5 Cấp Độ LÀM PHIM & VIDEO AI (Hướng Dẫn Chuyên Sâu)

Summary of Key Technological Concepts, Features, and Guidance (AI Video Filmmaking)

1) Five Skill Levels for AI Video Creation (“Learning Map”)

Level 1: Basics → Quick Viral-Style Results

• Create videos from text (text to video) using AI prompts.
• Create videos from photos (image to video).
• Reuse viral prompt designs, then tweak prompts for your own characters and settings.

Level 2: Consistency & Control via Image-First Workflows

• Use an image-first pipeline: photos → AI images → then AI video to keep character/background consistency.
• Use reference images and image editing models to swap faces/props/clothes while maintaining continuity.
• Learn core photography/filmmaking composition:
  • Shot sizes: wide / medium / close / extreme shots
  • Camera angles

Level 3: Prompt Control + Multi-Shot Storytelling + Debugging

• Write more precise prompts to control outcomes (timing, camera movement, actions).
• Run tests to compare/select models for different goals (emotion scenes vs. action scenes).
• Use final frame / start-end control to precisely manage the beginning and ending of a video.
• Use multishot to build longer sequences by switching shots (often with time markers).
• Debug prompts using:
  • Positive prompt sections
  • Negative prompt sections
  • Structural adjustments
• Optimize cost/quality:
  • Test at lower resolution (e.g., 480p), then debug/upscale at 1080p.

Level 4: Advanced Control (CDEN 2.5, References, Audio, VFX, Real Footage)

• Use structured, detailed prompts for CDEN 2.5 to produce cinematic scenes up to ~30 seconds.
• Build:
  • Character reference tables (multiple angles, tilt angles)
  • Background reference images for better spatial synchronization
• Follow a CDEN 2.5-specific prompt template structure:
  • Overview/events → references → spatial setup/equipment → timing/actions → lighting/sound/constraints/physics
• Focus on audio synchronization (voice and performance matching).

Techniques for handling voice issues

• Extract character dialogue, export it as a reference clip, then import it to drive a cloned voice.
• Note Vietnamese voice-command limitations and work around them using dialogue references.

Upscaling (“firefighting”)

• Use a built-in upscaler (mentions Tens Pro up to 4K) instead of regenerating everything at higher resolution.

Real-footage applications (two methods)

1. VFX special effects on real videos
   • Upload background references + short video references
   • Generate a VFX-enhanced AI prompt
2. Mimic human motion/expressions from real acting footage
   • Record actor performance
   • Use motion control
   • Understand quality limits versus CDEN 2.5 notes

Level 5: Product Workflow + Canvas + Competition-Grade Output

• Use a professional production pipeline:
  • overview/lockline → scripts/scene shortlist → reference design (setting, angles, characters, props) → storyboard → frame/video refinement → final assembly
• Use Top View’s Canvas for a single workspace that supports iterative production across:
  • images, videos, voiceovers, notes, and text
  • linking prompts/references to assets
• A competition workflow is mentioned for CDEN 2.5:
  • 30-second videos
  • prizes and an entry submission process

---

2) Product/Platform/Tool Mentions (and What They Do)

Top View AI (Core Platform)

• Text → Video (“Convert text to video”)
• Text → Image (“Convert text to image”)
• Image → Video (“Convert images to video”)
• Omni reference / reference table workflow to improve realism and enable element replacement
• Final frame control to set precise start/end frames
• Multishot for multiple shots with timing and composition control
• Select from table to reuse generated images in video pipelines
• Canvas workspace for integrated multi-asset production

CDEN 2.0 / CDEN 2.5 / “Fast version”

• Used as video models.
• CDEN 2.5 supports longer outputs (up to ~30s) but requires more detailed prompt structure.
• Fast version + 480p testing reduces credits/cost while validating prompt behavior.

Clot (AI Assistant for Prompt Generation/Analysis)

• Upload a short clip and ask it to structure prompts for:
  • image creation (GPT)
  • video creation (CDEN 2.0)
• Later used as a “director assistant” to revise/debug prompts based on feedback.

Skills Download / Prompt Writing Skills

• “Skills” are specialized chatbots trained for specific models (e.g., CDEN 2.0 prompting) to improve prompt detail quality.

---

3) Key Techniques Highlighted for Better Results

• Viral prompt reuse + editing

  • Find viral prompts for specific models (example: CDEN 2.0 prompts listed on a website).
  • Copy and adapt them for new characters/locations and adjust details (scores/outcomes/etc.).

• Element replacement using references

  • Use Omni reference to replace specific parts (face, logos, scoreboards, etc.) using aligned reference assets.

• Image-first for continuity

  • Create high-quality AI images first, then convert to video to reduce distortion and maintain identity and props.

• Language strategy for prompt writing

  • Vietnamese for general character/background descriptions.
  • English recommended for photography/technical keywords (lighting, color palette).

• Camera & cinematography vocabulary for prompt control

  • Shot types: wide/extreme wide, medium, close-up, extreme close-up
  • Angles: eye-level (E-level), high angle, low angle, bird’s-eye, over-the-shoulder

• Model selection via structured tests

  • Run emotion tests (slow zoom, screaming/tears) and action tests (handheld fight/sword vibrations).
  • Select the best-performing model for the specific scenario.

• Debugging with prompt structure + negative prompts

  • Include instructions for exact starting frames.
  • Use negative prompts to prevent artifacts (e.g., incorrect eye reflections/emission).
  • Adjust shot composition based on problem segments.

• Resolution/cost optimization

  • Generate at 480p first, debug, then regenerate at 1080p only when needed.
  • Mentions credit savings depending on model (e.g., CDEN 2.0 Fast at 480p loses fewer points).

---

Main Speakers / Sources (As Stated in Subtitles)

• Tao (Teacher Tao / Mr. Tao) — mentor and “teacher” referenced throughout.
• Clot — AI assistant tool/software used to analyze/generate prompts.
• Top View AI — platform used for text/image/video generation and reference features.
• Paulo Prom — creator whose viral video is referenced for recreation examples.
• Film directors referenced for inspiration:
  • Victor Vu
  • Christopher Nolan