Author Topic: Talking Shop: the thread for technical discussions, tools, and updates  (Read 18454 times)

0 Members and 5 Guests are viewing this topic.

Offline raphaelsdream

  • Member
  • **
  • Posts: 54
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #25 on: October 19, 2026, 09:46:44 pm »
Also, although I love making this stuff, and kinda 'coming out' as kinky after many years of hiding it, I love jokes more than anything - shouldn't this threat be called 'Talking Slop"?  :D

Offline Lord of Puppets

  • Member
  • ****
  • Posts: 1780
  • Gender: Male
  • May I pull your strings?
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #26 on: October 21, 2026, 10:54:33 am »
I'm finding making longer videos with Minimax H3 to be a nightmare. Effect precede cause no matter how clearly I lay out the chain of events, dialogue comes out of the wrong character no matter how clear I am on who is speaking. Any tips?
Obedience brings experiences you might enjoy.

Offline Wendall

  • Member
  • **
  • Posts: 368
  • I am a llama!
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #27 on: October 21, 2026, 01:32:47 pm »
Also, although I love making this stuff, and kinda 'coming out' as kinky after many years of hiding it, I love jokes more than anything - shouldn't this threat be called 'Talking Slop"?  :D
lol, nice catch
Quote
I'm finding making longer videos with Minimax H3 to be a nightmare. Effect precede cause no matter how clearly I lay out the chain of events, dialogue comes out of the wrong character no matter how clear I am on who is speaking. Any tips?
I don't have a ton of experience with the most recent versions of Minimax but this was a huge issue across many AI platforms in the past. Grok Imagine (my main go-to) has recently gotten much better in maintaining coherence to prompt guidelines (eg maintaining the look of a character or a room, getting the right characters to say the right lines in the right tone of voice), with the caveat being that videos are still limited to 15 seconds (so, however much context you can cram into that time frame). Depending on specifically what you're going for, it might be the kind of thing you could break up into several distinct prompts, and then weave all the raw outputs into a single video after words. For instance, generate one video which is *just* the hypnotist saying his lines, a second video which is just the hypnotizee saying her lines. Could even do a shot/reverse-shot like in a real tv show.

Offline raphaelsdream

  • Member
  • **
  • Posts: 54
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #28 on: November 12, 2026, 12:37:06 am »
I don't know if I'd ever accept money for making these story/image vignettes that I make, but I've had quite a few requests, either for image-stories on my DA, or private requests, which I've just done 'for the love of the game' and the number is increasing.

If I was ever to try and accept actual commissions, how could I do that?

I'm not sure if I think it's ethical - the words and ideas are mine but the images are (absurdly carefully constructed) AI. But if I did, I have no idea how. Patreon? Ko-Fi? That is my knowledge of money and the internet exhausted. Those two exist, and I've never used them.

Offline Svengli

  • Member
  • **
  • Posts: 106
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #29 on: November 13, 2026, 01:54:54 am »

I'm finding making longer videos with Minimax H3 to be a nightmare. Effect precede cause no matter how clearly I lay out the chain of events, dialogue comes out of the wrong character no matter how clear I am on who is speaking. Any tips?

I think longer video is simply extremely difficult, if not basically impossible. I've looked at a moderate number of both MC video and regular video and think Grimtiding's Cheer Teacher is still the best AI video I've ever seen. That's a compliment to them but it's also a statement of how bad most longer AI videos are. I mean, the showcase videos by companies making AI tools suck (inconsistent characters and landscape, etc). That's significant.

There are a bunch of tips you can find on youtube; character sheets, cinematic language (shot-description). break-up into smaller scenes and be very exact in your language - use positive language first but negative language as needed. Choose a model for your purposes (I've heard Gemini is best for long, consistent scenes and I think that's what Grimtidings uses). But I think these only go so far.

Of course, another factor is the feature length film and video is normally very, very expensive and time-consuming. I had a roommate years ago who spent ten years trying to finish a ten minute film (and failed). Google tells me Hollywood movies cost between $400K and $4M per minute. And much of that money is people doing stuff (lighting, costumes, scripts etc, etc). So AI saves a huge amount of effort but something decent is still going to require a lot of effort. 


 



Offline Svengli

  • Member
  • **
  • Posts: 106
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #30 on: November 13, 2026, 05:39:52 pm »

I don't want to be negative, I'm just thinking out loud. I follow several AI generated accounts on Instagram and they can effectively small vignette level video with consistent characters. We'll see how this technology progresses. 

Offline raphaelsdream

  • Member
  • **
  • Posts: 54
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #31 on: November 14, 2026, 08:01:28 am »

I don't want to be negative, I'm just thinking out loud. I follow several AI generated accounts on Instagram and they can effectively small vignette level video with consistent characters. We'll see how this technology progresses.

Can you share which ones - both for the 'talking shop' side and enjoying seeing them?

Offline Wendall

  • Member
  • **
  • Posts: 368
  • I am a llama!
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #32 on: November 14, 2026, 09:57:12 pm »
A little trick I've been having success with lately in Nanobanana (but I imagine it'd work with most current models) - go to https://gridmakerpro.com/ , upload a plain white jpg with the dimensions of whatever you want your final AI image to be. Select a grid template (some ones I've had success with are the golden spiral, the golden triangles, and most of the sacred geometry options). Download the white jpg with the grid over top. Now, in the AI generator, write your prompt like usual, upload all your character references etc, but also upload the grid image. Add something to the start of the prompt along the lines of "Use the line drawing as a guide for image layout - everything in the image is guided by the lines, but we don't actually see the lines themselves" (might have to be worded slightly differently depending on the model). I find doing this instantly creates much more dynamic image compositions.

This method seems to work much better than simply prompting the AI to follow composition rules like "rule of thirds" - sometimes it'll get in the ballpark, but usually not very precise. And using more artsy prompting like "cinematic masterpiece" will produce a cool result - which is repeated exactly the same way every time you use those keywords in a prompt. But actually uploading a grid or a layout plan causes the AI to be, on the one hand, much more precise, while on the other, not simply adding the same superficial polish to every single image.

Offline Svengli

  • Member
  • **
  • Posts: 106
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #33 on: November 14, 2026, 10:45:32 pm »

I don't want to be negative, I'm just thinking out loud. I follow several AI generated accounts on Instagram and they can effectively small vignette level video with consistent characters. We'll see how this technology progresses.

Can you share which ones - both for the 'talking shop' side and enjoying seeing them?

Here's a few:

https://www.instagram.com/ssunshinessmiless/reels/
https://www.instagram.com/stories/ai.myikigai/
https://www.instagram.com/charli.runs/



Offline Dazz1212

  • Member
  • **
  • Posts: 26
  • I am a llama!
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #34 on: November 15, 2026, 04:24:56 am »
I'm finding making longer videos with Minimax H3 to be a nightmare. Effect precede cause no matter how clearly I lay out the chain of events, dialogue comes out of the wrong character no matter how clear I am on who is speaking. Any tips?

I don't use Minimax so I can't be much help, if any. For more dialogue heavy scenes it may just be easier for you to just do a ton of shorter scenes and stitch them together at the point when a different character speaks. That way it may not be so noticeable and if Minimax lets you choose a starting image then you can keep continuity without each take looking different. 

Offline Hypno-Canuck

  • Member
  • ****
  • Posts: 2120
  • Gender: Male
  • Those Thrilling Days of Yesteryear
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #35 on: November 16, 2026, 06:35:20 am »
I find that if you specify that a hypnotized subject's eyes have tiny pupils, you can often get a nice enhancement to an image if you're using Grok or Bing.
"I am the Master, and you will obey me . . ."

Offline Wendall

  • Member
  • **
  • Posts: 368
  • I am a llama!
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #36 on: November 16, 2026, 09:26:42 am »
I find that if you specify that a hypnotized subject's eyes have tiny pupils, you can often get a nice enhancement to an image if you're using Grok or Bing.
Interesting, I'll have to try that. Grok & Nanobanana both tend to give good results for 'vacant expression', 'spaced out', 'dazed', 'hypnotized', 'mesmerized', 'brainwashed', 'mindless stare' etc ... although I do notice that w/ NB specifically, sometimes phrases like 'mindless stare' combined w/ descriptive sexy language in a prompt causes the censor to kick in (but not always, in fact not even most of the time). I'm actually not sure where the NB censor is at these days - it seems fairly permissive (for PG-13 kinky material). The new approach seems to be, if there's too much horndog language in the prompt, rather than blocking the output it'll just ignore the offending parts of the prompt & give something much milder. So the trick is to walk right up to that line of suggestive language (which can be pretty suggestive!) without crossing over & causing it to revert back to "stock photos on the wall of an HR department" mode

Anyways, other terms I've used - 'glassy eyes' or 'glassy-eyed stare' gets great realistic results in NB, in Grok it tends to produce those magical solid-colour eyes one would typically see in old DA manips. 'spiral eyes' i haven't tried in a while - i remember it giving good results on more cartoony images, but looking quite bad on realistic images (literally just spirals super-imposed over regular eyes). Perhaps it works better on newer models. 'dilated pupils' is another one I haven't tried in a while - I remember it mostly giving results that made the subject look high or really excited, as opposed to brainwashed. Probably would be appropriate in certain contexts.

Offline raphaelsdream

  • Member
  • **
  • Posts: 54
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #37 on: November 17, 2026, 09:33:38 pm »
Is it me, or is Gemini imagegen having serious issues?

Offline Hypno-Canuck

  • Member
  • ****
  • Posts: 2120
  • Gender: Male
  • Those Thrilling Days of Yesteryear
Re: Talking Shop: the thread for technical discussions, tools, and updates
« Reply #38 on: November 18, 2026, 08:01:58 am »
What isn't having issues online this week?
"I am the Master, and you will obey me . . ."

Offline Anostus

  • Member
  • **
  • Posts: 141
  • I am a llama!
I'm curious if people with local LLM setups have opinions on what models are best for MC stories?

I've experimented with various abliterated models like the Heretic versions of Qwen 3.6 35b and Gemma 4 26b a4b (and a few other fine tunes to boot), and enjoyed some of the results, but I'm curious if anyone has a favorite model that always hits the right notes for them?