How to choose source images for pets, people, and characters
Practical advice on clarity, occlusion, backgrounds, resolution, and using multiple reference images for a stable character.
Generation quality is not determined by pixel count alone. A readable subject, complete identity cues, and consistent references matter more than extreme resolution.
Choose a clear image with a complete subject
Use a front or slight three-quarter view where the eyes, face shape, and main accessories are visible. For pets, show the ears, chest, and coat pattern. For people, avoid hair or hands covering most of the face.
Professional photography is unnecessary, but avoid motion blur, heavily compressed screenshots, or images where the subject occupies only a tiny area.
Simple backgrounds help—manual cutouts do not
A visible contrast between subject and background makes the outline easier to understand. Everyday indoor photos are fine when clutter does not overlap the subject.
You do not need to remove the background. Rough cutout edges, halos, and missing fur can accidentally become part of the generated identity.
Keep multiple references consistent
When using several images, make sure they show the same character with reasonably consistent hair, clothing, or accessories. Conflicting references make it unclear which version should be preserved.
Use one main reference for the face and overall feel. Additional images should only clarify a side view, marking, or accessory.
A quick pre-upload check
Zoom in to verify that the eyes and outline are clear and that no sticker, watermark, or caption covers the subject. Then shrink the image to thumbnail size and check whether the character is still recognizable.
JPG, PNG, and WebP all work. Normal photographs do not need conversion or artificial upscaling before upload.