Saturday, June 10, 2023

Three AIs I Use A Lot, Two I Soon Will and One I'll Be Staying The Hell Away From...


As of today, I count more than twenty apps or websites, bookmarked in my AI Tools folder. It feels like I'm adding more every day. 

Most, I treat as toys. Get them out occasionally. Play with them. Put them back. Forget about them for weeks. Others, I use more constructively but also situationally, only when I have something specific I need to do.

There are three that I use regularly, by which I mean just about every day. I'm beginning to lean on all of them, more and more, just to get the posts out.

They are:

Snapedit  

Tipa put me onto this one a few months back and I've come to rely on it heavily. It solves so many problems and saves me so much time. Tough to recall how I got along without it. 

Snapedit gets rid of the stuff you don't want in the picture. You can upload any image and with a few quick sweeps of the mouse erase just about anything that doesn't fit. So far, so Photoshop. Not that I can afford Photoshop but I'm sure there's some free knock-off I could use. If could be bothered to look, which so far I never have. 

That's not the good part, though. The good part is that Snapedit uses AI to fill in the gaps with something that looks right. Right, not perfect. I'm sure no professional commercial artist would want to put their name to it but it's very much good enough for a blog post. 

The huge advantage I see, the reason I use it almost every day, is that I don't have to worry about taking screenshots with the UI on show. I also don't have to pay so much attention to other players or NPCs wandering into shot with their dumb names and guild tags cluttering up the scene. I can just wipe them out with Snapedit.

It must have saved me literally hours already. What with Snapedit and all the post-processing I do in Paint.net, few screenshots I use these days bear close comparison with the originals, let alone the game. Authenticity is so last decade.



Stable Diffusion  (Hugging Face version) 

My current favorite text-to-image generator. There are so many. I used to use Craiyon, which runs off the old Dall-E Mini, but the results, while amusing, weren't always practical. For a while I used the regular Stable Diffusion. It was fine but then it morphed into DreamStudio and that wants money.

Fortunately, there's still an unrestricted version of the original Stable Diffusion 2.1 build running under the rubric of Hugging Face, an open source platform, whose mission is to "Build, train and deploy state of the art models powered by the reference open source in machine learning." Snappy. That's what I'm using to produce most of the AI-generated images you see in posts here nowadays.

I like Stable Diffusion mostly because it's simple, fast and gives reliably accurate results. In sets of four big pictures. I can usually get at least one I like on the first or second try. They fit very nicely on the blog without much editing. The AI is responsive to prompts on style, reliably returning 50s magazine illustrations or 70s comic book art without needing a lot of tweaks or instruction.

It has the most minimal of interfaces. All you do is type a sentence or a phrase. You can also add a negative instruction to avoid results you don't want. There's an "Advanced" toggle but all that gets you is a slider that affects the Guidance Scale, the degree to which the AI follows the prompt. I generally leave that where it is.

Best of all, it's completely free for unrestricted use. Can't ask for more.



Bard

Bard is Google's chatbot. It's one of the Big Three, the others being ChatGPT+ and Microsoft's Bing. Of the three, I strongly prefer Bard, even though ChatGPT+ is definitely more versatile and Bing probably more accurate. 

The increasingly noticeable issue with ChatGpt+ is that it was trained exclusively on a data set that was gathered prior to September 2021. It's never been updated. Since most of what I want to ask relates to current events, that makes it fairly useless and getting more so by the day. 

I would always go to ChatGPT+ first for AI-generated fiction. I think it does a more creative job making stuff up than the others. Fiction is rarely what I need for the blog, though. 

Usually I want a precis of some news item or story to save time paging through multiple web pages. Bing can bring the data and quote the source but it seems to me to be nothing more than a minimally-enhanced search engine. An enhanced version of a search engine I would never normally use, at that. If I wanted to click on links and read primary sources I'd stick to Google Search.

Bard is a lot closer to the kind of AI-driven research assistant I'm looking for. It's chatty and friendly and communicates in reasonably demotic English. It offers opinions and suggestions, which I find both amusing and helpful. Bard also likes the sound of its own voice and never stints on the quantity, whereas Bing always acts like it's got better things to do with its time than chat and has to pay for every word out of its lunch money.

I can often quote whole paragraphs from Bard verbatim, although so far I've stopped short of just dropping one in and pretending I wrote it. That day is coming though, count on it, if only so I can see if anyone notices and calls me out.

The problem with using Bard as a research assistant - and it's a very big problem indeed - is that when it doesn't know something it's more than happy to make it up. Worse still, when challenged it's most likely to double down and insist it's right, trash-talking anyone that claims otherwise.

In this way I suspect its even more like an actual research assistant - some slacker with a know-nothing boss who can't tell and wouldn't care anyway. But I can tell. Every time Bard tells me something I don't already know, I go check with Bard's prissy elder sibling, Google. 

Still makes using Bard much quicker than going to Google right off the bat. Mostly, when I ask Bard to give me chapter and verse on something, it's because I already know most of the details. I just want them laid out neatly for me. Bard is good at that. It just needs watching. Closely.

Those are the three AI services I rely on but more are coming on stream every day. Here come two more I plan on using pretty often in days and weeks to come - and one I don't plan on using at all.



NightCafe

I can't remember how I came across NightCafe, even though it can only have been a few days ago. Insert Ferris Bueller quote here. It's yet another text-to-image generator but one that comes with a wide array of ready-to-use filters, everything from anime to neo-impressionism. Handy timesaver but more important is the way NightCafe allows you to produce linked sequences of images, each new one iterating on the last. 

Very useful for making slideshow videos but also for illustrating blog posts on a theme. I expect to be using this one often although how often depends on how many free Credits I can snag. 

The payment system operates on tokens, boringly called Credits. The number of iterations you can generate costs a set number of tokens. A single image is free but multiple iterations cost .25 of a credit each. The stages available are 4, 9 or 16.

You can buy credits either in packs or through a variety of monthly subscriptions but you can also earn them for free by doing various things that feel eerily similar to dailies in an mmorpg. Without doing anything much at all, I managed to generate more than enough credits for the images I needed yesterday and right now I have eighteen and a half credits left. Where the half came from I have no idea.

If I can keep myself supplied with enough free credits I'll be visiting the NightCafe often. 



Uncrop

I found this one yesterday, when I was looking for an image generator that could output results in landscape format. All the text-to-image generators I've tried like either square or grid or portrait format, none of which is ideal.

I wasn't having much luck finding one. Googling "Text-to-image generators - landscape format" mostly gets links to apps that generate landscapes. As I was going through the possibles I chanced upon Uncrop, which kinda-sorta solved my immediate issue even though it didn't match the search terms at all.

Uncrop is a new app from Clipdrop. Clipdrop is a more sophisticated Snapedit. it runs on Windows, Android and iOS. It has a free version but you have to take a watermark, which pretty much makes it useless for the intended purpose, I'd say.

Uncrop, for the moment at least, comes free with no watermark. You need to register an email address to get more than a few turns but I did all of this post with it and it hasn't asked me for money yet. It's an "Aspect Ratio Editor". Allegedly "The Ultimate Aspect Ratio Editor". I wouldn't know. It's the only Aspect Ratio Editor I've ever seen.

The website says it can "change the ratio of any image by creating an expanded background to complement any existing photo or image". Sounds fancy. What it actually does is use Stable Diffusion to generate supposedly suitable images to fill the extra space created when you specify a different size of canvas.

I think the idea is to make the end result look like a single, seamless whole. Mostly it works although it has a habit of throwing in completely off-topic elements, like the leaping duck and friends in the picture above. Sometimes it just adds panels on either side of the original to create a tryptych. Whatever it does, for my purposes the results I'm getting are absolutely fine.

And finally, one I almost feel guilty mentioning. Do not take this as any kind of recommendation. It's more of a warning as well as a foretaste of things to come. 



Paraphrasingtool

This is far and away the dodgiest suite of tools I've found so far. It's obviously meant for academic and professional use, by which I mean students who don't want to write their own essays and journalists who don't want to write their own copy. And bloggers, too, obviously.

It's priced accordingly but there's a free option. You get 20,000 characters free but it doesn't say if that's a one-off deal or if it's 20k every so many hours, days, weeks or months. That's for the main function, paraphrasing. There are lots of specialised subfunctions, too, some of which cost, some of which come free.

There wouldn't seem to be any immediate reason to use this set-up over ChatGPT+ or Bard, both of which are perfectly capable of banging out a thousand words of artless twaddle on any topic you care to give them, except that the awkwardly-named Paraphrasingtool (It really needs to paraphrase itself.) comes with a bunch of presets that save you the trouble of laying out your terms of reference beforehand.

I was there because I'd googled "AI Podcast Generators" and I'd done that because I'd seen an article suggesting there was about to be a glut of podcasts written and voiced by AIs and I wanted to see if I could get in on the act. I haven't found an app yet that claims to be able to do the whole thing - produce a fully-voiced podcast from a single line of text - but I've gotten it down to two stages, of which this is the first.

I got PPT's (Not typing that stupid name out every time.) Podcast Script Generator to write me the script for a podcast on SW:TOR's move to Broadsword. Then I had Elevenlabs voice it for me. 

The script was dull and factually inaccurate but so are most podcasts. It was at least formatted correctly and useable straight out of the box, provided you didn't expect anyone to enjoy listening to it. It was also about three times too long for Elevenlabs free service so the audiofile I made cuts off in the middle of a sentence. It's a blessed relief.

I could have cut and pasted and edited my way to a complete and finished version but why would I? No-one would ever listen to it, not even me. The point was to prove concept and the truncated version did that just fine. Here it is on my YouTube channel. I don't recommend listening to it out of anything other than curiosity.

 

PPT has presets for all kinds of short-form writing, from poetry to cover letters. It can offer you Blog Ideas and Blog Outlines and even a Blog Generator. All of them are free. I haven't tried any. I don't plan on trying any. If I ever do try any, I won't be talking about it.

Curiously, there are also presets both for detecting content written by AIs and for bypassing other apps designed to do the same. Got you covered going both ways, there. I can imagine either of those getting plenty of use as the AI wars heat up.

Which they undoubtedly will. These are just the opening skirmishes. This isn't the mythical metaverse or numbskull NFTs or crappy crypto we're talking about. AI's not going to be forgotten in a week or two, when the next dumb idea shows up. 

No, this is the actual future, like it or not and we're all going to need a good set of tools if we're going help build it - or hide from it. These are just a few of mine. Happy to share!

And I'm always looking to add some more, so if you find any good ones, don't keep them to yourselves.

5 comments:

  1. Kipple Online: now with extra AI! And remember:

    Kipple Online is a way of expressing creativity, humor or satire by creating or sharing digital content that is absurd, random or nonsensical.

    Kipple Online is a form of digital recycling, reusing or repurposing of existing content that would otherwise be discarded or forgotten.

    Kipple Online is a challenge -or a game- to find meaning, value or beauty in seemingly worthless or chaotic content.

    ReplyDelete
    Replies
    1. Heh! You had me googling "Kipple Online" there to see if it was real. It sounds absolutely amazing. I'd sign up in a heartbeat.

      When I was in my teens I read a lot of Philip K Dick. I thought even then that what he called "kipple", supposedly the accretive residue of entropy, particularly in a domestic environment, seemed like a fascinating and essential resource, something to preserve and cherish, not fear and avoid. I've always tried to keep as much of my past as possible, first physical, now both physical and digital but I've made little effort to sort or order it. Just allowing to build up naturally creates a wonderful, mesmeric chaos. When I'm in the mood I dig into it for inspiration or entertainment and I can always rely on synchronicity and randomness to feed me what I need.

      I find the AI illustrations I'm using in these posts evocative, emotive, mysterious and frequently beautiful. Also sometimes silly, weird and confusing, which are also excellent qualities. I also feel a very strong sense of achievement and satisfaction from having made them - a *lot* stronger than I ever got from anything I ever drew or painted, that's for sure. I'm keeping far more in folders on my hard drive than you'll ever see here, just so I can look at them and feel good about having made them or just that they exist.

      I haven't done a lot of work on trying to produce interesting prose using AI but when Tipa was doing that a year or so back, the results she got seemed well up to par with fantasy written by humans. I think she probably did a lot in the edit but the AI did the heavy lifting. I do want to work on some AI fiction but I'm not sure yet what I need to do to train the AI on my prose style. I need to look into that.

      It's still so very early in the process, though. The stuff we're seeing and hearing now is going to look quaint and primitive in just a few years, I'm sure. Unless there turns out to be a technical ceiling that's not aparent yet. In which case, I'll still be pretty happy to play with the toys we've already got.

      Delete
  2. TMW had to get to it sooner or later...

    ;)

    -- 7rlsy

    "Chatted Up"

    https://thenib.com/author/tom-tomorrow/

    ReplyDelete
    Replies
    1. oh dear... I've only just now found this music film (vid, in later times) of George at it and... bits of it look like horrid AI swings.

      Mebbe Eric Idle has a Hawking take on this existential confusion... Or mebbe it's just a "Ripping Yarn" with additional material by Terry Southern, Phil Dick, Ernie Kovacs; art direction by Gilliam; Produced by Ian MacNaugton for 92p and a bottle of Bell's whisky...

      Directed, of course, by whoever is Eric's current solicitor

      Delete
    2. https://www.youtube.com/watch?v=Coz0TmK2ZIg

      Delete

Wider Two Column Modification courtesy of The Blogger Guide