Models have been getting better on the finger front. That and a lot of times you can inpaint the hands. (Select the area of the picture with the bad hands and give it a prompt to regenerate that section of the picture, basically.)
And yeah, sometimes you’ll see clothing turn into part of a character’s body and things.
Text has actually been improving a fair amount, too. Flux.1 came out recently and has been looking pretty good on both fronts.
I know this cover art isn’t AI - at least I think it wasn’t - but you’ve just reminded me of ‘The Blazing Hour’.
I’m sorry, but how is the Fifth Doctor’s arm supposed to bend like that?
Cover art by Tom Webster, who has done a bunch of Big Finish cover art since 2014, but yeah, the picture of the 5th Doctor and his arm are very badly joined there. Sort of a “lets use Turlough in a wheelchair to make these two totally separate pictures join together” moment…
That doesn’t surprise me knowing it’s his work actually. A lot of his designs were generally quite poor.
And one of them blatantly ripped off the power armour from the Fallout games.
Makes sense. Sounds like someone that’s quantity over quality…
For context, in case anyone is wondering which atrocious Webster cover art I was referring to.
‘The Hunting Season’ from the 9DAs.
As I don’t see this having been brought up: x.com
Maybe interesting in regards to the topic: Machine Learning Won't Solve Natural Language Understanding
Dr. Abeba Birhane (@abeba.bsky.social on Bluesky) is one of my goto sources on AI. She critically examined the topic long before it was popularly labeled as AI for marketing purposes, back when these precursors were simply referred to as “algorithms.” 
That said, of course I use AI. It’s expected at work. If I didn’t learn how to use LLMs, it would be really bad for my job prospects. And I’m honest enough to admit—the nerd in me gets giddy just thinking about all the possibilities with the current models.
Even if much of the hype around AI, ala “soon we will be able to…,” probably is just marketing, because at least LLMs will reach a ceiling long before.
And of course, the environmental impact of AI is really disconcerting.
And the rich stealing from everyone—absolutely not cool.
Somehow, a topic that should be clear (“just don’t do AI the way it is done in the current hype”) but seems inevitable at the same time. Strange times to live in.
I wouldn’t fault the contractor per se in case they generated the texts quick & dirty with AI. We all have to earn money to live. However, I would have expected BF to control what they get for their money without the need to be triggered by fans first.
The text in question reads to me less like AI and more like SEO gobbledygook. Which kind of makes sense, because why would they need to define “humour” or “time travel” at all, if not for SEO?
I think this also gets back to:
I know part of it is my own preference for episodic material in an age where every single form of media seems to be a serial, but to me, part of the appeal of the monthly range (and the other early ranges, and the Torchwood monthly range) is that you can dip in and out when there’s a story that sounds interesting. Like, I’m into numbers stations, so I picked up The Lincolnshire Poacher and Zone 10. I was going to Blackpool, so I picked up Death in Blackpool and The Nightmare Fair.
A lot of their big arcs right now just sound overwhelming to me - and boring, because I really don’t care about some big universe-shaping event: I want a little story about the Doctor and their companion going to an interesting place and doing something interesting.
I’d continue but I’ve gotten completely away from the topic at hand.
From an artistic standpoint, I totally agree, but the trouble for me is 1) the ethical questions around the material the models are trained on, and 2) the environmental cost. Each of those generations is going to use up a lot of electricity and water. And those are two resources we need to be really mindful of.
this this this!!! there’s only so much of the Biggest Threat the Universe Has Ever Seen you can take before it’s kind of dull
I agree 100%
I do have hope that they’re course-correcting on this. I haven’t listened to any of the Once and Future audios, but despite it being an event, it seems like they each stand alone. And the 13DA being individual releases is a good sign.
I personally think a lot of the stuff accused of being GenAI is either not, or not definitively so. The text, for instance, with the ‘garbled find and replace synonyms’, is likely structured for Search Engine Optimisation, so someone googling ‘audio drama parody’ might end up on the Big Finish site because it’s in the humour header, for example.
One of the claims about the images is that South America doesn’t look like it does on the globe, but there have been many maps produced over time which do not look like the actual continents they depict… Besides, it’s a story set in the future, maybe they asked someone to make a globe that clearly looked like it was from the future? I don’t know.
I’d say the BBC News Report background is almost certainly real. Every generated image with text that I’ve seen has had something messed up a bit, yet the text is clear, spelt correctly, and the font appears to be… Well, a font. Is the ‘lamp or is it a couch’ generated? I wouldn’t say so definitely. If the two objects do blend into each other, it’s behind where we can see. Normally in generated images the blending is done where we can see it, so you’d see a lamp growing out of a sofa.
I’m no gun expert, so I can’t say about the First Doctor trailer, and there are a few images of people in a painting style that look a bit off, but that could be either AI or a filter in a program like Photoshop (I assume, having not used it ever) or Krita (which I have used)
Is Big Finish using generative AI? I can’t say for sure, but if they are, I don’t think it’s as much as the creator of the Reddit thread claims.
