Make your characters actually HOLD their weapons! ⚔️
Ever generated a beautiful fantasy warrior, only to discover that her sword is floating beside her hand, her fingers are fused into the handle, or her dagger is somehow passing straight through her palm?
Object Logic – Fantasy Weapons was created to address exactly these problems.
This concept LoRA focuses on improving hand-to-weapon interaction, natural gripping, and spatial consistency in FLUX.1-dev generations.
Rather than teaching a particular combat pose or visual style, it helps characters interact more convincingly with the weapons they carry.
What does it improve?
Natural weapon grips: More believable finger placement and hand positions around handles, hilts, and shafts.
Hand-to-object contact: Reduces floating weapons, awkward connections, and objects passing through fingers or palms.
One-handed and two-handed grips: Supports different ways of holding and carrying weapons.
Spatial consistency: Helps weapons sit more naturally in relation to the character's hands, arms, and body.
Pose flexibility: Works with different stances and compositions without enforcing a particular combat pose.
Supported Weapons
Trained on a diverse dataset featuring different fantasy weapons and gripping styles, including:
Swords, greatswords, and daggers
Battle axes and other hafted weapons
Spears and polearms
Magical staffs and wands
Bows
Shields and weapon combinations
The LoRA is designed around the shared principles of gripping and holding weapons rather than memorizing a single weapon type or pose.
Example Prompt
objectlogic_weapons, an adult female fantasy warrior standing in an ancient castle courtyard, wearing fitted dark leather armor and tall boots, holding a long steel sword in her right hand with her fingers wrapped naturally around the leather-covered hilt, the blade pointing diagonally downward beside her body, confident expression, lips closed, dramatic late afternoon lighting.
Limitations
This is not a weapon design LoRA. It focuses primarily on how characters interact with weapons, rather than guaranteeing historically accurate weapon shapes or proportions.
Bowstrings and complex weapon mechanisms may still produce inconsistent results. Crossbows were not included in the training dataset and are currently not reliably supported.
The LoRA improves the likelihood of realistic interactions, but does not guarantee perfect anatomy in every generation (we're still on flux...).
Showcase
The first 8 images show only base flux with the LoRA. The next 10 images show what it can do with some of my other LoRAs.
Would love if you post your creations here with it.
The LoRA was trained on the default flux1-dev model for maximum compatibility.
Dataset and training data:
90 images
7,5 steps per image
10 epochs
bucks with 1152px max size, no auto scaling
cosine scheduler with 0.2 warmup and 0.8 decay
Captioning was done with OpenAI
alpha == dim: 32
If you use it with other LoRAs, for example a character LoRA, try to lower the weight — I had good results at around 0.4.
As always: I did not polish any of the images in the showcase, they all came out like this. I did 4 images per prompt and picked the better looking one.







