Model Overview
This LoRA is trained on Boa Hancock's red outfit from the Amazon Lily arc. The dataset is curated using official One Piece anime frames alongside selected images sourced from Pinterest.
Eye Tagging System
To fix poorly-drawn and inconsistent eyes in frames, four distinct, canon-sourced iris designs were extracted from high-detail anime frames and applied across the dataset. 

Instead of using one generic eye type, four different cornea designs were applied to match specific moods and the pre-timeskip style.
nelvoth iris: Gives Boa a beautiful look; best suited for cute, in-love, and normal expressions.
blue eyes: Standard pre-timeskip design with simpler shading.
dark blue eyes, black pupils: Sourced primarily from the One Piece: Stampede movie, best suited for cold, neutral, and slightly angry expressions.
light blue eyes, black pupils: Best suited for shocked, terrified, and intense angry expressions.
Snake Earrings (ss_earrings)
To generate the earrings, use the tag ss_earrings.
Usage Instructions:
When prompting for
ss_earrings, addpearl earringsto your negative prompt. The base model sometimes attempts to generate pearl earrings simultaneously, causing structural deformation.To generate the character without earrings, remove
ss_earringsfrom your positive prompt and placess_earringsin your negative prompt.
The snake earrings in standard anime frames are frequently inconsistent and lack fine detail. To correct this, a clean canon design from high-quality frames was used to redraw and standardize the earrings across the dataset.


Outfit Generation & Tags

Topwear
Trigger Tags:
veskarn crop top,flower pattern on long sleeves,white cape,epaulettes.
Sleeve Details: The
veskarn crop topsleeves feature base snake-scale patterns on the forearms alongside a floral emblem with circular petals. Theflower pattern on long sleevestag specifically controls this floral emblem. If you only want the snake-scale design without the floral pattern, addlong sleevesto your positive prompt and placeflower pattern on long sleevesin your negative prompt, as the model often struggles to render this floral emblem accurately.
Outfit Swapping: The bottom section of the crop top has concept bleed. To generate entirely different topwear, you must add
veskarn crop topandpink belt with golden buckle on crop topto your negative prompt.Chest Pattern Correction: The patterns around the breasts were inconsistent in the original anime frames. The dataset was manually edited to replace them with a consistent, refined design across all images.


Bottomwear
Trigger Tags:
mirsolen sarong with flower patterns,light purple waist sash.Tip: When prompting for the sarong, add
bad flower patterns on sarong,deformed flower patterns on sarongto your negative prompt, as the model often struggles to render these pattern details cleanly.
Outfit Swapping: To generate custom skirts or pants, add
mirsolen sarong with flower patternsandlight purple waist sashto your negative prompt to fully remove the default lower design.
Salome the Snake
To generate Boa Hancock's pet snake, use the base trigger tag Salome the snake. The dataset uses specific sentence-style tags to control Salome's position, expression, and interaction with Hancock.
Positional & Interaction Tags:
Salome is coiled up: Generates Salome resting in a coiled position.Salome is behind Boa Hancock: Places Salome in the background.Boa Hancock is resting on Salome like a throne: Addthroneto the negative prompt.Boa Hancock is standing on Salome the snakeSalome is in dynamic movement: Generates Salome actively moving rather than resting.
Expression & Detail Tags:
Salome has open mouthSalome has closed mouthSalome has clenched teethSalome is angry
The Claw of the Celestial Dragons
Trigger Tags:
red circular tattoo with three triangles at top and one triangle below the circle,back tattoo
Signature Pose: Mero Mero Mellow
Base Pose (No Effects): To generate the hand gesture without any glowing hearts, use the tags
thumbs on top heart hands,outstretched arms,foreshortening.Tip: Add
heartto negative prompt.
Frontal Attack: To generate the hearts shooting directly at the screen, use the tags
thumbs on top heart hands,heart hands releasing multiple glowing pink hearts towards viewer,outstretched arms,foreshortening.Side Attack: To generate the character shooting the hearts sideways, use the tags
thumbs on top heart hands,heart hands releasing multiple glowing pink hearts to the side,outstretched arms,foreshortening.
Looking Down Pose
Trigger Tags:
head back,face hidden,leaning back,pointing down,outstretched arm,hand on own hip,foreshortening.
Expressions & Other Poses
Younger Appearance: Use the tag
aged downto give Boa Hancock a younger look.
Hair Flip: Use the tag
hair flipto generate her running her hand through her hair. This tag works exceptionally well when combined withdynamic pose.
Comedic Angry Face: Use the tags
triangular blank eyes,shaded faceto recreate her specific anime-style angry expression with blank white eyes.
Intense Shadowed Face: Use the tags
shaded face,black shaded eyes.
Wet Look: Use the tags
wet,wet hairto render her drenched hair and skin appearance.
Training Details & Parameters
Base Model: Anima Base v1.0
Dataset Size: 194 images (1 repeat)
Trained Version Checkpoint: Epoch 41 / 3,977 steps
Hardware & Batch Size: 2x GPU (Effective Batch Size 2: 97 steps per epoch)
Training Resolution: 1152 × 1152 (with bucketing enabled)
Network Rank / Alpha: Dim 64 / Alpha 32 (Trained UNet only)
Learning Rate: 0.0002 (Cosine scheduler, AdamW8bit optimizer)
Recommended LoRA Weight: 1.0
📌 Custom LoRA Commissions Open!
Want a specific anime character, outfit, or art style trained? I take custom AI model commissions optimized for Illustrious-XL, SDXL, and Anima base models.
🔗 Order here on Fiverr: https://www.fiverr.com/s/p33zBVZ
Description
FAQ
Comments (8)
Probably the best Boa lora ever. So many details and the looking down pose is perfect
what prompts did you use for the purple dresses?
Initially, I tried training a multi-outfit LoRA (4 cannon outfits), but the logos and patterns kept bleeding into each other. To fix this, I made it a single-outfit model. The purple dress images were only included as close-ups/upper-body shots to train specific facial expressions and dynamic poses, not the full outfit itself. In fact, for the looking-down pose, I edited the purple dress into a plain white dress so the model would learn the anatomy rather than the clothing logos.
I'll create the LoRA for the 3D2Y movie, Marineford Arc, and Stampede movie outfits later, since I already have the dataset ready for them.
beautiful lora, exceptional, just one thing, what is the input for her purple outfit?
Initially, I tried training a multi-outfit LoRA (4 cannon outfits), but the logos and patterns kept bleeding into each other. To fix this, I made it a single-outfit model. The purple dress images were only included as close-ups/upper-body shots to train specific facial expressions and dynamic poses, not the full outfit itself. In fact, for the looking-down pose, I edited the purple dress into a plain white dress so the model would learn the anatomy rather than the clothing logos.
I'll create the LoRA for the 3D2Y movie, Marineford Arc, and Stampede movie outfits later, since I already have the dataset ready for them.
@chitra Perfect, thank you so much <3 your loras are just amazing



















