공학 쇼츠는 이렇게 만들 수 있습니다 (ft. 구글 플로우, 코덱스)

Summary of the subtitles (technological workflow + product features)

The video explains a step-by-step workflow for generating engineering shorts (YouTube-style short videos) using two AI tools: Codex and Google Flow—with voice generation via Google Generay Speech. The example topic is an animated depiction of air/water-like fluid behavior, and later in the video, an explanation of a supersonic shockwave / sound barrier (Mach 1).

---

Tools and roles

• Codex: generates the script, image prompts, and video prompts (including how many frames/images to create based on script length).
• Google Flow: performs image generation and video generation using Codex’s prompts.
• Google Generay Speech: generates the narration audio (TTS).
• CapCut or Premiere: final assembly and editing (timing, audio mixing, rendering).

---

Production workflow (tutorial-style steps)

Step 0: Prepare the project inputs

1. Download an MD file (provided via the creator’s comments).
2. Paste/attach the MD file to Codex input and submit to begin video production.

Step 1: Script → TTS → determine keyframe count

• Enter/finalize the script text in Codex.
• Codex generates narration using TTS, and reports the speech length.
• That speech length determines how many keyframe images to create.
• Subdivision rule (as described):
  • 1 scene/image roughly corresponds to 2–4 seconds.
  • Codex generates ~18–32 images worth of prompts (depending on script length).

Step 2: Generate keyframes (images without infographic overlay)

• Ask Codex for an image generation prompt for images that do not include infographic elements.
• Copy the resulting prompt into Google Flow and generate all keyframe images at once.
• Agent mode must be enabled in Google Flow:
  • The description says: if the “light is white,” Agent Mode is on.

Step 3: Generate infographic versions (overlay infographics onto the keyframes)

• Ask Codex again to create an infographic prompt that includes infographic elements for the previously generated images.
• The key technique described:
  • Regenerate with the same image while overlaying an infographic on it.
• Paste that prompt into Google Flow to generate infographic-enabled images in one go.

Step 4: Generate the final video (using both image sets)

• Ask Codex to write a video prompt.
• Copy the full Codex video prompt into Google Flow.
• Google Flow generates a video clip with motion graphics “all at once” using:
  • Two images per segment: one with infographic, one without.

---

Video generation settings, model, and cost info

• Example output:
  • A 4-second clip generated using the Google OmniFlash model.
• Credits and pricing mentioned:
  • 7 credits for a 4-second video clip.
  • 2,500 credits ≈ 36,000 won
  • Approx. per-credit cost: 14.4 won
• Cost scaling example:
  • 15 clips × 4 seconds → ~1 minute total → ~1,512 won
  • A 2-minute video would be roughly double.
• Bulk generation / sequencing:
  • Multiple videos can be created at once.
  • To recover correct order, download the project and rename files with sequence numbers.

---

Downloading, organizing, and editing

1. In Google Flow, click three dots (top right) → Download Project.
2. Unzip the downloaded project and save generated videos locally.
3. In Codex, provide the folder path and request filename prefixes like 01, 02, 03 to ensure correct narration/video order.
4. Voice generation details:
   • TTS voice is Google Generay Speech
   • Script audio up to 2 minutes is free
   • Voice model: Google Jaminai 2.5%
   • Voice name: Achiid
   • Speed: Rapid Fire
5. Save the generated .wav audio.
6. Import video clips + audio into CapCut or Premiere:
   • Arrange clips to match script/audio timing.
   • Adjust clip lengths to match narration.
   • Lower sound effect volume so narration stays clear.
   • Render/export the final video.

---

Content example / analysis embedded in the shorts

The creator demonstrates how the method can narrate and visualize scientific concepts:

• Supersonic effects explained as a shockwave phenomenon:
  • Sound is wave-like (compared to water waves).
  • At Mach 1, emitted sound waves bunch up and create a high-pressure “barrier”.
  • As the plane passes, pressure/temperature drop quickly, causing moisture condensation into a cloud band.
  • The shockwave sweeps the ground causing a loud “explosion-like” effect (e.g., breaking glass).
• References to:
  • The “sonic boom” look (clouds burst first visually, but the fire/roar follows).
  • The Concorde passenger jet and its sonic-boom issues over land.
• The video ends with the creator encouraging viewers to enjoy the process.

---

Main speakers / sources (as stated)

• Creator / narrator (single main speaker describing the workflow)
• Codex
• Google Flow
• Google OmniFlash (video model)
• Google Generay Speech (TTS / narration)
• CapCut / Premiere (editing software)