AI Video Character Consistency Remains a Challenge
ke1lle · reddit · 2026-07-13
The author encountered a typical issue while making AI videos: the same character gradually "changes face" across multiple shots.
Their workflow involved generating a character with Midjourney, then using the reference image to generate 5 video clips in Kling. However, by the 3rd clip, face drift began, and by the 5th, even the jacket color and hair had changed. Although creating front, three-quarter, and side profile reference images improved the results, it remained unstable.
They believe the only truly stable method might be training a LoRA, but they lack the VRAM and patience. Currently, they mostly use Framia to manage storyboards and reference images, which at least prevents asset chaos.
More from Multimodal
- Magnific teases October 28 release: 'We're making it move' — cuenca · 2026-10-27
- Midjourney --sref demo: retro Americana Batman art series — gizakdag · 2026-10-27
- AI video is a new medium, not cinema or animation — and it's perfect for depicting dreams — Kyrannio · 2026-10-27
- Dad turns his toddler's 31-year-old quote into a song via SUNO kids' choir — NathanpmYoung · 2026-10-27
- Band's local-AI music video hits 200k views: 900 generations kept 64, 3-second clips beat face drift — Hwang0500 · 2026-10-27
- Adobe Firefly video editor walkthrough: one-sentence Quick Cut turns 32s into 20s — LudovicCreator · 2026-10-26