NOTE: Model was brained on base SDXL1.0, I recommend Juggernaut RunDiffusion v9 (not X), as it most closely replicates SDXL 1.0 data. Other models will work but characters' likenesses will diverge.
I was inspired upon using Dall-E3, seeing how well it recreated the look of the classic show Star Trek The Next Generation, but being disappointed that it could only recreate Picard.
This lora includes changes to the text encoder, and the training data was captioned by hand by me, so you'll be able to invoke different characters by name, different sections, different clothing, etc. I hope to demonstrate this in images, but here are some of the key terms:
Main keywords: Star Trek TNG, TV still
Characters:
Picard
Riker
Data
Geordi
Worf
Crusher (Beverly, I left out Wil)
Troi
Romulan
Klingon
Borg
(there are some more, leaving them a secret for fun (to whom?))
Clothing:
(red/yellow/blue) Starfleet uniform (other colors can be invoked if you try to varying degrees of success)
Klingon armor
Romulan uniform
(and some more)
Objects:
Geordi wears a 'visor', Worf has a 'metal sash', Borgs may have a 'holographic eyepiece'. I tried to train the lora on many different things.
Environments:
'Enterprise' - generically included for on the ship
'In a Klingon building'
'crew quarters', 'main engineering', 'command center' (instead of the bridge, due to bridges being objects), 'sitting in the captain's chair' etc can be invoked to a degree
I included terms like 'atmospheric', lighting colors, 'dark', viewing angles, etc.
I called all the interfaces on the ship 'control panels', the screen on the ship is the 'viewscreen', you can try to invoke different ship exteriors (to little success)
Prompt 'solo' and negative 'multiple people' if you get more than one person and you don't want that. Future versions will make this even stronger.
Description
New training methodology
Slightly augmented training data
New prompting style
FAQ
Comments (14)
how do I reference the lora? The trigger words don't really work.
i downloaded this LoRA and now my "Generate" button says "Engage"
I appreciate all the work you put into this. It's awesome!
Brilliant, thank you so much! Can't wait for a DS9 one!
This works shockingly well.
I am still using this and it's still amazing
oh and one of the "secret names" has to be tasha yar fyi, because i can see it in faces when i don't prompt a specific character. xD
the more I try to make Star Trek stuff across various checkpoints and other LoRAs, the more I realize how amazingly well done this LoRA is. I hope you continue to work on it. Whatever training you did was perfectly executed.
However, it doesn't know what a Vulcan is or how to do Vulcan ears. Which I guess makes sense, since Vulcans had limited participation in TNG.
Does this work with SD 1.5? It says XL in the title but I don't see an XL tag on it. Was it trained on XL but works on 1.5?
Great Lora but any idea how to avoid clones of the characters?
A few (not so secret) keywords:
('space', 51)
('profile', 51)
('facing away', 52)
('transporter room', 58)
('crusher', 59)
('main engineering', 62)
('riker', 68)
("captain's quarters", 71)
('romulan uniform', 76)
('miles', 79)
('a klingon building', 79)
('talking', 83)
('ready room', 95)
('wearing klingon armor', 95)
('metal sash', 97)
('crew quarters', 99)
('visor', 105)
('geordi', 114)
('worf', 116)
('picard', 140)
('data', 144)
('dark', 157)
('sitting', 167)
('red starfleet uniform', 209)
('standing', 252)
('command center', 287)
('yellow starfleet uniform', 410)
('enterprise', 1111)
('star trek tng', 1508)
('tv still', 1509)
Fantastic lora, thank you
any chance at a pony version or sharing your dataset?
Pretty amazing for the most part, but I think I'm doing something wrong. Crusher and Troi look like Crusher and Troi, but about thirty years older. Why do I keep getting old lady versions?