Video summary

How to Make AI Video

Main summary

Key takeaways

Technology

Summary of the Video (AI Video Creation Workflow)

ChatGPT Pro setup & model availability

  • The workflow starts with ChatGPT Pro (~$100/month) to generate and manage the pipeline.
  • The speaker mentions GPT-6 release timing and notes that in India GPT-6 may not be available yet.
    • If GPT-6 isn’t available, they suggest starting with GPT-5.6.
  • They claim a newer GPT-6 / Astra-style model is significantly better than GPT-5.6 for their current workflow.

Using ChatGPT “projects” and reusable “skills”

  • The system is organized using ChatGPT projects plus custom skills.
  • Example skill creation:
    • A prompt is used to generate a skill for selecting the best mythology topics/dates.
    • Output is produced in a specific “1970s” style.
  • Multiple skills can be created (e.g., Instagram hook master) and reused by calling the skill with a topic.

Generating visuals with ChatGPT + Midjourney

  • The speaker generates video/story images inside ChatGPT, but warns that ChatGPT images can be imperfect, such as:
    • incorrect hands/arms
    • visual artifacts
  • They use Midjourney to improve image quality.
  • After generating images, they use iterative fix prompts, for example:
    • “Choose one image size/style”
    • “Fix artifacts/arms/hands”
    • “Change only accessories, don’t change the overall style”
  • They note that style can drift across iterations, so they correct it by enforcing constraints.

Video assembly + narration/dialogue

  • They create multiple clips (e.g., “four clips”) that combine into a roughly 2-minute reel.
  • They instruct ChatGPT to:
    • add narration/dialogue
    • keep delivery polished (more realistic)
    • generate final visuals + audio together

Audio generation with 11Labs (voice selection/cloning)

  • 11Labs is used for text-to-speech:
    • choosing inbuilt “old school / radio-style” voices
    • voice cloning using ~10 seconds of audio
  • The speaker emphasizes that 11Labs credits/subscription are required and suggests it’s relatively affordable (they cite ~$5/month / ~500 INR, though exact plan details are unclear).
  • They test different voices, then tell ChatGPT to use the selected voice (example preference: “Danish Khan deep …”).

Plugins to integrate audio into video

  • The workflow mentions ChatGPT plugins to help:
    • generate audio (e.g., via 11Labs)
    • assist with web/audio/video integration
  • The key principle is to stay inside the ChatGPT workflow and use plugins rather than constantly switching tools manually.

Hook strategy for Instagram Reels

  • A major focus is hook engineering for the first 3–4 seconds.
  • The hook concept: what makes a viewer stop scrolling, such as:
    • camera action
    • talking
    • celebrity presence
    • unusual events
  • They use an “Instagram Hook Master” skill to generate multiple hook variations.
  • Hooks are inserted before/at the start of the video, creating multiple reel versions, such as:
    • generating 6 hook variations
    • optionally making 4-second hook variants

Credits/cost management vs. experimentation

  • Hook variants and video variants consume generation credits.
  • The speaker argues credits aren’t “wasted”—they’re spent on experimentation to find what performs best.
  • They compare:
    • using a team (more expensive)
    • using personal credit subscriptions
  • Their approach: generate many trials, then publish only the best-performing variants.

Caption/hashtags generation

  • The workflow includes generating:
    • captions
    • viral hashtags (sometimes specified as only 5 viral hashtags)
  • They note repetition can happen and adjust instructions to prevent it.

Quality control loop (“nudging” outputs)

  • The speaker repeatedly reviews and refines outputs, asking the AI to:
    • fix anatomy/accessories
    • prevent unwanted style changes
    • correct audio/voice choices
    • trim weak sections and polish
  • The emphasis is that the process depends on human guidance + constraints, not fully hands-off automation.

Troubleshooting distribution limits

  • Instagram may flag content as duplicate/copied even when visuals differ.
  • The speaker suspects factors like background music and plans to ask ChatGPT to modify the background audio to improve distribution.

Named planning for upcoming content

  • They plan 9-day Navratri uploads:
    • one reel per day
    • a goddess theme
  • They mention deadlines tied to Pakshya-related dates, including Sept 17 and Sept 26 (context unclear but used for scheduling access/audio readiness).

Final “Prompt Recipe” / Tutorial-Style Instructions (High-Level)

Reference videos are used as inputs, and the goal is a consistent style with old-school audio and strong hooks.

  1. Assemble reference videos into a single polished final reel (~2 minutes; sometimes ~2m30s is mentioned).
  2. Use 11Labs to create old-school narration (narration only; no talking dialogue).
  3. Add subtitles, then edit/trim.
  4. Generate multiple hook variations (e.g., 4-second hooks using camera movement/zoom/pan).
  5. Produce final reel versions by inserting hooks into the base story/video (multiple versions from the same base).

Main Speakers / Sources (as Referenced)

  • Primary speakers
    • The video creator/instructor (explaining ChatGPT Pro, skills, hooks, and the Midjourney/11Labs workflow)
    • The listener/collaborator (asking questions and confirming understanding)
  • AI tools/sources mentioned
    • ChatGPT (GPT-5.6 / GPT-6 / “Astra”-type upgrade)
    • Midjourney
    • 11Labs
    • Instagram Reels

Original video