Baldur’s Gate 3 Party Pack | MiniMax H3 LoRA

Version 0.5 Beta — FREE

This is an early 0.5 Beta release of my Baldur’s Gate 3 multi-character LoRA for MiniMax H3.

The LoRA is still unfinished and there are currently plenty of inconsistencies and generation errors, especially in more complex multi-character scenes. Because of that, I’m releasing this beta version completely free while I continue improving the dataset and training.

The LoRA currently supports four characters:

• Karlach

• Lae’zel

• Minthara

• Shadowheart

Activation Tags

KarlachV01MH3
LaezelV01MH3
MintharaV01MH3
ShadowheartV01MH3

There is no single global trigger sentence. Use the activation tag of the character you want directly inside your prompt.

For multi-character scenes, simply use multiple activation tags in the same prompt.

Prompting Tips

For best results, describe the scene in detail after the activation tag.

Useful things to describe:

• clothing / armor

• pose and body position

• facial expression

• gaze direction

• hand position

• action

• camera angle

• shot type

• environment

• lighting

• cinematic or realistic style

For group scenes, describe each character separately and clearly state what they are doing and how they interact.

Shadowheart

Shadowheart was trained with both her black-hair and white-hair appearances.

Specify the version you want directly in the prompt, for example:

ShadowheartV01MH3, white hair tied back...

or

ShadowheartV01MH3, black hair...

Example — Single Character

KarlachV01MH3, wearing fitted dark leather battle armor, standing in a rocky mountain clearing, relaxed confident expression, looking toward the camera, medium full-body shot, eye-level camera, warm golden-hour sunlight, cinematic realistic fantasy style

Example — Multiple Characters

KarlachV01MH3, LaezelV01MH3, MintharaV01MH3, ShadowheartV01MH3, the four characters standing together in a ruined fortress courtyard, each wearing distinct fantasy adventuring clothing, interacting naturally with each other, cinematic realistic fantasy style, dramatic dusk lighting, medium-wide group shot, detailed environment

The LoRA is mainly intended for MiniMax H3 Text-to-Video and can be used for:

• cinematic character showcases

• realistic fantasy interpretations

• party interactions

• dialogue scenes

• combat

• spellcasting

• group scenes

• gameplay-inspired sequences

Beta Notice

This is Version 0.5 Beta.

Expect errors.

Character identity can still drift, complex interactions can fail, hands and anatomy are not always perfect, and multi-character scenes are especially demanding for the model.

This version is being released for free so people can experiment with it, test different prompts, and see what MiniMax H3 can already do with the four characters.

More training and improvements are planned for future versions.

Have fun experimenting with it! 🚀