Kling 3.0 이렇게 안 하면 무조건 실패합니다 | 영화처럼 보이는 AI 광고 영상 제작

Overview

The video is a tutorial on making more reliable AI video ads with Kling 3.0, focusing on:

• Prompt engineering for multishots
• Avoiding wasted credits in the Kling 3.0 Store

The speaker proposes a structured workflow and compares Kling 3.0 vs Kling 3.0 Omni, based on how their underlying behaviors handle “reference vs prompt.”

---

1) Action-mapping prompt formula (most stable prompt structure)

Core recommendation

• Structure the video storyboard as shot-by-shot actions.
• For each shot:
  • Define one clear action
  • Ensure the camera angle/composition matches that action

This “action → camera planning” approach is presented as the prompt structure that yields the most stable Kling 3.0 results.

Why this matters

• If the sequence is structured incorrectly (for example, around emotions instead of discrete actions), the generated result can become “tangled,” and the procedure/order can get mixed up.

Example (laundry routine)

• Sequence actions: insert coin → open door → load laundry → close door → return to seat
• With this structure, the generated shot order aligns properly with the intended shot list.

---

2) Why generating within ~10 seconds is more stable (despite 15-second capability)

• Kling 3.0 can generate up to 15 seconds, but the speaker argues stability is better when you keep generation under 10 seconds.
• Longer generations increase drift, where identity/consistency breaks as the video progresses.

Drift example

• A character’s identity changes (e.g., switching from the intended person/appearance to another gender/face), and scenes deviate from the initial shot list.

Key takeaway

• Use ~10 seconds (or less) to keep generated content aligned with the designed cut/shot plan.

---

3) Kling 3.0 vs Kling 3.0 Omni (structural differences and when to use each)

Kling 3.0 = “Prompt-first / infer scenes from prompts”

• The model primarily follows the prompt’s instructions, inferring scenes accordingly.
• Best for: designing cut structure and camera angles shot-by-shot.
• Limitation: longer shot lists increase drift because the model must infer intervening details.

Kling 3.0 Omni = “Reference-first / maintain consistency by direct referencing”

• Omni is described as reference-based, prioritizing what must remain consistent.
• Best for: sequences where maintaining references (characters/products) matters more than strict prompt-driven cut inference.

---

4) Practical feature differences (multi-element binding, UI workflow, and limitations)

Multi-element binding

• Kling 3.0: binds/references up to three elements
• Kling 3.0 Omni: can directly reference three or more elements (multi-model aspect emphasized)

Speaker’s rule:

• If you need 3+ elements, prefer Kling 3.0 Omni.

---

How prompting is written/applied (UI workflow notes)

• Omni prompt behavior differs in the UI.
• The speaker warns that copy/paste may not apply immediately, and instructs users to press the relevant @ button / reference directly.

Workflow includes selecting:

• total generation time
• Native Audio Multishot
• then Generate

---

Extract Frame function difference

• Kling 3.0: supports Extract Frame (extracting a frame from the generated video to reuse later)
• Kling 3.0 Omni: does not support Extract Frame in the same way

If you need the image:

• you must capture/upload it separately.

---

5) Decision guide: which mode to use when

Choose Kling 3.0 when:

• You’re primarily designing cuts/camera angles up front
• You can keep generation around ≤ 10 seconds to reduce drift

Choose Kling 3.0 Omni when:

• Maintaining reference consistency (characters/products/linked elements) is the higher priority
• You need direct reference for 3+ elements
• You accept that Extract Frame workflow may require manual capture/upload

---

Main speakers / sources

• Main speaker: The single presenter/author of the tutorial (no other named source is clearly identified in the subtitles).