Direct AI fight scenes through controlled iteration: Direct scale and transformation with measurements

2026-08-10 · Academy lesson · Adil · recording, 3:52 · original

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

This academy lesson breaks down two prompts for scale and transformation shots. The first raises an army in fog. It is built in three layers, with the action timed second by second and a frozen finale, and one big batch gave the two clips used. The second arcs the camera around Roko and gives his transformation in exact numbers.

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 army shot: scale in AI filmmaking against traditional production
• 00:00:26 — Anatomy of the army prompt: fog, three layers, action second by second, the frozen finale
• 00:01:59 — Production note: one big batch, two clips used
• 00:02:10 — The army shot played
• 00:02:32 — The camera arc around Roko and the transformation in numbers
• 00:03:30 — The transformation shot played

Threads in this video

• restate · Which LLM should be used to write video prompts? — For the army shot Adil gives Claude a quick text description through the same prompt refinement skill, and Claude returns the final prompt. (00:00:26)
• refine · How do you make a large crowd or army in a generated shot? — Adil puts the army scene in thick fog that caps visibility at 20 meters on purpose: at that scale the model glitches in the background, and the fog hides those mistakes. (00:00:26)
• restate · How do you make a large crowd or army in a generated shot? — The army prompt never says "show a thousand samurai": it builds the frame in three layers (one sharp warrior up front, a dense crowd behind him, silhouettes fading into the mist), so only about 40 soldiers are clear and the brain fills in the rest. (00:01:08)
• restate · How do you get precise motion (dance, action, choreography)? — Adil writes the action second by second: the foreground samurai's rise goes in half-second steps (hand at 0.5, helmet at 1.5, standing by 3), and with the timing locked the model cannot rush it or skip stages. (00:01:08)
• restate · When a generation fails, should you generate more takes, fix the prompt or start over? — Once the army prompt clicked, the team ran a big batch, got several great takes and used two different clips from that one batch in the final edit. (00:01:59)
• restate · How do you prompt a sequence of camera moves in one shot? — For the arc around Roko, Adil writes the camera orbit in raw numbers (8 meters down to 4, 3 meters high down to 2), because video models prefer arithmetic over vague words like "dramatically". (00:02:32)
• refine · How do you make a transformation shot? — Roko's transformation gets a hard number, 0.4 seconds, and the prompt calls it a weapon deploying, not a magic sequence: Adil says a transformation left to take its time becomes a glowing anime power-up every time, and the instant snap keeps it real. (00:02:32)

Mentioned

Central: Claude · Roko

Also: Hell Grind prompt adherence skill · rebatching · Seedance

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

Also centred on Roko: Hell Grind: Rebuilding Our Biggest Fight Scene from Scratch

← Previous: Build an AI Ad Agency with Claude + Higgsfield: Design the client intake · Next → Add AI VFX to Real Footage: Set your head on fire