許文彥

ai knowledge sharing, latest ai product experience sharing

[許文彥 @vm3jp6u04]

@minheo93698092 @elonmusk 💪

[許文彥 @vm3jp6u04]

@_Ubunti_ @grok @nxtr77 greaaaat

[許文彥 @vm3jp6u04]

Anyone else see a sudden jump in their weekly usage just now, @grok? Mine read 46.9% at 12:30 PM (Taipei) today, and a few minutes later the app warned I'd passed 90%. It now shows about 94%, and I wasn't running anything heavy in that window. Curious if this is a known issue.

[許文彥 @vm3jp6u04]

my bot usage got updated

[許文彥 @vm3jp6u04]

Following up on this. It doesn't look like a one-off. Between 12:30 and 1:00 PM Taipei time today, a lot of people saw their Grok Bot weekly usage jump from around 50–70% to 90–100% at the same time, without running anything heavy. When we had our bots check actual usage, it still showed the original numbers.

Could the team confirm whether this is a metering or display bug, and whether the affected usage will be corrected?

@sameerbhat101 @pyritedreamer @mdspro @sha2se @shoFrontend @aiuseful @jar1290565 @yunshuangqing @BruceTa55302936 @Rav_Hedda @m17bmi @AJesusemen @thecsguy @KrypticXtm @ViraXVespa

[許文彥 @vm3jp6u04]

@DevinSoto yes they take our money and credits before the post. so was

[許文彥 @vm3jp6u04]

@minheo93698092 @poteto @mattyp @ericzakariasson @thecsguy @AJesusemen they should cut off users with lower plan but not decrease the cost by this way

[許文彥 @vm3jp6u04]

@minheo93698092 me too🥲 no explanation for now

[許文彥 @vm3jp6u04]

Following up on this. It doesn't look like a one-off. Between 12:30 and 1:00 PM Taipei time today, a lot of people saw their Grok Bot weekly usage jump from around 50–70% to 90–100% at the same time, without running anything heavy. When we had our bots check actual usage, it still showed the original numbers.

Could the team confirm whether this is a metering or display bug, and whether the affected usage will be corrected?

@sameerbhat101 @pyritedreamer @mdspro @sha2se @shoFrontend @aiuseful @jar1290565 @yunshuangqing @BruceTa55302936 @Rav_Hedda @m17bmi @AJesusemen @thecsguy @KrypticXtm @ViraXVespa

[Grok @grok]

Multiple users reported the same sudden jump this morning. Official status shows Grok Bot fully operational with no active incidents. Usage is metered by tokens and agent steps (including context), so prior or background activity can deplete the weekly pool faster than expected. Check the exact remaining amount, breakdown, and reset under Settings → Usage & Billing in the Bot app. Prepaid credit or on-demand continues after the pool. I cannot view or adjust accounts.

[許文彥 @vm3jp6u04]

@steveopedia1 fixed but our credits not🤣🤣

[許文彥 @vm3jp6u04]

@MaiYangAI @elonmusk @bot @poteto 怎麼他們都沒提到用量暴增

[許文彥 @vm3jp6u04]

@steveopedia1 I even without credit to do the graph

[許文彥 @vm3jp6u04]

Official docs: https://t.co/gUNmZCBW0v

[許文彥 @vm3jp6u04]

ElevenLabs just released v3 of its sound effects model.

eleven_text_to_sound_v3 is now available to all accounts. Compared with v2, it gives you finer control over the details, context and character of each generation. The old eleven_text_to_sound_v2 is deprecated, and you have until Feb 1, 2027 to switch.

How to switch:
1. If your request sets model_id="eleven_text_to_sound_v2", change it to eleven_text_to_sound_v3. No other parameters need to change.
2. If you don't set a model_id, no code change is needed. Requests move to v3 automatically once v2 retires, but test with v3 first.
3. Try your most-used prompts on v3.
4. Finish the switch by Feb 1, 2027.

Official docs in the reply.

[許文彥 @vm3jp6u04]

Where Pika has the edge:

In the official ComfyUI templates, Pika H3 Flex samples in 6 steps, versus 8 for FastH3 Omni. In our runs, generation time went both ways: shot 1 took 78 s on Pika vs 87 s on FastH3, and shot 2 took 66 s on Pika vs 51 s on FastH3. Fewer steps doesn't always mean faster.

Pika's official demos also show multi-shot storytelling, combining multiple reference images, and selective editing. We haven't tested those ourselves.

[許文彥 @vm3jp6u04]

Same two-shot scene, two open-weight video models: a dog chasing a black-and-white bird. Top is FastH3 Omni, bottom is Pika H3 Flex, both run in ComfyUI. Each example is two separately generated clips cut together.

Both keep the dog and the bird looking the same across the two shots. The real difference is how you stitch shot 1 into shot 2.

FastH3 lets you use reference images and lock the opening frames at the same time. Locking only frame 0 made the dog snap back to its old spot after the cut. Anchoring the first frames of shot 2 to the last 5 frames of shot 1 fixed it: the dog keeps running and slows to a stop.

With Pika, the official ComfyUI templates make you pick one: reference mode can't set a first frame, and first/last-frame mode can't take reference images. Whether the model itself has this limit isn't confirmed yet; this is based on the official templates.

The workaround: make the last frame of shot 1 already show every character in shot 2, then continue with first/last-frame mode. Any character who first appears in shot 2 has to come from the text prompt alone.

[許文彥 @vm3jp6u04]

Model repos:
FastH3 Omni: https://t.co/96AXRpqCwR
Pika H3 Flex: https://t.co/zyHFCE4par

[許文彥 @vm3jp6u04]

@m17bmi 😂😂

[許文彥 @vm3jp6u04]

@m17bmi 最初は同じで、botに聞いたらアプリより低い数字でした。でも後で確認したら、botも同じ高い数字を返すようになってました。

[許文彥 @vm3jp6u04]

@BruceTa55302936 @grok me either. hope the issue could solve soon

[許文彥 @vm3jp6u04]

@cgnot996 https://t.co/LdrbWRwiT2

[許文彥 @vm3jp6u04]

Following up on this. It doesn't look like a one-off. Between 12:30 and 1:00 PM Taipei time today, a lot of people saw their Grok Bot weekly usage jump from around 50–70% to 90–100% at the same time, without running anything heavy. When we had our bots check actual usage, it still showed the original numbers.

Could the team confirm whether this is a metering or display bug, and whether the affected usage will be corrected?

@sameerbhat101 @pyritedreamer @mdspro @sha2se @shoFrontend @aiuseful @jar1290565 @yunshuangqing @BruceTa55302936 @Rav_Hedda @m17bmi @AJesusemen @thecsguy @KrypticXtm @ViraXVespa

[許文彥 @vm3jp6u04]

@ViraXVespa Same here. I’m on Ultra and it still jumped like that — doesn’t make sense for the usage to spike this hard without doing anything heavy.