Auto Blend in LAPLACE Chatterbox continuously watches the danmaku in the current live room's chat. When several viewers repeat the same content within a short window, it follows along with the matching danmaku, keeping your sending rhythm in line with real viewers
How this differs from AI Companion Chat: Auto Blend only follows along with danmaku that has already appeared in chat, and with YOLO mode on an LLM rewrites that content before it goes out — the point is still blending into the crowd. AI Companion Chat instead has an LLM write new danmaku from scratch out of the Simultaneous Interpretation transcript and viewer danmaku, playing an "AI viewer" that interacts with the streamer. The former depends on real viewers to create an opening; the latter keeps producing on its own even when chat is quiet
Features
- Monitors the live room's chat stream in real time, with no external API involved
- Trigger thresholds are configured independently, and both the number of distinct senders and the total occurrences within the window must be met
- A per-content cooldown keeps it from following along with the same danmaku again too soon; Auto Cooldown can adjust that window to the room's current danmaku rate (2 to 60 seconds)
- Avoid Consecutive Repeats can discard later danmaku identical to what was just sent, so continued spam from viewers cannot retrigger it
- @-reply danmaku is always excluded from the counts
- Danmaku you send yourself is never counted, so the feature cannot feed itself
- A cross-room user blocklist, where danmaku from blocked users stays out of the counts
- A cross-room message blocklist, where a specific string can be added in one click and no longer counts
- Platform-locked emotes (fan club / captain / admiral / governor / level-gated) are skipped, never counting toward a trigger and never sent as follow-along content
- Danmaku identified as a large emote is skipped so it cannot pollute the counts
- Replacement rules can optionally be applied to follow-along content
- Optional YOLO mode, where an LLM rewrites the danmaku after a trigger, with one LLM call per trigger
- Reuses the Auto Send panel's send interval, max length, random characters, random color, and interval randomness settings
- The follow-along count is configurable, so one trigger can send several times in a row
- The on/off state can be remembered per live room and restored the next time you enter it
Usage
Enable Auto Blend
- Open a bilibili live room and click the Danmaku Assistant floating button in the lower right to expand the panel
- Switch to the Send tab and expand the Auto Blend section
- Click Start Blending to begin monitoring, and click again to stop
- While Auto Blend is running, the Send tab and the Auto Blend panel title show a purple 🟣 indicator
Configure Trigger Conditions
"Trigger: X people repeat it Z times within Y seconds" at the top of the panel sets the thresholds, and all three numbers must be met before it follows along
| Option | Description | Default |
|---|---|---|
| Viewers (X) | How many different viewers must send the same danmaku to trigger | 3 |
| Window (Y) | Length of the counting window, in seconds | 15 |
| Repeats (Z) | Total occurrences of the same danmaku within the window | 3 |
Configure Follow-Along Behavior
| Option | Description | Default |
|---|---|---|
| Follow-Along Count | How many times to send in a row after one trigger (up to 20) | 1 |
| Same-Content Cooldown | Cooldown in seconds after following along with a given danmaku, during which it cannot trigger again | 10 |
| Auto Cooldown | Derives the cooldown from the room's current danmaku rate (CPM), between 2 and 60 seconds; the fixed value above is ignored while this is on | Off |
| Avoid Consecutive Repeats | After a follow-along, danmaku identical to what was just sent is discarded so it cannot trigger again; the state lasts only for the current session | Off |
Other Options
| Option | Description |
|---|---|
| Apply Replacement Rules | Whether cloud and local replacement rules apply to follow-along sends |
| YOLO Mode | An LLM rewrites the follow-along content before it goes out, with one LLM call per trigger (repeated sends share the same rewrite); configure the API under Settings → LLM Settings first |
| Keep Auto Blend State for This Live Room | Remembers the on/off state by room ID and restores it the next time you enter that room |
Auto Blend reuses the Auto Send panel's Send Interval, Max Length, Random Color, Random Characters, and Interval Randomness settings, so there is nothing to set twice
Block Specific Users
Some accounts spam chat or post noise. Use the Blend Blocklist to keep them out of the counts:
- Click any username or danmaku in the chat panel to open bilibili's built-in menu
- A 🟣 Add to Blend Blocklist item appears at the bottom of the menu, and clicking it adds that user to the list
- For a user already on the list, the item reads 🟣 Remove from Blend Blocklist, and clicking it takes them off
- Danmaku from blocked users does not count toward Auto Blend's viewer or repeat totals
- The blocklist applies across rooms, and you can review it, remove single entries, or clear it in one click under Auto Blend Blocklist in the Settings tab
Blocking is always available, so you can add known spam accounts ahead of time even while Auto Blend is off
Block Specific Text
Fixed strings that keep coming back — copypasta, ads, flamebait — can trigger Auto Blend over and over. The Message Blocklist drops them from the trigger counts outright:
- Manually add or edit blocklist text, or clear it in one click, under Auto Blend Message Blocklist in the Settings tab
- Matching danmaku is discarded before it reaches the candidate pool, so it contributes nothing to the trigger conditions
- Matching is on the full text, after leading and trailing whitespace is trimmed, and like the user blocklist it applies across rooms
Notes
- Follow-along sends go through the shared send queue, subject to the same bilibili rate limit as Auto Send, Simultaneous Interpretation, and manual sends, with at least 1 second between sends
- When Auto Send and Auto Blend both run, manual actions such as a normal send or a one-click meme send take priority, and the automatic follow-along yields
- If DOM changes make sender UIDs unreadable, it falls back to judging by total occurrences and keeps working
- Platform-locked emotes (fan club / captain / admiral / governor / level-gated) are filtered out at the counting stage, so they never affect a trigger and are never sent when following along
- Large-emote danmaku, such as official sticker images, is filtered at the same stage
- With YOLO mode on, the rewrite happens after the trigger and before the send; if the LLM call fails, the original content is sent instead
Last updated on September 3, 2026