3 Hours of Making an AI Film Start to Finish (Watch Me Fail)

$6,139 in one sitting. 800 generations. 8 shots that made the cut. This is Day 4 of trying to make a fully AI-generated 80-minute feature film for Cannes. In this episode I generate two scenes start to finish — no editing magic, no skipping the failures — so you can see exactly what the iterative AI filmmaking process actually looks like. You'll learn how to spot slop vs. cinema, when to abandon a generation, why the keyframe matters more than the prompt, and the framework I use to turn 800 failed attempts into shots good enough to (hopefully) premiere at Cannes. If you missed Episode 1, watch it first: https://higgsfield.ai/s/seedance-2-0-... The shotlist builder skill I use in this video (download the file and upload to Claude): https://www.dropbox.com/scl/fi/zy8vbm... 0:00 Intro: Making an 80-Minute AI Film 1:11 Plot Summary: Scenes 21-23 1:51 Building the Shot List in Claude 5:27 Collaboration Workflow for 15 People 9:44 Environment Concept: The Apartment 13:01 Set Refinement & Set Dressing 25:01 Adding Narrative Lore through Props 31:50 Patrick Kalyn (VFX Artist) Reaction 33:25 Designing Polaroid & Note Assets 45:52 Generating Video with Seedance 2.0 58:01 Day 2 Update: Scrapping Scenes & Credit Burn 1:31:10 Meet the Team: Specialized Roles & DOPs 2:05:32 Editing Hacks: Frame-by-Frame Continuity Fix 2:45:00 Scene 21 Complete 3:05:01 Final Status: 4.4 Million Credits Spent Links 💬 Discord:   / discord   🐦 Twitter / X: https://x.com/higgsfield 📸 Instagram:   / higgsfield.ai