GOAL OF PART 1

By the end of this guide, you will have:

one stable AI face
60–100 images of the same person
ready dataset for training
images prepared correctly (size, quality, structure)

🧩 TOOLS YOU WILL USE (NO ALTERNATIVES)

1) Image generation

Choose ONE:

Midjourney
https://www.midjourney.com

Leonardo AI (recommended for dataset)
https://leonardo.ai

2) Image preparation app (IMPORTANT)

✅ Best option (simple + free)

👉 Photopea (online Photoshop)
https://www.photopea.com

Why:

works in browser
batch resize possible
no installation
perfect for dataset prep

Alternative (if you prefer):

Canva
Photoshop
GIMP

But in this guide I use Photopea.

✅ STEP 1 — CREATE BASE FACE (1 IMAGE)

PROMPT (copy-paste)

ultra realistic portrait of a young woman, 24 years old, european facial features, symmetrical face, natural skin texture, subtle freckles, neutral expression, minimal makeup, soft natural window light, 85mm lens, shallow depth of field, cinematic color grading, realistic skin pores, professional photography, high detail --ar 2:3 --stylize 30

What you do:

Generate 10–20 images
Choose ONE face you like
Save it → this is your AI character

⚠️ Rule:
You NEVER change this face later.

✅ STEP 2 — GENERATE FULL DATASET (60–100 IMAGES)

Now we create the same person in different situations.

CORE DATASET PROMPT (identity lock)

same person, same facial identity, identical face structure, realistic skin texture, professional photography

You always ADD this to prompts.

🧩 DATASET STRUCTURE (COPY THIS)

You must generate images in 6 categories.

1️⃣ BASIC PORTRAITS (20 images)

PROMPTS

A. Front view

same person, front view portrait, neutral expression, soft natural light, realistic skin texture, professional photography

B. 3/4 angle

same person, 3/4 angle portrait, natural expression, soft window light, realistic skin pores

C. Profile view

same person, side profile portrait, natural lighting, realistic skin texture

D. Slight smile

same person, slight natural smile, realistic skin texture, soft lighting

2️⃣ LIGHTING VARIATIONS (15–20 images)

Window light

same person, soft window light from left, realistic shadows, professional portrait

Studio light

same person, soft studio lighting, beauty portrait, realistic skin texture

Outdoor light

same person, outdoor natural daylight, realistic skin tones

Cinematic light

same person, cinematic lighting, soft shadows, film color grading

3️⃣ CAMERA & LENS VARIATIONS (10–15 images)

85mm portrait

same person, 85mm lens look, shallow depth of field, professional photography

50mm portrait

same person, 50mm lens, natural perspective, realistic portrait

Close-up

same person, close-up portrait, detailed skin texture, realistic pores

4️⃣ CLOTHING VARIATIONS (10–15 images)

Minimalist outfit

same person, full-length, minimalist outfit, neutral colors, professional portrait

Casual style

same person, full-length, casual clothing, natural style, realistic portrait

Business style

same person, full-length, business outfit, corporate style, professional photography

5️⃣ BACKGROUND VARIATIONS (10–15 images)

Neutral background

same person, full-length, clean neutral background, studio portrait

Lifestyle background

same person, full-length, lifestyle background, blurred environment, natural light

Urban background

same person, full-length, modern city background, realistic lighting

6️⃣ HIGH-QUALITY PORTRAITS (10 images)

These will be your “gold images”.

same person, ultra realistic portrait, editorial photography, soft natural light, cinematic color grading, realistic skin pores, professional fashion portrait

✅ RESULT OF STEP 2

You must have:

minimum: 40 images
ideal: 60–100 images

All with the SAME FACE.

✅ STEP 3 — PREPARE IMAGES (IMPORTANT!)

Tool: Photopea

https://www.photopea.com

What you do with each image:

1) Crop face correctly

Face must be centered.

2) Resize

Choose one size:

✅ 512 × 512 (recommended)
or
✅ 768 × 768 (better quality)

Photopea:
Image → Image Size → set 512×512 → Save as JPG.

3) Remove bad images

DELETE images with:

distorted face
weird eyes
unnatural skin
strong filters
extreme makeup

Dataset quality = model quality.

✅ STEP 4 — DATASET STRUCTURE (READY FOR TRAINING)

Create folder on your computer:

AI_MODEL_DATASET
│
├── 01_portraits
├── 02_lighting
├── 03_camera
├── 04_clothing
├── 05_background
├── 06_premium

Put images inside.

🧠 IMPORTANT SECRET (WHY MOST FAIL)

Most people generate random images.

You did something else:

✔ same identity
✔ controlled variations
✔ realistic lighting
✔ clean dataset

This is EXACTLY what LoRA needs.

Сontinuation of this article in PART 2 — https://aiworkflowtips.com/create-a-realistic-ai-model-part-2-train-your-ai-face-in-leonardo-ai/