Liara T’Soni - Krea 2 Character LoRA
This character LoRA was created for Krea 2 and focuses on reproducing Liara T’Soni’s recognizable appearance, including her facial identity, blue skin, expressive eyes, natural Asari cranial crests, athletic proportions, and iconic science-fiction aesthetic.
The LoRA performs especially well across different poses, actions, outfits, environments, camera distances, and visual styles.
What This LoRA Can Do
Recognizable Liara T’Soni facial identity
Natural Asari head shape and cranial crests
Detailed blue skin with visible texture and tonal variation
Portraits, close-ups, full-body images, and environmental compositions
Sitting, walking, running, kneeling, swimming, and action poses
White-and-blue uniforms and futuristic armor
Casual, modern, formal, and experimental clothing
Spaceships, laboratories, alien planets, cities, and everyday environments
Cinematic science-fiction scenes
Illustrations, anime, paintings, and stylized interpretations
Convincing photorealistic and live-action results
The LoRA also responds well to asymmetrical poses and compositions. Liara can be positioned away from the center, interacting naturally with objects and environments instead of simply standing in a frontal character pose.
Activation Token
The primary activation token is:
Liara T'SoniPlace it near the beginning of the prompt.
Activation Sentence
No dedicated activation sentence is required.
A reliable prompt starter is:
Liara T'Soni, an adult blue-skinned asari womanThe word asari is recommended when describing her species.
Basic Prompt Example
Liara T'Soni, an adult blue-skinned asari woman standing beside a holographic navigation table, white-and-blue fitted uniform, focused expression, asymmetrical composition, cinematic spaceship lightingAfter the activation token, you can freely specify her clothing, pose, activity, environment, expression, camera perspective, and visual style.
Photorealistic Prompting
The LoRA can produce surprisingly convincing realistic interpretations of Liara. The strongest results use photographic language, everyday materials, natural poses, detailed skin descriptions, and multiple believable light sources.
Recommended realism starter:
Liara T'Soni, documentary photograph of a real living adult asari woman, biologically natural blue skin, fine pores, subtle pigment variation, realistic cranial crests, natural anatomy, available light, full-frame photographyFor wet or highly tactile skin, try:
naturally textured blue skin with fine pores, subtle translucency, a slightly slick hydrated surface, irregular water droplets and narrow specular highlightsUsing two or more light sources can make the blue skin appear much more physical and believable. Useful combinations include:
Warm sunlight with cool environmental bounce
Amber streetlight with cyan storefront light
Cool window light with warm tungsten interior light
Direct sunlight with turquoise reflections from water
Warm practical lamps with a cool rim light
The skin reacts especially well to rain, mist, swimming-pool reflections, humid environments, and mixed-color lighting.
Suggested Subjects
Liara works well as:
A scientist or researcher
A spaceship officer
An explorer or astronaut
A diplomat
A casual everyday character
A futuristic investigator
A swimmer or underwater explorer
A fashion or editorial subject
A cinematic action character
Tested Settings
Base model: Krea 2
Aspect ratio: 1:1
Resolution: approximately 2 megapixels
Activation token:
Liara T'SoniActivation sentence: none required
LoRA strength: adjust according to the desired balance between character accuracy and creative freedom
The attached examples demonstrate the LoRA across different clothing, poses, environments, styles, lighting conditions, and realism levels.
Have fun experimenting with Liara in completely new settings.
Description
FAQ
Comments (5)
HELL YEAH. My queen.
I quess its now our queen xD. Have fun with it. Its planed do increase the dataset by 4x but for that I will need way more time. So fare the lora is super and flexibel
@Astroburner for Krea2 I've found 20-30 images is the best with almost no captions depending on the character/dataset. Like any of my realistic characters I literally only use a trigger word.
If you have a lot of variation in outfits, you dont need to caption them. But for a cartoon character with one or two repeating outfit I caption a little bit more but very minimal like=
trigger, cartoon character, outfit
I've found without something like 'cartoon character' they get really stuck as a cartoon character lol.
I've tried larger datasets or more detailed captions, and almost everything is better with only a trigger word.
Even my Shion LoRA which I'm about to upload for Krea2 was only 20 images and just a trigger word. I was really surprised at how well it worked. You have to be detailed when you prompt with it after you trained it, but the training part has become super simple.
Thank you!
Would you do Jack and Miranda for Krea2 too?
I could do it, will need some time to collect the dataset. I think i could finish it in 1 or 2 weeks , Miranda and Jack are lovely xD


















