Direct AI fight scenes through controlled iteration: Preserve physics through the impact cut

2026-08-13 · Academy lesson · Adil · recording, 2:02 · original

Topic: Video generation & shot direction (also: Prompting & AI assistants) · mode: Tutorial · tags: course chapter, series, one creator's experience, prompts shared, fiction

In the impact-cut lesson of Direct AI fight scenes through controlled iteration, Adil breaks down the prompt for the bullet-time first strike: a pose match lock across the cut, the tension written in centimetres and the impact written as physics, then reruns a good-but-imperfect take as a batching game.

Part of another source

This recording is almost entirely part of Hell Grind: Rebuilding Our Biggest Fight Scene from Scratch (2026-08-05); what is said in it is covered on that page.

What is said

• 00:00:00 — The bullet-time first strike and why its prompt works: match lock, tension in centimetres, impact as physics
• 00:01:16 — The first result: a bad roll, not a broken prompt
• 00:01:50 — The final cut of the strike

Threads in this video

• restate · Which LLM should be used to write video prompts? — For the first strike Adil gave Claude a quick text description and got back the final prompt that made the shot. (00:00:01)
• restate · How do you direct a multi-shot sequence? — Adil gives the model a strict match lock: the poses at the end of shot one exactly match the start of shot two; otherwise the model treats the cut as a brand new scene and throws the poses away. (00:00:28)
• restate · How do you get precise motion (dance, action, choreography)? — For the slowed-down tension Adil maps the distance between the blades in centimetres, second by second, instead of writing "time slows down"; the Claude skill calculates the numbers. (00:00:28)
• restate · When a generation fails, should you generate more takes, fix the prompt or start over? — The first bullet-time take was decent and its movement right, so Adil treated it as a batching game, not a prompt issue, and reran it until he got the take; knowing a broken prompt from a bad roll saves credits. (00:01:16)
• restate · How do you spend fewer credits? — The first bullet-time take was decent and its movement right, so Adil treated it as a batching game, not a prompt issue, and reran it until he got the take; knowing a broken prompt from a bad roll saves credits. (00:01:16)

Mentioned

Central: bullet time (concept) · Claude

Also: Hell Grind prompt adherence skill · rebatching

Only in the description or credits: Direct AI fight scenes through controlled iteration · Higgsfield Academy

← Previous: Direct a cinematic AI car commercial: Finish the story in the edit · Next → Seedance 4K: Cinematic Realism: Ads, Brands & Sharp Text