AI 영화 제작 가이드⚡AI 영상 퀄리티 3배 높여주는 8가지 비법 공개

Main ideas, concepts, and lessons

• Goal: Achieve movie-like AI video cuts—not just with “flashy prompts,” but by mastering three fundamentals:
  1. Character sheet basics
  2. Camera basics (movement, angle, and camera grammar)
  3. Framing basics (frame size / shot type) + cut duration
• Debunking: Movies aren’t only for “majors” (professionals). Many famous Korean directors are portrayed as non-majors; anyone can make decent AI videos if they follow the workflow correctly.
• Best learning approach: Don’t just learn final results—learn the full production process (sheets, camera, cut choices, shot lengths).

---

Sources / learning recommendation (method)

The creator claims Fixfield / Expild (name appears inconsistently in subtitles) is presented as the best educational resource because it:

• Shares not only finished outputs
• Reveals the entire production process
• Includes projects and original series with full short stories
• Shows sheets, camera setups, and cut structures

Note: The service supposedly has no mobile app; everything runs on the web.

---

Methodology / instructions (detailed step-by-step)

1) Character sheet basics (to prevent “doppelganger” issues)

Problem

If the character sheet contains too many images/information, the AI may interpret them as multiple people, causing:

• Slightly different versions of the protagonist
• The same face appearing multiple times in one scene
• Outputs that feel like a “collection of memes” rather than a coherent character/movie

Correct approach (structure)

Use exactly three panels:

1. Full-body front without the head (mannequin/head removed)
2. Full-body rear
3. Face closer

Key rationale

• The 3rd panel already carries strong frontal facial information.
• Duplicating frontal face info wastes information and increases the chance the AI treats it as multiple identities.
• More character images on one sheet → higher probability of “multiple people” interpretation.

---

2) Camera basics (why AI videos look “bland” and how to fix it)

Common symptom

Characters and image quality look fine, but the video doesn’t feel cinematic/movie-like.

Core claim

The biggest cause is camera movement/grammar, not taste.

“Movie grammar” lesson

Movies convey emotion through camera movement. For example:

• Pushing in can feel like awakening
• Stepping back can feel like loss

Even with the same character and expression, the camera movement can change the emotional feel.

Camera prompt grammar: split into 4 parts

For each clip, the camera prompt should be organized into four components:

1. Movement: what the camera does
2. Speed: how fast it moves
3. What is maintained while moving: what the camera keeps consistent during motion
4. End: where the camera stops

Critical “missed piece”

• If you don’t specify the end point, the AI may stop awkwardly or drift, making the last moment feel wrong.

Don’t combine camera directions with scene descriptions

• Rule: Separate camera directions from scene descriptions.
• Reason: If bundled together, you may have to rewrite the whole block every time the scene changes.
• Benefit: If separated, you can reuse the same camera setup across frames/scenes more easily.

---

3) Camera movement library (10 ways)

The video lists 20 total camera-related items (movement + angle + techniques). For movement, it gives 10 examples:

1. Runner / Zoom-in close (used at realization/awakening moments)
2. Slow zoom / dolly-style pull of lens (compression/pressure; quiet tension)
3. Arc shot (circling the figure; explains space when the character first appears)
4. Follow shot (behind at shoulder height; audience feels “behind” the character)
5. Reverse tracking (stepping back from the person; “speaking” walking style)
6. Handheld (subtle hand tremor; realism/anxiety/documentary feel)
7. Whip-pan (fast swing with afterimage; used to transition/signal impact)
8. Crane (vertical rise while horizontal orientation stays; character shrinks; often near endings)
9. Snorricam (camera attached to body; character fixed while background shakes)
10. Push fast / pass-through (camera “passes through” a door/window to transition spaces)

Beginner mistake to avoid

• Don’t pack two or more movement types into one shot.
• If you combine movements like “move up” + “dolly/rotate together,” the camera may “tremble awkwardly” because it doesn’t know where to go.

---

4) Camera angle basics (movement vs angle are different axes)

• Movement = how the camera moves
• Angle = where the camera is looking (a separate axis)

Beginners often change movement but not height/angle, so the image doesn’t change emotionally.

Angles: 5 types

1. Rear shot (fixed behind person; no face shown)
2. Low angle (near ground looking up; weakness/hero framing; character feels strong)
3. High angle (above looking down; helplessness/isolation/being watched)
4. Overhead (~90° straight down; space looks flat/fateful overview)
5. Dutch (Birch) angle (~20° tilt; twisted world; anxiety/distortion)

---

5) Special techniques: distortion/focus/perspective effects (5 items)

1. Dolly zoom (Dollijum): pull back while zooming in so character size stays while the background stretches/distorts
2. Rack focus / Lag focus: camera stays still; focus shifts front→back to force attention and reveal hidden info
3. Side tracking: side-by-side formation; profile view repeated; background flows at different speeds (rhythm/motion)
4. First-person POV: eye-level; face not shown; hand appears at bottom (puts audience “inside” character)
5. Drone fullback: backward+upward while character stays centered; space expands (scale/height; often an “ending cut”)

---

6) Add one more foundational item: still shot

Before movement and camera language, include still shots:

• The camera can be completely static; “not moving” is also a valid cinematic choice.

---

Framing basics (third “basic cut” and how to choose it)

What the “third basic frame” means (frame size / framing crop)

• The “third basic” is frame size:
  • Where the protagonist is cropped within the frame
  • Presented as the essence of framing

Shot types: 9 frame sizes (as listed)

1. Extreme long shot (character tiny like a dot; emphasizes space; opening/ending)
2. Long shot (full body with margins top/bottom; where you are; “staged shot”)
3. Full shot (head-to-toe fits perfectly; shows costume + movement)
4. Medium full shot (from below knee; “cowboy shot”; good for two-person talking)
5. Medium shot (cut at waist; basic for dialogue)
6. Medium close-up (chest) (from mid-section; good for news/interviews; facial + hands readable)
7. Close-up (head/shoulders) (emotion conveyance)
8. Closer (forehead to chin) (emotion at peak)
9. Extreme close-up (eyes) (maximum detail/tension)

Emotional effect rule (distance → role)

• Far away → audience becomes an observer
• Close → audience becomes involved/participant

Relationship-aware framing

Framing isn’t only crop size—it also depends on:

• Who is in the shot
• Relationships

The subtitles mention “eight reactions” category names (some may be unclear, but includes):

• Single, Two Shot, Three Shot, Over the Shoulder, Insert, Cutaway

These names are described as enough to start.

---

Cut duration basics (fourth basic cut duration)

The video argues AI does not automatically generate consistent cinematic lengths (e.g., rejecting a rule like “maximum of 15 or 30 seconds” as not the point).

• Duration = how long the cut lasts.

Example workflow claim from a Fixfield project

A Fixfield-made project (described as “Zapier like this,” including a plot about “Pieros Squad” protecting against aliens”) used:

• Cut lengths specified per car/segment, not all at once
• Some segments: 7 seconds, others: 10 seconds

Why cut length matters (3 reasons)

1. AI tries to fill the given time
   • If you request 30 seconds but described actions cover only ~7 seconds, AI may invent/extend ~23 seconds → risk of awkward silences.
2. Editing grammar in film
   • Commercial films often use shorter cut lengths:
     • Often 3–5 seconds
     • Action sequences often shorten further
3. Modification cost
   • If something breaks at 20 seconds inside a 30-second cut, you must redo the whole 30 seconds.
   • If you use 7-second chunks, only the failed chunk is redone → reduced loss to 1/4.

Practical guideline for selecting duration

• Object close-ups / inserts: 3–4 seconds
• Character movement / action: 5–7 seconds
• Emotional facial expressions / dialogue: 7–10 seconds
• Setup shots showing space: 5–8 seconds
• Extend to 10+ seconds only if intentionally going for a long take
• If using longer takes, keep camera movement as a single unit

Key principle

• “Length is design, not intuition.”
• Choose duration based on the purpose of the cut.

---

Putting it all together (final assembly order)

The creator summarizes a production sequence:

1. Create character sheet using three panels
2. Lock the beginning of the prompt
   • Include only the face + costume
   • Don’t remove the background
   • Only one character
3. Attach the third camera prompt (camera grammar part) / camera instructions
4. Ensure each clip includes only one movement unit:
   • movement, speed, framing should be treated together as one coherent camera setup
5. Set duration according to the cut’s purpose (e.g., 3s, 7s, 10s+ for long takes)
6. Paste the “fifth letter in the order of the timeline” (subtitle unclear; intended meaning is: assemble clips in the correct timeline order)

Final emphasis: camera angles/proportions weren’t “skill” issues—people just lacked the correct sequence/workflow. The hardest part after technical fundamentals is the story/content material.

---

Speakers / sources featured

• Video creator / narrator (unnamed): the person giving the AI filmmaking guide.
• Fixfield / Expild (website/service mentioned): claimed educational resource; also described as producing an “Original Series” and projects used as examples.