Bilibili Scraper | Videos, Comments & Creators

Collect Bilibili video, comment, creator, danmaku and live-room data for Chinese video research. Choose a collection mode and export structured rows with explicit source-coverage information.

Quick start · Workflows · Pricing · Input reference · Output reference · API and automation · FAQ · Technical reference

What you can do

• Multiple public-data modes — Search, popular/ranking, video details, comments, creator profiles/catalogs, relations, danmaku and live rooms.
• Thread and reaction context — Request comment replies and optional per-video danmaku reaction profiles.
• Migration and monitoring — Accept mode or operation aliases and use persisted delta state for recurring collection.
• Visible completeness — Warnings, missing fields and source requirements distinguish partial retrieval from complete output.

Who uses it

• Creator and agency research: examine public creator profiles, catalogs and engagement.
• Brand and content teams: monitor topic searches, popular videos and public comments.
• Analysts: study timed danmaku reactions and export data for external analysis.

Quick start

1. Open Input, enter your target and a small result limit.
2. Click Start and inspect the dataset and run log.
3. Export JSON/CSV, or save the input as a Task for recurring runs.

{
  "mode": "popular",
  "maxResults": 5
}

Form prefills, schema defaults and empty API input can differ. Use explicit targets and limits; see the technical reference for source and default behavior.

Import an existing configuration

Use zhorexInput or sianInput to paste the corresponding competitor's complete input JSON. Put additional options inside the selected object. Native form defaults do not override it. For SIAN operations, omitted result budgets are derived from maxPages; explicit native caps still take precedence.

{"sianInput": {"operation": "searchVideo", "keyword": "人工智能", "maxPages": 1, "maxResults": 10}}

The import fields preserve documented input names, types and defaults. Output metadata and source availability can differ. Full subtitle text is not provided; available caption metadata is a separate source-dependent result.

Common workflows

Search recent videos

{
  "mode": "search",
  "searchQuery": "人工智能教程",
  "sortOrder": "pubdate",
  "maxResults": 10
}

Collect comments and replies

{
  "mode": "video_comments",
  "videoUrls": [
    "BV1rpWjevEip"
  ],
  "maxComments": 20,
  "includeReplies": true
}

Inspect one video

{
  "mode": "video_detail",
  "videoUrls": [
    "BV1rpWjevEip"
  ],
  "maxResults": 1
}

Output at a glance

Illustrative field subset below; these are example values, not a captured customer review or a claim about source coverage. Native and compatibility output profiles can differ.

{
  "type": "video",
  "bvid": "BV1rpWjevEip",
  "title": "Illustrative video title",
  "url": "https://www.bilibili.com/video/BV1rpWjevEip",
  "coverage": "partial",
  "warnings": [
    "Illustrative warning: one optional source field was unavailable."
  ]
}

Full subtitle retrieval is not guaranteed and is outside the current parity target. Source access and mode availability vary; a successful run is not proof that every requested field was retrieved. Optional enrichment can add output rows and costs.

Pricing

Pay per delivered event, with automatic discounts based on your Apify subscription plan. These are result fees in USD, not separate subscriptions sold by this Actor. All tiers have the same Actor features. Apify compute, storage, transfers and any proxy usage are charged additionally.

Billable event (1,000 events)	Free	Starter	Scale	Business
Video Discovery Result	$3	$2.5	$2	$1.5
Video Detail Result	$5	$4	$3	$2.5
Danmaku Profile	$10	$8	$6	$5
Profile Result	$5	$4	$3	$2.5
Relation Result	$5	$4	$3	$2.5
Live Result	$5	$4	$3	$2.5

Platinum and Diamond currently use the Business rate. The Pricing tab is the source of truth for the active schedule and your applicable plan.

• Video Discovery Result: Charged for each delivered video discovery result. Failed result rows carry no custom event fee. Platform usage is additional.
• Video Detail Result: Charged for each delivered video detail result. Failed result rows carry no custom event fee. Platform usage is additional.
• Danmaku Profile: Charged for each delivered danmaku profile. Failed result rows carry no custom event fee. Platform usage is additional.
• Profile Result: Charged for each delivered profile result. Failed result rows carry no custom event fee. Platform usage is additional.
• Relation Result: Charged for each delivered relation result. Failed result rows carry no custom event fee. Platform usage is additional.
• Live Result: Charged for each delivered live result. Failed result rows carry no custom event fee. Platform usage is additional.

Example on the Free plan: 10 / 1,000 / 10,000 Video Discovery Result events cost $0.03 / $3 / $30 in result fees, plus platform usage. Mixed Bilibili event types are summed separately.

There is no additional developer setup or monthly rental fee. A run with no billable results can still consume platform resources. Start small, cap results and review run costs before scheduling a larger workload. Requested counts are ceilings, not guaranteed deliveries.

Input reference

Fields below are grouped like the Input form. Optional fields need not be supplied. Compatibility input envelopes and transformations are explained in the technical reference.

Targets and collection

Field	Type	Default	Meaning
mode	string	Not set	Canonical mode. Leave empty when using the SIAN-compatible operation field.
operation	string	Not set	Optional compatibility alias: searchVideo, videoDetail, videoComments, userDetail, userRelationStat, userVideos, videoDanmu, videoCaption.
searchQuery	string	Not set	Keyword used in search mode.
keyword	string	Not set	SIAN-compatible alias for searchQuery.
autoLocalize	boolean	true	Also search a small set of common Chinese brand-name aliases when available.
searchAliases	array	[]	Extra terms to search and merge with the primary keyword.
pubtimeBegin	string	Not set	YYYY-MM-DD
pubtimeEnd	string	Not set	YYYY-MM-DD
videoUrls	array	[]	One or more Bilibili video URLs, BVIDs, or numeric AIDs.
bvid	string	Not set	Single BVID compatibility alias for video modes.
aid	string	Not set	Single numeric AID compatibility alias. Caption/transcript can also use aid together with cid.
cid	string	Not set	Single video CID, used directly by danmaku/caption/transcript modes when applicable.
userIds	array	[]	One or more Bilibili creator MIDs.
userId	string	Not set	Single creator MID compatibility alias.
liveRoomIds	array	[]	One or more Bilibili live room IDs or live-room URLs.
roomId	string	Not set	Single live room ID compatibility alias.
category	string	"all"	Category filter for popular and ranking modes.
includeComments	boolean	false	For discovery modes, append public comment rows for discovered videos while respecting the global maxResults row cap. The Actor reserves enrichment capacity before discovery so this flag is not a silent no-op when discovery could otherwise fill the cap.
includeReplies	boolean	false	When scraping comments, expand available child reply chains.
includeDanmaku	boolean	false	For discovery modes, append one aggregate danmaku profile per discovered video.
danmakuSampleSize	integer	25	Maximum number of representative danmaku texts kept in a danmaku profile.
includeCreator	boolean	false	For discovery modes, append public creator profile rows for creators in the video results.
includeTranscript	boolean	false	For discovery modes, append native transcript rows when Bilibili exposes public subtitle tracks.
enrichDetails	boolean	false	More complete engagement fields at the cost of additional requests.
requestDelayMillis	integer	120	Minimum spacing in milliseconds between HTTP request starts across the client.
sentimentAnalysis	boolean	false	Local SnowNLP polarity and -1 to +1 score for comments and video titles/descriptions. Use at least 512 MB memory. No external API fee. This heuristic is not a validated accuracy guarantee; non-Chinese text is tagged no_chinese_text.

Filters and ordering

Field	Type	Default	Meaning
sortOrder	string	"totalrank"	Sort video search results by relevance, views, publish time, danmaku, favorites, or score.
durationFilter	string	"any"	Limit search results by Bilibili duration bucket.
sortComments	string	"hot"	Preferred public comment ordering.

Result limits and cost

Field	Type	Default	Meaning
maxResults	integer	100	Primary result limit. In discovery modes, optional enrichment adds rows without reducing requested videos. maxOutputRows and the Apify charge budget remain hard caps. Set maxResultsScope=global for legacy total-row semantics.
maxPages	integer	Not set	SIAN-compatible page cap. maxResults remains the hard row cap.
maxComments	integer	20	Maximum number of comment/reply rows requested per video.
maxRetries	integer	3	Maximum retries for transient HTTP blocks, rate limits, network errors, and Bilibili risk-control responses.
maxResultsScope	string	"primary"	primary preserves requested discovery video count; global preserves legacy combined-row limit.
maxOutputRows	integer	10000	Additional safety cap including optional rows. Reaching it may truncate enrichment; charge budget can stop earlier.

Monitoring and state

Field	Type	Default	Meaning
deltaMode	boolean	false	State is isolated by deltaStateKey and persisted across runs in the named bilibili-delta-state store. Use a separate key per workflow and avoid overlapping runs with the same key.
deltaStateKey	string	"default"	Namespace used to isolate persisted delta-mode history between independent monitors.

Compatibility and transformations

Field	Type	Default	Meaning
outputSchema	string	"extended"	extended adds coverage/warnings; zhorex-strict removes extension fields for easier migration.

Connection and advanced settings

Field	Type	Default	Meaning
proxyConfiguration	object	{"useApifyProxy": false}	Direct HTTP is the default. Enable Apify Proxy if your network is rate-limited.
requestConcurrency	integer	5	Maximum number of independent target requests processed concurrently.

Additional compatibility options

Field	Type	Default	Meaning
zhorexInput	object	Not set	Paste the complete competitor input JSON here to preserve its field types and defaults. Put all options inside this object. This is optional; normal native inputs remain available.
sianInput	object	Not set	Paste the complete competitor input JSON here to preserve its field types and defaults. Put all options inside this object. This is optional; normal native inputs remain available.

Output reference

The table describes native fields; fields can be absent or null when the source does not provide them. Error rows and selected compatibility profiles have different shapes. Nested objects remain structured in JSON; CSV exports may serialize them.

Field	Type	Meaning
type	string	Result kind such as video, comment, danmaku, user, relation, caption, transcript, live, or error.
bvid	string	Bilibili video identifier when the row belongs to a video.
aid	integer	Numeric Bilibili archive ID when available.
title	string	Video or live-room title when applicable.
url	string	Canonical public URL for the result when available.
authorName	string	Public creator name for video/comment rows when available.
authorMid	integer	Numeric Bilibili creator ID for video/comment rows when available.
viewCount	['integer', 'null']	Public video view count when supplied by the source endpoint.
text	string	Comment, danmaku, or transcript text depending on the row type.
available	boolean	For optional source data such as captions/transcripts, whether the public source exposed the requested data.
notAvailableReason	string	Why optional public data was unavailable, for example login_required_by_source or no_public_track.
sourceRequiresLogin	boolean	Whether Bilibili explicitly indicated that the requested optional source data requires a logged-in source session.
coverage	string	Completeness assessment for this row: complete, partial, best_effort, rate_limited, or not_available.
warnings	array	Machine-readable-adjacent human warnings explaining fallbacks, truncation, or source limitations.
sourceMode	string	Actor mode that produced this row.
retrievalMethod	string	Public retrieval path used to obtain this row, including fallback paths when applicable.
errorCode	string	Structured Actor/Bilibili error classification for error rows.
scrapedAt	string	UTC ISO-8601 time at which the row was produced.
missingFields	array	Fields unavailable in exact-search detail fallback are null, never fabricated zero counts.
catalogCountGap	['integer', 'null']	Reported total minus unique observed source IDs, clamped at zero; not verified missing videos.

API and automation

The API tab provides snippets for this Actor. For scheduled or larger collections, start a run asynchronously and poll its status instead of holding a synchronous request open. Keep your API token in an environment variable.

import os, time, requests

base = "https://api.apify.com/v2"
headers = {"Authorization": "Bearer " + os.environ["APIFY_TOKEN"]}
actor_input = {'mode': 'popular', 'maxResults': 5}
response = requests.post(
    base + "/acts/HcxOZArmyg6ZyjOBJ/runs",
    headers=headers, json=actor_input, timeout=30,
)
response.raise_for_status()
run = response.json()["data"]
for _ in range(720):
    if run["status"] not in ("READY", "RUNNING", "TIMING-OUT", "ABORTING"):
        break
    time.sleep(5)
    response = requests.get(base + "/actor-runs/" + run["id"], headers=headers, timeout=30)
    response.raise_for_status()
    run = response.json()["data"]
else:
    raise TimeoutError("Polling ended; inspect the existing run before starting another.")
if run["status"] != "SUCCEEDED":
    raise RuntimeError(run.get("statusMessage", run["status"]))
response = requests.get(
    base + "/datasets/" + run["defaultDatasetId"] + "/items",
    headers=headers, params={"format": "json", "limit": 1000, "offset": 0}, timeout=30,
)
response.raise_for_status()
rows = response.json()
print(rows)  # Paginate with offset for larger datasets; inspect per-row coverage.

Scheduled monitoring and integrations

1. Save a working input as an Apify Task.
2. Attach a Schedule in your timezone; keep runs sharing an incremental state key sequential.
3. Add a successful-run webhook or connect your downstream workflow to the dataset export/API.
4. Inspect missing/partial data before sending alerts or treating absence as removal.

Use the platform Integrations/API options to connect external workflows such as Make, n8n or a data warehouse. An external sentiment or summarization step is separate from this Actor and may have its own charges. The platform also supports Actor tools through Apify MCP; availability and client setup depend on your integration.

FAQ

Do users need to use the developer’s Bilibili account?

No shared personal account is promised. Public collection paths are used; source-restricted fields are reported as unavailable or partial.

Does this guarantee complete subtitles?

No. Caption and transcript modes depend on source availability; full subtitles are not a guaranteed product feature.

What is a danmaku profile?

It summarizes a video’s retrieved timed reactions into a profile. It is billed per delivered profile, separately from a video row.

Can I export without writing code?

Yes. Run from the Input form, open the dataset and use the export controls for JSON or CSV. Select your preferred output profile before running when migrating an integration.

Does a successful run guarantee every field?

No. Inspect coverage, warnings and source-specific status fields. Optional enrichment may be missing even when usable rows were delivered.

Troubleshooting

Symptom	What to check
Empty dataset	Validate the target, source availability and filters; try the small quick-start input.
Fewer rows than requested	Check source limits, pagination and filtering. Increasing a cap cannot create unavailable source records.
Partial fields	Inspect coverage/warnings and the relevant source or enrichment status.
Access or throttling errors	Check the run log and documented proxy configuration; retry at a lower request volume.
Migration mismatch	Check the explicit input/output profile and technical field mapping below.
Unexpected bill	Separate result events from platform resources; inspect enrichment, result caps and the Pricing tab.

Technical reference

Supported modes

Mode	Input	Result
search	searchQuery	Video rows
video_detail	videoUrls / bvid	Full video metadata
video_comments	videoUrls / bvid	Comment rows, optional replies
popular	optional category	Popular video rows
ranking	optional category	Ranking video rows
user_profile	userIds / userId	Creator profile rows
user_relation_stats	userIds / userId	Follower/following rows
user_videos	userIds / userId	Creator upload rows; coverage is explicit
video_danmaku	video target or cid	Raw danmaku rows
danmaku_profile	video target or cid	One aggregate reaction profile per video
video_caption	video target	Native subtitle metadata
video_transcript	video target	Native subtitle text when available
live_info	liveRoomIds / roomId	Public live room state

Quick start

Search:

{
  "mode": "search",
  "searchQuery": "人工智能教程",
  "sortOrder": "pubdate",
  "maxResults": 50
}

Video details:

{
  "mode": "video_detail",
  "videoUrls": ["BV1rpWjevEip"]
}

Comments with replies:

{
  "mode": "video_comments",
  "videoUrls": ["BV1rpWjevEip"],
  "maxComments": 100,
  "includeReplies": true
}

SIAN-style input is accepted too:

{
  "operation": "searchVideo",
  "keyword": "python",
  "maxPages": 3
}

Migration compatibility

The canonical API intentionally keeps familiar fields such as mode, searchQuery, videoUrls, userIds, maxResults, sortOrder, includeComments, includeReplies and includeDanmaku.

By default, maxResults limits primary discovery videos. Optional comments, danmaku profiles, creators and transcripts add rows without reducing the requested video count. maxOutputRows and the Apify charge budget cap total delivery. Set maxResultsScope="global" to restore the legacy shared row budget, which reserves enrichment slots by reducing the initial discovery batch.

For multi-target modes that can emit many rows per target (user_videos, comments, raw danmaku, captions, transcripts), the remaining global row budget is distributed across the remaining targets and reallocated when an earlier target returns fewer rows. A single prolific creator or video therefore cannot silently starve the rest of a batch.

Set:

{"outputSchema": "zhorex-strict"}

to suppress extension fields on documented video/comment rows. This helps pipelines that reject unknown JSON properties.

Strict video output preserves authorFace (the creator avatar URL) when present in the source row. Comment avatars remain in authorAvatar; missing video avatars are not fabricated.

The default extended schema adds:

{
  "schemaVersion": "1.0",
  "sourceMode": "video_comments",
  "coverage": "complete",
  "warnings": [],
  "retrievalMethod": "public_http"
}

coverage can be complete, partial, best_effort, rate_limited, or not_available.

Rate limits and proxies

The Actor uses public Bilibili HTTP endpoints. Direct access is the default. Bilibili can return HTTP 412, API -352, or -799 under risk control, especially during deeper pagination.

The client includes request pacing, exponential retry and optional Apify Proxy sessions. Enable a proxy in the standard proxyConfiguration input if your environment is blocked.

Video-bound modes resolve CID/AID separately from full video detail: comments use /x/player/pagelist and /x/player/v2; danmaku/caption modes use /x/player/pagelist. Source restrictions can still prevent retrieval.

video_detail itself still prefers the full /x/web-interface/view source. If both full-detail APIs are blocked, the Actor performs an exact BVID/AID search, verifies the returned ID, enriches it with /x/player/pagelist, and returns that row as coverage: best_effort with retrievalMethod: public_exact_search_fallback. This is a deliberate downgrade rather than a false error or a false claim of completeness.

user_videos attempts the WBI creator catalog, then public media-list, dynamic-feed and exact-MID creator-name search fallbacks. Partial catalog results can trigger supplements. Fallback rows remain coverage: best_effort; a complete creator archive is not guaranteed.

Comments

Comment coverage depends strongly on Bilibili's response for the caller's egress IP. The Actor first tries the current WBI cursor API, then falls back to the legacy public comment endpoint when necessary. It deduplicates comment IDs and exposes sourceReportedCommentCount, returnedCommentCount, coverageRatio, and retrievalMethod when the source provides a total. If the source exposes only top/pinned comments, those rows remain usable but are explicitly marked coverage: partial.

A requested comment cap below the source-reported total remains partial coverage even when every requested row is returned.

Danmaku

video_danmaku returns the public XML danmaku payload as individual rows. danmaku_profile reduces the same public payload into count, unique senders, per-minute rate, peak minute and an evenly spread text sample.

Native transcripts

video_caption returns subtitle metadata/URLs when Bilibili exposes them publicly. video_transcript fetches those native subtitle JSON files and produces joined text plus timestamped segments. When the anonymous player API exposes no subtitle track, the Actor returns a status row with available: false and coverage: not_available instead of pretending the scraper failed. If Bilibili explicitly reports need_login_subtitle, the row also carries notAvailableReason: login_required_by_source and sourceRequiresLogin: true. These unavailable status rows are not charged through a custom pay-per-event result event. The GA Actor intentionally does not request a user's Bilibili session cookie and does not invent an ASR transcript when native subtitles are unavailable.

Run success vs. data completeness

A successful Actor run only means the requested workflow executed normally. It does not imply that Bilibili exposed every row that exists on the website. Extended rows therefore carry an explicit completeness signal:

Coverage	Meaning
complete	The source endpoint returned the requested object/set without a known truncation signal.
partial	The source reported or strongly indicated more data than was returned.
best_effort	A public fallback supplied useful rows, but it is not a guaranteed complete archive.
rate_limited	Bilibili risk control prevented full retrieval.
not_available	The optional public data does not exist or was not exposed, for example a video with no public native subtitle track.

The run-level OUTPUT record separately reports partialRows, unavailableRows, errorRows, HTTP request/retry counts, and the number of rows pushed to the dataset.

Run summary status is also explicit:

• succeeded: no known coverage limitation or error was observed.
• succeeded_with_limited_coverage: the run completed but one or more rows are partial, best-effort, or unavailable.
• succeeded_with_errors: some targets failed while other targets still produced data.
• failed: every produced row is an error or a top-level mode failure occurred.
• budget_limit_reached: Apify stopped additional charged rows at the configured event budget.

Delta mode

For search, popular and ranking, set:

{
  "deltaMode": true,
  "deltaStateKey": "my-monitor"
}

Delivered video IDs are retained in the named key-value store bilibili-delta-state; later runs with the same scope return unseen videos. See incremental state below.

Current public-data limitations

• Creator upload catalogs are one of Bilibili's most aggressively risk-controlled public surfaces. user_videos therefore treats fallback results as best_effort and verifies the creator MID before returning a discovered upload.
• Full-detail sources can be blocked. Exact-ID search fallbacks remain best_effort; unavailable deep fields remain null with missing-field diagnostics and no full-detail custom charge.
• Anonymous/datacenter comment access can expose only a tiny set of top or pinned comments even when the video reports a much larger reply count. The Actor reports the observed coverage instead of manufacturing completeness.
• Native captions/transcripts are returned only when Bilibili exposes a subtitle track to the anonymous public player API. If Bilibili marks subtitle metadata as login-required, the output says so explicitly rather than asking for or silently depending on a Bilibili account cookie.
• Optional proxies do not guarantee access or improve every source; direct access remains the default.

Data scope

This Actor is built for publicly accessible Bilibili data. Users are responsible for using exported data in accordance with applicable law and the source platform's terms.

Incremental state

Delta mode persists delivered video IDs in the named bilibili-delta-state key-value store across runs. Use a distinct deltaStateKey for each query/filter workflow and do not overlap runs sharing the same key. State is committed after successful delivery, so budget-excluded rows remain eligible on the next run. A crash between delivery and state commit may repeat rows; consumers should deduplicate by BVID. The latest 20,000 IDs are retained per key.

Missing-field semantics

Exact-search detail fallback now returns null for unavailable engagement counts and lists them in missingFields. Consumers must accept null; this corrects previous unknown-as-zero values. Complete detail still depends on source availability and is not guaranteed. Extended video output includes authorFace when provided by the source.

Discovery completeness correction

maxResults now counts primary discovery videos by default. includeComments/includeDanmaku/includeCreator/includeTranscript add output rows without reserving slots by reducing the video request. maxOutputRows (default and maximum 10000) and the Apify charge budget remain total caps. Extra paid enrichment can increase the bill. Set maxResultsScope="global" to retain the previous combined-row limit. Source unavailability may still prevent enrichment.

The schema no longer injects a default mode that conflicts with operation aliases. Creator pagination keeps page size fixed across pages; profile enrichment failure no longer discards already recovered video rows.

All-error runs now finish as platform FAILED after saving diagnostic rows and OUTPUT. Limited-coverage non-error records are still explicitly marked and are not a claim of full task completion.

Creator catalog recovery

If WBI creator search is unavailable, the public media-list endpoint supplies creator uploads with cursor pagination. Every author MID is checked; rows with a different primary author are excluded and counted in catalogExcludedOtherAuthors (source lists may include collaborations). Source coin/share/like counts and video part IDs are preserved; missing statistics remain null. catalogTotalCount and catalogHasMore describe the source response, not a guarantee that all uploads were fetched. Pagination failures after successful pages retain partial results with warnings.

When an exhausted public creator media list reports more videos than the unique source IDs observed, catalogCountGap records that discrepancy. An exhaustion flag alone no longer disables the existing exact-author dynamic/search fallbacks in this case. The gap is a source-count discrepancy, not proof of that many missing public videos. Supplemental sources may still return nothing new. Source warnings and best-effort coverage remain; complete creator-history recall is not promised.

Optional sentiment scoring

Enable sentimentAnalysis to add sentiment: {polarity, score, method} to video and comment rows, including strict compatibility output. SnowNLP 0.12.3 runs locally with no third-party API fee; allocate at least 512 MB. Scores range from -1 to +1; the neutral band is [-0.1, 0.1]. This is a heuristic trained primarily on product reviews, not a claim of validated Bilibili sentiment accuracy. See the model documentation.

Non-Chinese or empty text uses method: no_chinese_text. Model failures have null scores and a warning. Exceptionally long inputs are scored on the first 8,000 characters with sentimentInputTruncated: true; the scraped title/description/comment remains complete. Existing runs keep this feature disabled by default.

Support

Open an issue from this Actor’s Issues tab with the run ID, expected result and a redacted input. Do not include tokens, cookies or private credentials. Include the output profile and affected field when reporting a migration mismatch.

Presentation update

The current listing adds workflow examples, grouped input documentation, output-field explanations and plan-based result discounts. Source-specific limitations above still apply. This is an independent tool and is not affiliated with the source platform.