Trigger: metallic gold AlessaDress
Model Type: WAN 2.1 14B T2V LoRA (fully compatible with WAN 2.2 and I2V)
The Alessa Dress LoRA delivers a high-glamour metallic aesthetic with reflective gold panels and elegant, form-fitting design. It was built from the ground up to capture the look of modern performance and festival couture — equal parts fashion, fantasy, and stage-ready brilliance.
While trained for WAN 2.1 14B T2V, it performs flawlessly in WAN 2.2 and Image-to-Video (I2V) workflows.
🔧 Recommended Settings
High Strength:
0.6(or try 0.5 for subtle results)Low Strength:
0.8(or try 0.7 for smoother blending)
These ranges maintain rich metallic detail without overpowering the scene.
🎨 Prompting Tips
For general use:
wearing an AlessaDressFor exact control:
wearing a metallic [color] AlessaDress
→ works with gold, silver, red, green, blue, etc.Add or modify sleeves for variety:
sleeveless,short sleeve,long sleeve, etc.
(Be consistent across prompts for best results.)Lower LoRA strength to repurpose outfit details into new designs such as swimsuits, lingerie, or bodysuits.
🧠 Training Details
Base Dataset: A small curated set of real-world images of a metallic festival outfit (head cropped, backgrounds removed).
Data Expansion: Additional reference images were generated with Google’s Nano Banana, guided by a structured routine covering multiple angles, poses, and perspectives for diversity.
Image Cleanup: Every dataset image was refined in Photoshop to remove visual aberrations and artifacts.
Captioning: Captioned using Qwen2.5-VL-7B-Instruct in BooruDatasetTagManager, with a custom QUERY template ensuring structured, detailed descriptions.
Training: Trained in AI-Toolkit on RunPod for 3000 steps, optimized for visual clarity and material fidelity.
Description
Trigger: metallic gold AlessaDress
Model Type: WAN 2.1 14B T2V LoRA (fully compatible with WAN 2.2 and I2V)
The Alessa Dress LoRA delivers a high-glamour metallic aesthetic with reflective gold panels and elegant, form-fitting design. It was built from the ground up to capture the look of modern performance and festival couture — equal parts fashion, fantasy, and stage-ready brilliance.
While trained for WAN 2.1 14B T2V, it performs flawlessly in WAN 2.2 and Image-to-Video (I2V) workflows.
🔧 Recommended Settings
High Strength:
0.6(or try 0.5 for subtle results)Low Strength:
0.8(or try 0.7 for smoother blending)
These ranges maintain rich metallic detail without overpowering the scene.
🎨 Prompting Tips
For general use:
wearing an AlessaDressFor exact control:
wearing a metallic [color] AlessaDress
→ works with gold, silver, red, green, blue, etc.Add or modify sleeves for variety:
sleeveless,short sleeve,long sleeve, etc.
(Be consistent across prompts for best results.)Lower LoRA strength to repurpose outfit details into new designs such as swimsuits, lingerie, or bodysuits.
🧠 Training Details
Base Dataset: A small curated set of real-world images of a metallic festival outfit (head cropped, backgrounds removed).
Data Expansion: Additional reference images were generated with Google’s Nano Banana, guided by a structured routine covering multiple angles, poses, and perspectives for diversity.
Image Cleanup: Every dataset image was refined in Photoshop to remove visual aberrations and artifacts.
Captioning: Captioned using Qwen2.5-VL-7B-Instruct in BooruDatasetTagManager, with a custom QUERY template ensuring structured, detailed descriptions.
Training: Trained in AI-Toolkit on RunPod for 3000 steps, optimized for visual clarity and material fidelity.