "sanitize": 3};
media decisions were {"drop": 436, "publish": 2189, "sanitize": 367}. Unsafe
rows and rows referencing unsafe assets are absent. Aggregate evidence is in
privacy_summary.json; raw prompts, provider responses, local source paths,
and model rationales are intentionally private.
Images are re-encoded without EXIF. Approved bounded redactions use expanded, irreversible Gaussian blur. Videos requiring visual redaction are dropped. Videos with audio require separate voice privacy and redistribution approval.
Audio
The IR model consumes [AUDIO_TIMELINE] text, not waveforms. The complete
ASR, forced-alignment, optional diarization, sound-captioning, and timeline
merge workflow is maintained at
https://github.com/IAmIronMan42/h3-ir-enhancer/blob/main/docs/audio-pipeline.md.
Loading
from datasets import load_dataset
dataset = load_dataset("StellarVoyager/H3-IR")
License
The released training records, annotations, and media are dedicated to the
public domain under CC0 1.0. See LICENSE for the complete legal text and
asset_rights.jsonl for the asset-level release receipts.
Limitations
Automated privacy review reduces risk but cannot prove that every subtle
identifier was detected. Dataset consumers must comply with the declared
license and the per-asset provenance in asset_rights.jsonl.
- Downloads last month
- 185