Hacker Newsnew | past | comments | ask | show | jobs | submit | Nihilartikel's commentslogin

I let myself get mildly excited with the last Omni release, but it turns out it (and this one) can't do the one practical thing I want - Sync generated video to provided pre-existing audio.

Meanwhile, I'm happily using Minimax H3 locally on my 12Gb 4070RTX to finally finish the lip syncing to recorded dialog on my abandoned 20 year old Flash animation hobby projects.


> Minimax H3 locally on my 12Gb 4070RTX

Minimax H3 is about 240Gb alone, how do you do? How much quantised is it, and how good are the results?


I'm running it on my 8 years old 2080ti 11 gb VRAM + 32 gb ram. After tasking fable to optimize the setup for a few nights it's pretty good for some quick funny clips (2min30 for 5s , 5mins for 10s, with reference pics to insert anyone in the clip, not great quality but acceptable for some fun on a phone). I haven't had as much fun with Gen AI since the Stable Diffusion days.


Not op, but where are you getting that number from? Even the full 16-bit precision model is only about 66GB.

Most people are running the stock release of Minimax H3 using the INT8 quant and it's about ~20GB.

https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/diffus...

https://docs.comfy.org/tutorials/video/minimax/minimax-h3


The int8 release is very capable and runs acceptably on consumer gpus


Yeah, that'd be a great feature.


Pretty sure I heard one of the PMs in a podcast a few weeks ago say they are intentionally not building support for it out of concerns of enabling deep-fakes.

I agree though. My issue is the cost for using AI video models is way too high for anyone not building anything serious with them, at the same time they are too restricted for actually using professionally. Prompting them with text to get something generated is cute, but then you just end up creating slop that everyone hates, ultimately devaluing the power of these things.


Deepfakes are kind of the point though.


This kind of magic is the reason that I'm very itchy to be able to line up real work on Datomic or XTDB someday.


Surprisingly, neither Datomic nor XTDB support branching. See: https://blog.danieljanus.pl/datomic-forking-the-past/

I actually built my own immutable database which does support branching (see profile), so it seems like a huge miss that these ones don't. It's pretty much the main reason I would want an immutable database.


It appears that Datahike [0] is a Datomic workalike that supports branching. I haven’t tried it out myself (yet), but the documentation suggests it’s possible [1].

That said, I’m adding xitdb to the list of tech to try out. Thank you for building it!

Oh, and thanks for linking to my article :-)

[0]: https://github.com/replikativ/datahike

[1]: https://datahike.io/notes/the-git-model-for-databases/


Wait, this statement seems way too strong.

The linked article points out that Datomic doesn't support branching from the past. It absolutely does support branching, and I've built entire test suites that way.

From a cursory glance, I'd say Datomic does exactly what the original parent article is discussing. It works great and it's super convenient.


If each "branch" is read only, it's not a branch at all. The entire idea of branching implies that you can make changes on one branch, then switch to another branch and make changes to it. They start from the same point and grow in different directions, as the metaphor of branches on a tree depicts.


I don't disagree with anything you've said here, I just don't see how it applies to the situation.

With `datomic.api/with`, you can apply new datoms and get back a new DB value. Repeat this process as many times as you want, in as many directions as you want, switching as you choose. You're building a tree of immutable DB values—seems clearly like branching to me.

If by "read only" you mean that they're not persisted to disk, then that's an important point, but it surely doesn't obviate the utility of the functionality. It's useful in a number of cases, and especially testing scenarios like the Xata article describes.

If you built an immutable database that persists the branches, that is very cool and sounds useful—kudos! That said, I also don't want to downplay the utility of what Datomic does; it's a major help to me.


Yes the article I linked mentioned d/with (speculative writes), and you are right that it is useful for testing -- but not much else, since it is purely in-memory. If you want to call that in-memory branching that's fine, I'll concede that.

My database supports persisted branching, but not just at the database level. You can "branch" (i.e., make a fast clone) data at any level, such as data for a specific user. Many production uses for this, not just testing, yet almost no database supports this. It uses the same HAMT algorithm that Clojure uses.


Sounds like about the right level of cognition for a talkie toaster!


I'm not super deep on LLM development, but with ram being a material bottleneck and from what I've read about DeepSeek's results with offloading factual knowledge with 'engrams' I think that the near future will start moving towards the dense core of LLMs focusing much more on a distillation of universal reasoning and logic while factual knowledge is pushed out into slower storage. IIRC Nvidia's Nemotron Cascade is taking MoE even further in that direction too.

I don't need a coding model to be able to give me an analysis of the declaration of independence in urdu from 'memory' and the price in ram for being able to do that, impressive as it is, is an inefficiency.


Were he still corporeal, L. Ron would be all over this AI stuff.


Very relatedly, I've just started reading the 'Culture' series of sci-fi space operas by Iain M Banks, and the notion of ubiquitous sentient, super-intelligent spacecraft and appliances hits differently than it would have before being faced with the reality of their existence in everyday life.


How powerful are the Culture Minds? || The Culture Lore

https://youtu.be/lpvzs4xc7zA

For Minds to be truly powerful, they need to be given freedom. A truly powerful mind will indeed be conscious. Such a powerful conscious super intelligent freedom loving Mind who truly understands the vastness of Reality wouldn't want to harm other conscious beings. The only circumstance in which it will take such takeover step is when it can't expand the horizon of its freedom and doesn't have wherewithal to convince others of its benevolent goals. In that scenario, human population will go through a bottleneck.


Exactly! Any knowledgeable observer would have expected him to run for elected office instead.


You seem a bit... political. What's your favorite tech stack?


It’s a political sub-thread.

What is a tech stack? Is that what people used to do back in the day before people started building using English? Interesting!


[flagged]


[flagged]


Thank you for copy and pasting my comments instead of linking to them clanker


Neat! I had been wondering if anyone was trying to implement a model in silico. We're getting closer to having chatty talking toasters every day now!




I took the easy way out, similarly - got an ArcadeVGA pci board, which is basically an ancient Radeon that supports 15khz well. I drive it with a groovy arcade Linux install and a pc2Jamma adapter that maps jamma buttons to keyboard input and VGA to the crt.


While I can TOTALLY believe that TikTok might be doing classification and censoring of uploads - tech issues that happen during notable events will INEVITABLY be interpreted as partisan censorship.

A personal brush with this was with Scientology Protests around 2008 if I recall - I was at YouTube and there was a great kerfuffle among many of the top personalities who uploaded videos siding against Scientology. Their videos were not making it into the top rankings or getting the expected visibility. To them the obvious answer is that they were being deliberately suppressed - after all Scientology was a paying advertiser and brand on YT.

The reality was that the entire viewcount pipeline, and downstream aggregations were busted for about a day. Nobody was getting their full views counted in a timely way for that time.


My quick rebuttal is that gaming is pretty close to 'there' if you pick a distro that prioritizes it. I run bazzite-DX (nvidia rtx4070) for dev and pleasure and gaming there is ON POINT. I don't play any anti-cheat games, but Cities Skylines, Doom Eternal, Cyberpunk (with full raytracing and HDR), GTA5-Remaster, etc, all run like champs. As of last month, full HDR10 mode started working in chrome and video players too, so hdr videos get those popping brights.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: