Best practices for AI video lip-sync? Getting weird mouth movements

I've been experimenting with AI lip-sync tools (D-ID, Synthesia, HeyGen, Wav2Lip) for creating talking head videos, and while the technology is impressive, I keep running into the same issue: weird, unnatural mouth movements that make it obvious it's AI-generated.**The Problem:**The lips technically sync with the audio, but:- Mouth movements look robotic or exaggerated- Sometimes the jaw moves but the lips don't (or vice versa)- Transitions between phonemes feel "snappy" instead of smooth- The lower face looks disconnected from the rest of the expression**My Current Setup:**- Tool: Primarily using HeyGen (also tried D-ID and Wav2Lip)- Source: High-quality 4K video of myself with good lighting- Audio: Clean voiceover recorded in a studio- Settings: Mostly default, but I've tried adjusting sensitivity**What I've Tried:**✅ **Better source footage** - Multiple angles, neutral expression, good lighting- Helped a bit, but still not natural✅ **Slower speech** - Thinking it might help the AI keep up- Made it worse, actually looked MORE robotic✅ **Different AI tools** - Tried 4-5 different platforms- All have similar issues, just in different ways❌ **Post-processing** - Tried smoothing in After Effects- Doesn't really help, just makes it blurry**My Questions:**1. **What are your go-to settings?** Are there specific parameters that make a huge difference?2. **Source video best practices?** Should I be recording with my mouth slightly open? Specific expression? Distance from camera?3. **Audio quality impact?** Does the quality/format of the audio file affect lip-sync accuracy?4. **Which tool is actually best for natural results?** I know they all claim to be the best, but from real experience?5. **Is manual cleanup worth it?** Should I just accept I'll need to do frame-by-frame fixes in post?**Specific Technical Questions:**- Frame rate: Does 24fps vs 30fps vs 60fps matter?- Resolution: Is 4K overkill, or does it actually help?- Compression: Should source video be uncompressed?**Use Case:**I'm creating educational content and need it to look professional. Right now, the uncanny valley effect is too strong and it's distracting viewers from the actual content.Has anyone cracked the code on making AI lip-sync look truly natural? Or are we just not there yet with the technology?