November AI Image Generation Tips

Make photo-editing instructions specific: don't just say "help me retouch this photo." Break the background, lighting, and color tone into clear prompts so the results don't all look the same.

Write portrait prompts in segments: identity, pose, props, composition, and skin and lighting. Specify poses down to angles and hand positions for higher reproducibility.

First choose references from the no-login pose and outfit libraries, then mix and match freely and lock the composition, instead of repeatedly drawing blind boxes.

The same GPT Image prompt produces noticeably different styles in chat mode and work mode, so it's worth trying both modes when generating images.

Write characters, outfits, background, and camera separately in the prompt, and let outfit and hair/makeup vary randomly. You can then reuse it repeatedly as a gacha-style template.

To mass-produce outfit street-style content, keep the character setup, angle, and background fixed, change only the clothes and movements, then use image-to-video to bring the scenes to life.

For picture-book-style shorts, use a high-contrast color anchor plus one core prop. Midjourney handles the art, then hand it off to a video model for motion.

To swap faces locally, use the BFS face-swap LoRA in ComfyUI to separately do face swaps, head swaps, and body swaps, preserving the original image's scene and lighting.