Skip to content
Lover Snap
English 简体中文 繁體中文 日本語 한국어 Español Español (México) Français Français (Canada) Deutsch Italiano Português Português (Brasil) Nederlands Svenska Dansk Norsk bokmål Suomi Polski Русский Türkçe العربية עברית हिन्दी Bahasa Indonesia Bahasa Melayu ไทย Tiếng Việt Filipino Українська Čeština Slovenčina Magyar Ελληνικά Română

Memorial · Preparation guide

How to prepare reference photos

The quality of your generated memorial photos depends almost entirely on the quality of your reference photos. Spending an extra hour selecting reference material is worth more than any prompt-tweaking after training.

Step by step

  1. 1. Pick one era of their life. The single biggest factor in generation quality is whether your reference photos span one era or several. Pick the era you want to remember them at — usually a 5-10 year span — and only use photos from that range.
  2. 2. Aim for 3-16 photos, with variety. Three photos can produce a recognizable character; more variety improves quality. Include close-ups (face and shoulders), medium shots (waist-up), and at least 2-3 full-body shots. Different settings and lighting help.
  3. 3. Scan physical prints at 300+ DPI. For photos from albums, scan at 300-600 DPI. Most home scanners default to 200 DPI which is too low for AI training. Even minor creases and color fading are fine — the AI works with imperfect inputs.
  4. 4. Convert HEIC to JPG. iPhone photos are stored in HEIC format by default, which Lover Snap does not accept. In the Apple Photos app, select the photo, tap Share, choose "Save to Files" or "Save Image" — it will export as JPG.
  5. 5. Avoid sunglasses, hats, and group photos for training. The AI needs to learn the face. Sunglasses, hats that shadow the face, and photos where the face is partially obscured all degrade quality. Group photos where the person is small in frame are weak training material — use individual or pair photos.
  6. 6. Skip heavily filtered photos. Heavy Instagram filters, beauty filters, and stylized edits give the AI inaccurate information about the person's actual features. Use natural-lighting unfiltered photos when possible.
  7. 7. Train, then evaluate before generating in burst. After training (~30 minutes), generate a single test photo to verify the character looks recognizable. If it doesn't, retrain with different reference photos before burning quota on burst generations.

What good reference photos look like

What to skip

Common questions

I only have one or two photos. Will it still work?
Single-photo characters tend to look generic — the AI struggles to learn a face from one angle. Two photos can sometimes produce usable results if they're from different angles. Three is the practical minimum.
My reference photos are very old. How old is too old?
There is no upper age limit. Photos from the 1940s scan well at 300+ DPI and produce recognizable AI characters. Black-and-white photos work as long as you have enough of them — the AI infers reasonable skin tones from context.
Can I use photos where the person is smiling vs not smiling?
Yes — mix expressions. A reference set that's all smiling photos will produce a character that always smiles. Mix in some neutral, contemplative, or candid expressions.
What if I have a great photo with sunglasses or a hat — should I use it?
Skip it for training. The AI needs the face clearly. Save those photos for inspiration on the style of clothing or setting you want to generate.