小寄

搞点AI。
但谁能懂我？我来推特是为了看男娘的！

[小寄 @Dead_Reality_]

有考虑过用本地模型吗？
使用minimax H3，租赁5090或者pro 6000算力卡，使用AI agent循环放置任务，并使用多模态AI截取生成帧自动检查镜头效果。
其次，因为生成方式（主要是成本）有变化，你可以考虑使用多轮生成。使用灰度图/强化轮廓图进行场景和人物抽卡，然后再使用视频生视频模型上色一遍，然后再使用某项速度极快的本地放大模型（比如LTX2.5）放大分辨率。
生成时声音生成使用ASMR效果，只生成环境音，后期使用其他AI进行音轨调整。

[ZOOT @LittleTinRobot]

Sadly I will be taking a back step now re AI film making. I have worked very hard as we all have to try and build a style and brand and have co promoted the companies that I have been working with. I have made little to zero money on any of this and now that the plans are being universally removed by these companies that enabled me to carry on, now is the time to sadly call a halt.

Why?

The cost to produce under current “rates” is prohibitive especially given the glaring elephant in the room along with my not being in work.

It seems that none of these companies are factoring in the “Cutting room floor” aspect of all of this. I don’t know about anyone else but it can take me up to 20 renders on some scenes to get a good result. That’s not me being fussy that’s me having to reject “brain fart system f ups”

I cannot justify $850-1000 or an X sub for each long format that on X now gets shown to a room full of people ever since that genius Nikita Bier threw his brain spanners into the algorithm engine.

So for now I’m on pause as showing long format on X now is pointless.

I’m trying to host something locally re $$$ so if anyone has any suggestions let me know…

💪💪💪🙏🙏❤️❤️

[小寄 @Dead_Reality_]

@MaimuAkemi Greeting

[小寄 @Dead_Reality_]

@ai_xiaomu 重要个锤子。真这么重要，搞个7900xtx本地部署个27b，上下文搞个10w~20w得了。

[小寄 @Dead_Reality_]

@wei_wang 最大问题是这玩意太拉了。
真打的过mimo2.5吗

[小寄 @Dead_Reality_]

@dearemon 问题是，你如果平均用车成本2W，你得考虑没车的情况下达到近似的通勤效果消耗。
而不是说有车你开车，没车你走路或者电驴能达到相同效果。完了一算净省两万。
一年用车成本2W，至少得按公里数0.5/公里算，保险4K，保养停车费等额外支出6K，油钱10K，一年得两万公里。
已知纯打车，快车2R，专车6R

[小寄 @Dead_Reality_]

@__fanqie__ 放车上。就说是车险/话费/宽带送的

[小寄 @Dead_Reality_]

@pingzi22333 可能是机器人发福

[小寄 @Dead_Reality_]

@Kenntnis22 简单，如果是线下交易，直接要求和对方在银行大厅取款并交易现金不就完了。
如果对方做不到在银行眼皮子底下大额取款，过不了银行安控那显然来路不正啊。

[小寄 @Dead_Reality_]

@Lamrrk 说来你可能不信。我plus也是sol一句话百分之5

[小寄 @Dead_Reality_]

@pjjin574832 这人以后离婚付不起律师费的。

[小寄 @Dead_Reality_]

@rennuzhen @jianlvup 你光在这叨逼叨。不看图的吗？

[小寄 @Dead_Reality_]

@chasen_liao 没有。换codex用luna了。死慢

[小寄 @Dead_Reality_]

@Kashiko_AIart 试试LTX2.5

[小寄 @Dead_Reality_]

@meetqy 我接的1k 0.02的

[小寄 @Dead_Reality_]

@Lamrrk 然后过段时间就发现白教了

[小寄 @Dead_Reality_]

@slow_loading_ 做不到就不要做了咩。接受自己是个没能力自残的废物。也不是什么大事。毕竟就算做到了。也不酷

[小寄 @Dead_Reality_]

@yeppppp24 没毛，但具体怎么做呢？有想做的和我组队吹牛吗？
我已经同时在尝试图生视频出精灵图和纯2D绑骨两条线路图了。

[小寄 @Dead_Reality_]

@Lea757651167628 @Lunarisozdp 骗骗自己就完了。就这种让人玩怀孕就拉黑的臭鸡，听你这语气被这么玩还挺自豪的？那你老牛逼了。

[小寄 @Dead_Reality_]

@Fomsky_Wei Luna巨慢，开极速比deepseekv4f慢几乎一倍算吗？