We got down to The Severed Hand the next day. It is an imposing sight!
Its a little daunting to think of something so big, being abandoned for so long. It is rumored that the final combat here between Elves and Orcs is replayed endlessly by the non-survivors.
We were barely in the doors when we were first attacked. But there was more in every room of every floor it seemed. Goblins, Orcs, Ogres and even the elves.Pikwell kept busy. There were plenty of traps, and he’s become an expert on the subject.The great rooms still had great meetings in session. And they would all turn against us.Not all the specters were hostile. I met Sehriya. I have a little one at home about this age. She wanted to play hide and seek, I think a specter may have an unfair advantage…One of the towers had something a little more solid to deal with.An elven priest asked for our help in restoring his sanctuary.We finally located the elven arch-mage Larrel. Initially he was insane, but he was obsessed with his damaged astrolabe. We helped restore it, and he recovered his wits.
Honestly, Charity was the only one of us knowledgeable about such things. But she made it happen. Now we knew we were heading to a place called “Dorn’s Deep” as the source of evil in the region. Larrel offered to teleport us there, but we asked to return to Kuldahar to resupply. He obliged.
We had another detour! A friendly beast, of a type we weren’t familiar with, asked for our help with an undead relative of his. This was not difficult and only cost us a day. Apparently his type is often regarded as evil and dangerous, his relative had become a lich of sorts. So we saw this as important detour.
The Severed Hand is probably my favorite chapter of IWD. Pretty much every time, but especially when I have someone like Tessa along who really feels the loss and tragedy.
My favorites here are Pikwell (as always) and both of the ones with Ila. That first spectre is awesome, and the setting there is great as well. And I always have a soft spot for the little ghost girl. The detail on Pikwell – particular things like the way his hands look as he is pointing – really sells his look.
I like your Astrolabe. That was a pain to get a good render of. I think I went with “brass Orrery” or the like. Can’t remember.
I do like this part a lot. I think Dragoneye is my favorite, but Severed Hand isn’t too far behind, and for a lot of the same reasons. They’re right at the sweet spot for levels, and they’re intensely atmospheric.
I really liked Ila with the little girl. I figure she’s probably the most homesick of the crew, add in the uncertainty of ever getting back, and I think the little orphaned spirit would hit her hard.
it is amazing to me how well Pikwell has worked for this! But he seems to be pretty easy for the AI to get right, and he adds just the right amount of silly.
The astrolabe was harder to get than I expected! I had to keep saying “bigger”! Finally “fills the whole room” and even that could have been bigger. But yeah, I’m pretty happy with this.
BTW, I should mention, I’ve been playing around more with stuff for my coming BG run. And a few more observations.
I cannot do a “character from the show Chuck” prompt on Co-Pilot. Its blocked as a violation. Difference in US vs Australian law maybe? I am still thinking I may use Co-Pilot for a lot of action scenes, and almost any multi-character scene. It is far faster and easier for these things. Also, I think I prefer Co-Pilot’s interpretation of fantasy looks. Even going with the “photographic” type style, Co-Pilot seems to do a better job making the photograph “fantasy”. That all means I’ll have to describe the main characters each time, and maybe hide/obscure faces as best I can? Or blurred action type scenes? I don’t know, even the couple renders you did were not convincing likenesses. Maybe I’ll crop the heads out… Or just full armor (with helmet) could work for Sarah. So I’m looking at Co-Pilot for a lot of the look, but not as much specifically for the main characters.
OpenArt does staggering high quality portraits, especially with the models. But I cannot get them into action scenes as easily. It really wants to fight on that. It does allow for fusing and editing renders. I think I can’t play with this much until Saturday due to no currently working models. But I think that means I’ll be able to pose my main characters together and maybe have them interact to some degree. BUT, this all looks complicated. Obviously its taken time to get more proficient with Co-Pilot, the same may happen with OpenArt. But I may not use it for much beyond posed portraits. A further issue is its clean vivid realism. That looks great for human characters with models. But Ogres and Orcs look so alarmingly clean they look plastic. I’ve played a little with better adjectives to fix this, that may require more work to get a mood and style I want. In some ways OpenArt seems less willing to create and is more of a slave to the typed description. This is a mixed blessing; its easier to command, but can’t be left to do very much on its own. I’d also mention with the models, of course none of the photos I have of the characters/actors are dressed and equipped as I want for a BG run. So time is spent describing attire every single time. And OpenArt seems to know even less about armor and weapons than Co-Pilot does. Seriously, I’ve had Sarah in armored high-heel boots. And a lot of the training photos I used were things like the actress doing an exercise video, lot’s of great action and movement. But OpenArt interprets that with showing a lot of leg, even in plate armor. Of course Co-Pilot sometimes does that too, especially when I describe skin color.
It will probably be another two or three weeks before this starts, I’ll be rebuilding the games again too (I’m using SCS now, and have found a few more quest mods I want). But surely before Christmas I’ll have something to show.
Note that it is extremely picky about the exact wording when doing characters like this. For me “woman who looks like Sarah from the show Chuck”, was blocked, but “Sarah from Chuck” sailed through. So you may need to avoid saying things like “movie” or “show” or anything that points out to the text filter that you are trying to reference other media.
I’ve heard a lot of folks use CoPilot to generate some of their training imagery, and then feed that in to their model generation for later consistency.
I think fundamentally what I like about CoPilot is the randomness/creativity. I can feed it a concept, get back a bunch of variations, and then pick out either the one most like what I asked for or simply the one I liked the best. I wasn’t able to do that with many of the OpenArt models. And while, as you say, I could get great portraits – I found it terrible at monsters and action.
I’m still going to experiment more. What I would *really* like is to be able to train “Actors” and then combine them in scenes. E.g. train a “Keira” model, a “Kord” model, and a “Yuan-ti” model, and then be able to say “Keira dodges a Yuan-ti swinging a scimitar while Kord smacks the Yuan-ti over the head” and such. But that looks like it would require diving into the world of compiling and running my own local AI or signing up for an expensive pay service. And I don’t like the latter idea very much because I haven’t seen one that let me play with it enough to see if it suited me before paying.
But yea. I think its time for me to branch out a bit as well and do some pages about experimenting with other services.
Okay, I may try again with the wording on Co-Pilot, it certainly seems like a good shortcut for getting some better action renders.
My experience so far with OpenArt would confirm everything you just said. Even to say, I’ve done exactly that with CoPilot for helping with training pictures. I was particularly concerned for the character Chuck, 50% of his show costuming is a uniform I do not want! So I played around to get his adventuring look for BG in CoPilot and then fed those into OpenArt. Sarah is a little easier, not only did she wear more variety on the show, but the actress did a fair amount of modeling. So a ton of high quality source material for her. Although I still used CoPilot to render some fitting armored views and fed those to OpenArt. Funny that the AI still decided to show her legs so often.
Anyway, everything you’ve said affirms my thought that I’ll want to continue using both.
BTW, I still just get blocks even with it phrased that way. So I think I’ll have to describe the characters for most scenes. I’m used to it now, and I’m expecting I’ll want to obscure their faces somewhat anyway. But it will be worth it for the better fantasy look of Co-Pilot.
That’s bizarre. I wonder if its also because I did it on my pay account instead of one of my free ones? I have notice there is a difference there. And I often actually prefer the results I get with the free ones! But maybe it is as you say, and it because I’m in Australia. Who knows?
Leave a reply to atcDave Cancel reply