AI production workflow

Two of the biggest complaints on X are credits burned on regenerations and characters that don't stay consistent from clip to clip. This workflow is built to fix both.

1. Research and idea

   Use an LLM (Grok, ChatGPT) to explore, but check facts against primary sources. Write down your own angle.

2. Script

   Write it yourself, or heavily rewrite it in your voice. A generic one-shot AI script is exactly what YouTube means by "mass-produced."

3. Voice

   Your own voice is best. If you use an AI voice, a clone of your own voice keeps your channel sounding like you; ElevenLabs (affiliate link) offers voice cloning and text-to-speech. Either way, tune pacing and pronunciation by ear.

4. Character and style sheet

   Before generating anything, define a sheet: features, wardrobe, color palette, a reference image, and a seed if the tool supports it. Reuse it in every prompt to keep things consistent.

5. Images and video

   Test with stills or short low-res clips first, and only render the final version once the shot works. Tools: Veo, Kling, Runway. The same prompt behaves differently across models, so adapt it to each one.

6. Editing

   Cut, trim, and add licensed music (CapCut, Epidemic Sound). Editing is where your contribution shows most.

7. Thumbnail and title

   Keep them honest. Misleading titles and thumbnails can fall under YouTube's spam and deceptive practices rules, and many viewers say they avoid content that looks like low-effort "AI slop."

8. Upload

   Turn on the altered or synthetic content disclosure when it applies (see Policies).

Tool names are examples. Pricing and credit costs change often, so we don't list them until they're verified.