There's a few more polls than usual this time around, mainly to make up for the big gap between releases and to give me enough data to work with in case the next one takes a while too. Some are related to the model, while others are more general questions about AI as a broader topic. As always I appreciate the feedback and support, and I'll try to respond to everything in due time, so check back a few days after leaving a reply.
(Last comment - 17/02/26)
My LoRAs:
Poll 1 - V7.0's performance
While I'd advise you to base your answer on the things you typically use it for rather than the entire model, you can see the current full showcase here, and V6.1's here for reference.
Pretty much the same ratio here as last time, which is good. This is just the obligatory before/after question, so I don't have much to say here, besides "thanks".
Poll 2 - Next model
This is mostly to guage interest as I'll probably do a smaller model or two before getting back to the AIO regardless of the results, because they tend to inform a fair number of changes based on what I learn from their performance. I've been sitting on an update for my slime-carrier model for a while too, which is nearly ready, so whatever's picked here will likely follow that.
Nice to see a fairly decent spread of results here, I'll look into the manticore/jabberwock model again when I get the chance. As I mentioned above, I've still got a rough order I'd like to do things in, but it's good to know there's a fair spread of support for various models rather than just my main one.
Poll 2.1 - Specific models
Refer to the MGE wiki for info if these categories are unclear. I've tried to pick all the ones I think I could make meaningful improvements for, but feel free to give a more specific example in the "other" box if none of these interest you, or you have a more specific suggestion for one of the listed options.
A surprising amount of interest in the "tentacle" model, more than I was expecting. I've got a couple of ideas for most of the popular ones so I'll probably give them all a shot at some point, I might even be able to squeeze the sandworm into the aformentioned tentacle one if I can find enough things to cross-reference, no promises there though.
Of these, I reckon slime-girls would probably be the easiest to do, considering both their current mostly-working state in the AIO, and the fact I've already got another slime-themed lora to refer to. Lamias should also be pretty easy, and I can probably get away with using a lot of non-explicitly-MGE-related images in the tentacle one. Plant-girls work reasonably well already, so I'd imagine they'll be fairly straightforward too, while the wight/wendigo will probably be the hardest to get working in a meaningful way, wherein the extra hands have good functionality.
On a somewhat related note, I've been slowly teaching myself how to draw for the sake of getting more good references. Obviously it'll depend on how well I'm able to do so in each case, as I'm essentially a beginner, but I'm fairly happy with my current attempts.
Poll 3 - Model usage
I have a feeling a lot of the posts under my model are either from bots, or from people using it unknowingly after copy/pasting other generations, as it's rare to see an actual monster girl down there. Select all that applies.
This one's interesting because I only see only see a couple of people going through the list, but it does make sense, in that you're probably just testing things versus trying to generate post-able stuff. In regards to wildcards, I've had a couple of people mention them with previous releases, I could look into an "official" one if there's enough demand. It'll be a while away though, given the other models I intend to work on, but if you're interested then keep an eye out for any future comments under my model, as that's usually the best place for me to ask single questions for feedback.
Poll 4 - Taglist Detail
While I probably won't make any retroactive changes based on this, it'd be useful to know what people prefer for future models. V7.0's taglist is closest to option two, and the prior ones are closest to three for reference, in case the questions are cut off.
Decent spread here again, good to see the current one is the most favoured though. I could try doing a table or something next time, with a section for each of the options here, but I'll refrain from promising anything as the taglist is already one of the more time-consuming things to deal with as-is, let alone with four or five times the amount of text.
Poll 5 - Other base models
This one's just to guage interest. If there's a lot of support for one then I'll look into it, but I'm still restricted to the options Civitai has available on their trainer.
This is about what I expected, apologies to the remaining Pony fans but I think I'm gonna stop making loras for it, unless they release some version that's on par with the newer models' performance.
I'll look into Z image, I've yet to use it but I've heard good things. I believe it's a caption-based model, as opposed to tags, so based on the effort that'd require I think it might be better to do it with V7.1 rather than now. Correct me if I'm wrong though, I haven't done much research yet.
General Questions:
Poll 6 - AI Monetisation
This is a personal topic I'm interested in for a bunch of reasons, but essentially I'd like something to base my model-usage stance on other than my own opinion. The current one is mainly in place to avoid association with low-effort and bad-actor-made content, but I'm aware it's quite pedantic, and I'm willing to make changes if I get some good feedback here. I'll save the rest of my thoughts for my analysis/response to avoid skewing the results.
As mentioned I had a couple of reasons for asking this one, great to see a range of opinions. A consensus would probably be easier to work with, as I could just update the policy and be done with it, but I think the potential for discussion is more interesting, and I'll probably ask some more specific questions related to it in future polls.
Aside from the association reason stated above, I've historically been in the middle-negative group of opinions, hence the current stance. There does seem to be slightly more favour than there are negative opinions which, although expected on a site specifically for AI-related things, will probably influence any changes I make.
Another reason is that, frankly, it's become quite hard to justify the time and effort required to make my models for free. To be clear, I'd never consider paywalling content, but I still thought it'd be a good idea to test the waters for something like a tip jar with more regular progress updates, or perhaps showing off the self-drawn images I mentioned below Poll 2.1. Obviously it'd be hypocritical to bar monetisation only to then seek it myself, and I'm still of the opinion that such practices are the main reason for the large negative sentiment around AI online, but I'm more focused on the quality of my releases than spreading my beliefs so I'm willing to look into something of the sort.
Poll 7 - General AI Usage
This is mostly for curiosity's sake, as I doubt I'll be moving away from LoRAs any time soon.
This one's also pretty-much what I expected, although audio being above video was somewhat surprising, given recent developments. I'm still bound to both Civitai's model-trainer and the "monster girl" subject matter, so don't expect to see me branching out into other technologies any time soon.
Poll 8 - Additional Comments
As with last time, add anything else you'd like to say here and I'll respond in due time.
1 - N/A - Thanks, and yeah it's in the queue. I'll try to aim for less than six months this time lol.
2 - Tentacle control - Glad you like it. There seems to be a lot of favour towards the tentacle model already in the other poll, so I'd assume it's a shared sentiment. For positions, I'm currently thinking of doing "tentacle on/in [thing]", "[thing]-coil" and a few more vague ones like "tentacle-hug" for spacing and such, all of which I've gotten working in my WIP slime-carrier model with varied levels of consistency. Performance will mostly depend on how many clear examples I can find, particularly with things like the night gaunt that have ambiguous or varied feature designs (the tail, in its case).
3 - More options - This'll just improve naturally as I find better ways to train certain details, particularly whenever I break existing vague tags into more specific stuff. I'm currently focused on clothing as it's usually a good middle-ground between size and consistency, and there's a finite number of ways you can draw a skirt or jacket so there's lots of similar-enough references. As mentioned elsewhere, I should also hopefully get some useful insights from the dedicated models I make between my main ones, so expect a lot of the improvements to follow the options in the other poll above. I've been trying to get things like the "tentacle-brush" seen on the tentacle-girl and manticore, and the "extra arms" of the wight and wendigo working for a while, and I've been looking to allow for finer control over wings, tails and such, just haven't had much luck with my current dataset.