Ch3: Martha, she’s great. here at Danklabs our new assistant is over performing. Swamp exploration for rare DNA, dark rituals. Man she’s got it all, even wrestled that thing that ate the last assistant… anyway.

Chapter 3 was made with 5 Lora’s, a thousand images each consisting of dark and moody landscapes, people, races, monsters and a dash of pose regularization. The result is, frankly, my favorite so far. read below for some prompting tips.

Ch2: Suppose I should write something substantial here now that these breakouts are becoming a regular thing. Yeah, it happened again, I'm running out of lab assistants, test chambers, and there's a huge hole in the wall but what can I say, I'm happy with the latest creation to escape Danklabs. So......

Ch1: There’s been a containment breach here at Danklabs. Been fighting with it all night but the little bastard got out anyway. It’s your problem now I’m afraid,,, I’m off to bed the big Intro will have to wait. GLHF.

MERGE WARNING!

Model trained on deliberately falsely tagged data set to blend styles.

Chapter 2 has a re-worked data set that includes more images styled in 3d, photographic and film still to push it towards a more realish uncannyishness. As ever the tagging is far from perfect using a combination of machine tagging and shallow-at-best adjustments carried out by my latest late lab assistant,,, think that thing swallowed her, good God. So merge with care if you choose to do so.

To Tame the Beast:

• CFG: 4 to 8 (I like 5)

• Step: 20 and beyond. (42 is the best... obviously)

• Dpmpp_3m_sde, Karras (just my fav, go wild)

PROMPTING:

Using tokens like…

(low quality, candid, low light, dark and moody, grainy, analog photograph, haunting, dark and moody atmosphere, natural dark lighting, low light:1.?)

Ch3 recommended tags: muted colors, dynamic angle,

… lean into the changes in the dataset. wouldn’t go as far as calling then trigger words but… give ‘em a whirl.