Dia Roy — Bengali Stylist v1
A character LoRA for Deepanwita "Dia" Roy, a virtual fashion stylist character based in South Kolkata. Heritage-led fusion aesthetic with editorial photography, warm earth-tone palette, and an oxblood signature accent.
Trigger word: diaroyv2
Base model: Flux.1 Dev
Best at:
- Editorial portraits and close beauty shots
- Festive ethnic wear (saree, Jamdani, Banarasi, traditional jewelry)
- Fusion silhouettes (linen shirts, wide trousers, kurtas with denim)
- Party wear with deep necklines and wear., nightwear, or intimate-apparel styling as content. "-as-focal-point composition" / body-as-subject framing. "Suggestive poses (lying on beds, hands on body in suggestive ways)"
- Warm candid lifestyle photography
- Kolkata heritage interiors and South Asian urban settings
Recommended settings:
- LoRA strength: 1.0 (workable range 0.8 to 1.2)
- Aspect ratio: 4:5 for portraits, 9:16 for full-body
- Steps: 20 to 28
- CFG: 3.5
- Sampler: default Flux scheduler
Prompt structure: Lead with diaroyv2, then describe the scene — pose, expression, outfit, setting, lighting. The LoRA holds face, body type, hair, and signature jewelry, so the prompt should focus on what varies between shots.
Notes:
This is a private character LoRA for a disclosed virtual creator project. Use is restricted to the original creator. Not intended for generating real people, content, or other characters. All published content using this model carries AI-generated disclosure per Meta and ASCI India guidelines.
Description
Initial release. Trained on 36 captioned images covering close portraits, medium shots, and full-body editorial photography across festive ethnic, fusion, and casual registers. Trigger word: diaroyv2.
Training: Flux.1 Dev base, Kohya engine, 10 epochs, 3,600 steps, network dim 16, alpha 16, learning rate 4e-4, cosine scheduler, 1024 resolution.
Captioning methodology: trigger-word-first, identity features (face, body type, skin, hair, age, ethnicity) intentionally left uncaptioned to bake into the trigger; only variable scene elements (pose, outfit, setting, lighting) described per image.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.