Gabrella: Update 15

Published by

atcDave

January 18, 2026

*My team setting out for “The Wailing Dwarf”.*

Our expedition was to learn the fate Amnish Captain Bryam Lancameth. His widow has wanted an answer for many years. The ancient Dwarven Kingdom in the Troll Hills has been over run by, wait for it, trolls. And other things. We will see what can be found.

*“If you boys were smart you know you’d just move along…”*

Eh, goblins are rarely that smart.

*“What kind of troll am I? I’m more of a troll antidote actually.”*

The red wizard was the first of several we met. They were doing experiments and cross-breeding trolls. We faced a bewildering variety of types (sub-types? Cross breeds? Whatever). They can conveniently be classified now as “ex-trolls”.

We explored deeply into the ancient Dwarven kingdom. Apart from the many trolls we faced orcs and giants.

*We also talked with the long dead members of Lancameth’s expedition.*

They all begged for us to find the traitor who had led them all to their end. And as we went deeper we began encountering dwarves. Those dark and twisted magic-using Derro Dwarves.

*Oh we found the traitor. He was one of the Derro, who had accompanied Lancameth’s trip in disguise.*

Defeating him brought peace to those we had sought, although Lancameth himself had been infected with a sort of undeath. With the help of the Church of Helm we were able to bring peace for him too.

Since we were back in town, it was time to check out our friend Bodhi’s old stomping ground.

*Even with my smaller size, overbearing sometimes works.*

*“Ah honey, who taught you such a lunatic fighting style?!”*

Imoen reported she felt whole and at peace after we slew Bodhi.

*****

This was done entirely with the new SeeDream 4.5. Mostly an excellent renderer with a few quirks.
I started by doing new “photo-realistic” renders with SeeDream of all the characters on my team. My first attempt at putting them in one portrait showed them all looking way too much alike. It was consistent on the *details* I mentioned, but the basic facial model seemed to be the same on all of them. So I reconsidered, and did new generic character renders with both DALLE and MAI-Image. This got a little more variety in characters and facial types. This helped, although I still find when you mash them all together with SeeDream it forces some conformity. Looking at the portraits I posted, well let’s just say Gabby and Imoen sure look like sisters! And Skie may be a cousin! Aerie has a slightly different face.
It also wanted to give Keldorn and Bull the same facial hair. I found I could differentiate by specifying “the bald one”. Funny.
Then, my bad, the first portrait, “heading out” has Keldorn wearing the new Mithral Plate he got on the quest. Oops.

I think I’ll still have plenty of use for the Co-Pilot renderers. For one, I’ve scaled back on points at OpenArt since my Chuck and Sarah run. And now I’m using it when I do video renders. SeeDream is also slightly more expensive than the other renderers. So it will have a place, but it will not be exclusive in the future as it is in this update.

4 responses to “Gabrella: Update 15”

Zeno

January 18, 2026

Its definitely a whole new world being able to work with pre-defined “actors” like this. And it seems to do a pretty good job interpolating the parts of the character it can’t see in the reference image. There’s definitely some cool shots there that would have been difficult or impossible with DALL-E. The group shot is great! The “after Bohdi” shot would have taken 100 or more takes. The derro under the dwarf statue looks great too!

I’ve been playing around with it a bit too. I can’t quite get the same dynamic motion out of it that I can out of DALL-E. And add too many characters together and it sometimes starts mashing them together. It also seems to have trouble with relative heights if they don’t match expectations. The old issue with male halfling standing next to female human. But it can definitely pull some things off (like the group shots) that would be utterly impossible in DALL-E. So I think it will end up being a tool in my kit for composing complex static scenes.

I tried doing a fight scene between Minsc and a Sahuagin with it, and while it could pose them both okay – and add extras in the background – the movement was never quite as smooth as I liked. And it had a disturbing tendency to turn the Sahuagin into an actual fish….

At best I got this:

https://openart.ai/share/MndSTnR3enAwQWdoZ3Z4NjU0SFc7aHR0cHM6Ly9jZG4ub3BlbmFydC5haS91cGxvYWRzL2ltYWdlXzN5TUFod3NoXzE3Njg3NDU1NzkwNTdfNTEyLndlYnA

At worst I got this. Yikes!

https://openart.ai/share/c3NsdE42a1F5NURSRExzdHlVR3g7aHR0cHM6Ly9jZG4ub3BlbmFydC5haS91cGxvYWRzL2ltYWdlX0t5REpaV3VlXzE3Njg3NDU0MDExODBfNTEyLndlYnA

LikeLike

Reply
1. atcDave
  
  January 18, 2026
  
  Oh man, that is one angry looking fish!
  
  I think the best part of SeeDream is working with multiple characters. Like yeah, that one with Imoen over Bodhi in the background only took a couple tries. Of course four renders from each, a little tweaking of the description and I was happy.
  I think you could do Minsc and the sahuagin if you did a stand alone render of the sahuagin in DALLE, then feed a Minsc and a sahuagin render to SeeDream and you can pose them easily until you like the scene. I think you could get something good much faster than in DALLE alone.
  Although as you suggest, it doesn’t seem as good at explosive/dynamic/frenetic action. Gabby jumping on Bodhi was the best I saw, but it lacks the blur or whatever that suggests energy and speed. Maybe, I could have played with it more and made that work better? But I’d already used some “dynamic” words and that was what I got.
  
  I was very pleased with how easily it handled the groups. It does mix up some details, and it did make the women look very similar. But when it wanted to give Bull and Keldorn the same goatee, it was easy to fix with a single line “bald man has full beard (brown with grey)”.
  
  so I will use it some more! It’s very exciting in many ways.
  
  LikeLike
  
  Reply
  1. Zeno
    
    January 18, 2026
    
    I actually did exactly that – handed it standalone DALLE-3 renders of Minsc and the Sahuagin and described the action with (character 1) and (character 2). I used exactly the same reference images for both renders. When you click on the link you can see the two reference images off to the side. It just takes it in weird directions sometimes.
    
    LikeLiked by 1 person
  2. atcDave
    
    January 18, 2026
    
    Oh yeah, I see that now. Not an expected behavior! Apparently it doesn’t know sahuagin.
    
    LikeLike

DavesGaming