Talking Cow with Attitude

⚙️ Tool(s) Used

• Runway Gen-4 – for AI-generated video

• ChatGPT – for crafting prompts

• DALL·E – to generate reference image for Runway

• Premiere Pro – for video editing

• Mureka – for music and voiceover

• Manual SFX – cowbells, chewing, wind, added during edit
  

🎯 Creative Goal

To create a short, expressive, cinematic video of a cow that looks up while chewing, locks eyes with the viewer, mimics speech using only eye movement and facial changes, and slowly steps back — as if unsettled by being watched.

🧠 Approach

• Script: Written manually by me — focused on comic timing and visual beats

• Prompt Development: We iterated prompts here in ChatGPT based on the script

• Reference Image: Created via DALL·E and uploaded to Runway to match visual style

• Used Runway Gen-4, which requires a visual reference and supports 5-10 sec clips

• The desired sequence was broken into modular scenes:

  • Cow eats grass

  • Looks up

  • Eyes move left/right

  • Talks (“What? You looking at me?”)

  • Steps backward

• All scenes were generated separately and stitched together in Premiere Pro

• Added sound effects manually and sourced music and VO via Mureka
  

📈 Outcome

• Runway Gen-4 successfully maintained cow consistency across clips when using the same reference image

• However, it failed to execute precise facial or eye-only movements — could not isolate eyes from head motion

• "Step backward" prompt broke — instead of a natural backward motion, the cow ran away awkwardly or glitched

• Script-based nuance was completely lost — expressions were random, not driven by tone

• Runway’s credit system is a major blocker:

  • Standard plan included 625 credits/month

  • Each 5-second video cost 125–250 credits

  • This short consumed almost all credits

  • Upgrading to Pro ($35/month) gives 2250 credits/month — still only 3 minutes of footage, max

  • No refunds for unsatisfying clips, which adds cost and friction

📷 Output Gallery

Glitching Cow 🐮 🤣

💡 Key Takeaways

• Even with Gen-4, Runway struggles with fine motor cues like “eyes only” or subtle emotional shifts

• Prompting must be extremely literal and visual, not conceptual

• Financially unscalable for short films — cost vs. output is wildly imbalanced

• Music and sound saved the mood — post work mattered more than the raw AI footage
  

🗒️ Creative Rating

• 🔥 Creativity Potential: 4/5

• 🤖 AI Quality / Realism: 3/5

• 😅 Time vs. Reward: 3/5
  

🧪 Would I Use It Again?

Maybe — for short experimental visuals only. Not suitable yet for performance-driven storytelling or budget-aware projects longer than 30–60 seconds.