Published September 25, 2026by vivekchahaman600

Vivekamarisvoss LoRA

81 views0 reactions0 comments on CivitAI0 collected

rajasthani woman, traditional,training guide

HOW TO USE

Vivekamarisvoss LoRA

A practical guide for SDXL 1.0 and DreamShaper XL

Vivekamarisvoss is a character LoRA trained for SDXL 1.0 and compatible SDXL checkpoints such as DreamShaper XL (standard, Turbo, and Lightning). It is meant to lock a consistent Indian woman look: long face, sharp elegant pointed nose, defined jawline, piercing eyes, long hair, and very fair skin.

This guide covers installation, the trigger word, starter prompts, recommended weights, and settings for Automatic1111 / Forge and ComfyUI.

1. What you need

An SDXL checkpoint (not SD 1.5, not Pony-only, not Flux). Good bases: stock SDXL 1.0, DreamShaper XL (alpha / Turbo / Lightning), Juggernaut XL, or similar photoreal SDXL mixes.
The Vivekamarisvoss .safetensors LoRA file.
A UI that can load SDXL LoRAs: Automatic1111, Forge, or ComfyUI.

Put the file here:

Interface

Folder

Automatic1111 / Forge

models/Lora/

ComfyUI

ComfyUI/models/loras/

Restart the UI or refresh the LoRA list after copying the file.

2. Trigger word

Always include the trigger in the positive prompt:

Vivekamarisvoss

Without it, the LoRA may barely affect the face. Put it near the start of the prompt.

Optional emphasis if the face is drifting:

(Vivekamarisvoss:1.1)

Do not stack many extra character LoRAs on top of this one. One identity LoRA at a time is cleaner.

3. Starter prompt (the official base)

Positive

Vivekamarisvoss, 1 Indian woman, long face, sharp elegant pointed nose, defined jawline, piercing eyes, long hair, very fair skin, orange embroidered Lehenga Choli dress, soft dramatic studio lighting, hyper-realistic, photorealistic, close up face front

Negative (recommended starting set)

deformed, extra fingers, extra limbs, fused fingers, bad hands, mutated hands, poorly drawn face, asymmetrical eyes, blurry, low quality, worst quality, text, watermark, logo, jpeg artifacts, oversaturated, cartoon, anime, 3d render

Keep the facial descriptors in the first half of the prompt. Change clothing, pose, camera, and location after the identity block.

4. LoRA weight

Start here and adjust by eye:

Goal

Weight

Balanced face + pose freedom

0.7 – 0.8

Strong face lock (portraits)

0.85 – 1.0

Softer influence / mix with another style

0.5 – 0.65

Above 1.0 often over-bakes skin, jewelry, or fabric and can flatten expression. If the lehenga or embroidery “sticks” when you want a different outfit, lower the LoRA to 0.65–0.75 and describe the new clothes more strongly.

Automatic1111 / Forge syntax (replace with your exact filename):

<lora:Vivekamarisvoss:0.8>

You can also set Unet and CLIP separately if your UI allows it, for example <lora:Vivekamarisvoss:0.8:0.8>. Keep both similar unless the face is strong but clothing is fighting the prompt.

5. Settings by base model

A. Standard SDXL / DreamShaper XL (non-Lightning)

Use these when the checkpoint is not a Lightning or Turbo distill.

Setting

Value

Resolution

1024×1024, or 832×1216 / 896×1152 for portraits

Sampler

DPM++ 2M Karras or Euler a

Steps

25–40

CFG

5–7

Clip skip

1 (try 2 only if the checkpoint card says so)

High-res fix: upscale 1.5× at denoise 0.25–0.35 after a good base image.

B. DreamShaper XL Lightning / Turbo

Lightning and Turbo need low CFG and few steps. Using 30 steps and CFG 7 on Lightning usually ruins the image.

Setting

Lightning

Turbo

Sampler

DPM++ SDE Karras

DPM++ SDE Karras (or LCM if the card says so)

Steps

3–6

4–8

CFG

Resolution

1024×1024 first pass

Keep LoRA weight the same (0.7–0.9). Do not raise CFG to “force” the trigger; raise the trigger weight or LoRA strength instead.

6. Automatic1111 / Forge — step by step

Load an SDXL or DreamShaper XL checkpoint.
Enable the LoRA from the LoRA tab (or type the <lora:...> tag).
Paste the starter prompt. Put Vivekamarisvoss first.
Set weight 0.8.
Match sampler / steps / CFG to the checkpoint type above.
Generate a batch of 4. Pick the seed that holds the nose, jaw, and eyes.
Then change only clothing, camera, or background.

Example full positive line:

7. ComfyUI — step by step

Load Checkpoint → your SDXL / DreamShaper XL model.
Insert Load LoRA between the checkpoint and the CLIP / sampler chain.
Set lora_name to the Vivekamarisvoss file.
Set strength_model and strength_clip to 0.8.
In CLIP Text Encode (Positive), start with Vivekamarisvoss and the identity tags.
Sampler settings must match the checkpoint: full SDXL about 30 steps and CFG 6; DreamShaper XL Lightning 4–6 steps, CFG 2, dpmpp_sde + karras.

If the LoRA “does nothing,” check two things: the checkpoint is SDXL-family, and the trigger word is in the positive CLIP node.

8. Prompt recipes beyond the close-up

Keep the identity block. Swap only the last part.

Three-quarter portrait, outdoor

Vivekamarisvoss, 1 Indian woman, long face, sharp elegant pointed nose, defined jawline, piercing eyes, long hair, very fair skin, orange embroidered Lehenga Choli dress, standing in a palace courtyard, golden hour light, photorealistic, medium shot

Different outfit

Vivekamarisvoss, 1 Indian woman, long face, sharp elegant pointed nose, defined jawline, piercing eyes, long hair, very fair skin, white silk saree with gold border, soft window light, photorealistic, close up

Full body

Vivekamarisvoss, 1 Indian woman, long face, sharp elegant pointed nose, defined jawline, piercing eyes, long hair, very fair skin, orange embroidered Lehenga Choli dress, full body, standing, studio backdrop, photorealistic

If the face weakens on full-body shots, raise LoRA to 0.9. Generate a portrait first and inpaint the body, or use a face-detail / after-detailer pass. Do not try to force a close-up token onto a full-body crop.

9. Quality tips

One identity LoRA. Stack style or lighting LoRAs at 0.3–0.5 if needed, not a second face LoRA.
Describe the face every time for the first few sessions. After the model locks, you can shorten to: Vivekamarisvoss, 1 Indian woman, very fair skin — plus outfit.
Skin: “very fair skin” is part of the trained look. For a different complexion, lower LoRA weight and write the new skin tone clearly.
Jewelry and embroidery can over-fit. If gold work appears on every dress, add “plain fabric” or name a specific textile, and drop weight slightly.
Eyes: “piercing eyes” is strong. For a softer look, replace it with “soft gaze” and use weight 0.7.
Seed-lock a good face, then vary only CFG or a single clothing token.

10. Troubleshooting

Problem

Fix

Generic face, trigger ignored

Trigger first in prompt; weight 0.85–1.0; confirm SDXL LoRA on an SDXL checkpoint

Melted or plastic face

Lower weight to 0.65–0.75; lower CFG; add “natural skin texture”

Same lehenga every time

Weight 0.7; put new outfit in parentheses, e.g. (red silk saree:1.2)

Lightning images noisy or dark

CFG 2, steps 4–6, DPM++ SDE Karras — do not use full-SDXL settings

Hands / extra fingers

Keep them out of close-up crops; use a hand-fix or inpaint later

Wrong base / garbage output

Do not load this LoRA on SD 1.5, Flux, or SD3

11. Short card you can paste under the model

Use this block on Civitai or in a pinned comment:

Trigger: Vivekamarisvoss

Base: SDXL 1.0 and DreamShaper XL

Weight: 0.7–1.0 (start at 0.8)

Base prompt:

DreamShaper XL Lightning: 3–6 steps, CFG 2, DPM++ SDE Karras.

Standard SDXL: 25–40 steps, CFG 5–7.

Note

This LoRA is for generating an original AI character. It does not depict a real person. Replace the filename inside <lora:...> tags with the exact name of your downloaded .safetensors file.

Guide prepared for Vivek Chauhan · Vivekamarisvoss character LoRA · September 2026