The 80/20 Rule: Your source image determines 80% of your final quality. Everything downstream is damage control.

📺 Tutorial: Midjourney V7 Complete Guide — Official documentation for V7 features including Draft Mode and Omni Reference

Example:

Woman in red dress answering phone, annoyed expression, 
shallow focus, practical lighting, 35mm film grain, hotel lobby

⚠️ Common Mistake: Long environment descriptions. Keep it minimal — complex backgrounds create animation problems later.

If a face looks "almost right" but something's off, fix it NOW. Motion amplifies every flaw.

NanoBanana (nanobanana.com) corrects:

• Facial proportions without changing identity
• Eye alignment and asymmetry
• Mouth/jawline for animation readiness

Image Generation → NanoBanana (1 pass only) → Animation

💡 Pro Tip: One pass only. Multiple passes flatten expression.

What it does: Transfer your recorded performance onto any character. Your body drives their body.

📺 Official Guide: Kling Motion Control User Guide

📺 Deep Dive: Higgsfield Motion Control Guide

1. Record yourself (3-30 sec, stationary camera, single person)
2. Upload reference video (your performance)
3. Upload character image (who you want to become)
4. Choose mode:
   • Exact = Static camera, precise match
   • Partial = Camera can move independently
5. Generate

Turn one performance into a multi-camera edit:

1. Record ONE continuous performance
          ↓
2. Generate 3-5 character images (different angles/backgrounds)
          ↓
3. Run SAME reference video against EACH image
          ↓
4. All outputs sync perfectly (they share timing)
          ↓
5. Edit together with beat-matched cuts

💡 Pro Tip: Segment into 10-second chunks. Generate all angles before editing.

Generate video (face visible, 5-10 sec)
          ↓
Isolate vocals from audio (Lalal.ai or Moises.ai)
          ↓
Apply lip sync tool
          ↓
Recombine with instrumental in editor

⚠️ Never feed a full music track to lip sync. Isolate vocals first. Ultimate Vocal Removal is incredible and free.

1. Generate base video with visible face
2. Enable "Match Mouth" tracking (~10 min processing)
3. Upload clean isolated vocal audio
4. Adjust frame offset in your editor if needed

📺 Tutorial: ElevenLabs Voice Cloning Guide — Official documentation for instant and professional voice cloning

1. Record 1-3 minutes of clean audio (no background noise)
2. Upload to Voices → Create Voice → Instant Clone
3. For pro quality: 30+ minutes audio, use Professional Clone (Creator plan required)

⚠️ Legal: Get written consent for any voice you clone commercially.

📺 Tutorial: Suno Complete Guide — Official guide to creating AI music

Describe song style in Suno
          ↓
Generate with isolated stems enabled
          ↓
Feed vocal stem to lip sync
          ↓
Recombine in Video Editor

📺 Tutorial: Topaz Video AI Documentation

Link: topazlabs.com/topaz-video

Different AI clips have different textures. Grain unifies everything.

In DaVinci Resolve:
1. Place grain asset above all footage
2. Blend mode: Overlay
3. Opacity: ~30%

This single step often does more than hours of per-clip color correction.

Lock framing → Test at low tier → Generate at full quality → Upscale ONCE → Lip sync LAST

Reordering these steps wastes money on content that gets regenerated.

• ✅ Generate at native resolution, upscale once at the end
• ✅ Test complex shots with Standard mode before Professional
• ✅ Batch similar shots before committing
• ❌ Don't use platform "enhancers" (Topaz is better and cheaper per clip)
• ❌ Don't lip sync before final framing is locked

A "failed" generation is often fixable. Repair costs time; regeneration costs credits.

Run before final export. If anything fails, regenerate — don't patch.

• Eyes track consistently, no micro-jumps
• Teeth stable across frames
• Hands don't partially disappear
• Clothing doesn't shimmer or crawl
• Background motion doesn't loop

• Hard consonants (p, b, t, d) match lips
• Breathing matches chest movement
• Room tone matches environment size

• Every cut has purpose
• Camera movement has intent
• Emotional state clear within 2 seconds

Teams that storyboard before generation report 30-50% fewer regenerations.

1. Define camera angle, lens, movement intent
2. Create reference boards for color/lighting
3. Map shot sequence with emotional purpose
4. Test with still images before video

Project/
├─ 01_Source_Images/
├─ 02_Reference_Video/
├─ 03_Generations/
│  ├─ v1_exploration/
│  ├─ v2_selected/
│  └─ v3_final/
├─ 04_Audio/
├─ 05_Upscaled/
├─ 06_Edit/
└─ prompts.txt

⚠️ Never overwrite generations. Version drift is how quality regressions sneak in.

AI output is mathematically perfect. Real footage isn't. Add controlled imperfection.

Apply after generation, before final grade. Adds halation, grain, highlight rolloff.

• Assume Rec.709 gamma 2.4 unless platform specifies otherwise
• Convert all clips to single working space before editing
• Never mix HDR and SDR without tone mapping

• Generate everything at 24fps or 30fps, never mixed
• Fix frame rate BEFORE lip sync, never after
• If jittery, apply optical flow AFTER upscaling

• Fine skin texture
• Subtle gradients
• Neon lighting
• Fog, smoke, rain

• Add light grain before export (gives encoders texture to preserve)
• Boost contrast slightly
• Avoid pure black backgrounds (macroblock badly)
• Export at higher bitrate than platform recommends

Tools change monthly. These don't:

1. Capture quality determines your ceiling. No tool compensates for bad inputs.

2. Lock framing early. Mid-process reframing cascades problems everywhere.

3. Modular separation. Treat body motion, face animation, and voice as independent tracks. Combine in editorial.

4. Regeneration beats repair. Fresh output usually costs less than fixing broken output.

5. Ambiguity multiplies cost. Know exactly what you want before generating.

Last verified: January 2026 Platform capabilities shift rapidly — confirm current features before production.