Direct AI fight scenes through controlled iteration: Diagnose and repair a complex shot

2026-08-10 · Academy lesson · Adil · recording, 4:23 · original

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

In the diagnose-and-repair lesson of Direct AI fight scenes through controlled iteration, Adil takes a one-take battle shot with Roko through three iterations in Seedance 2.0 in Cinema Studio: character tags, an environment block and a speed comparison in the prompt, character classes against a clone army, and asking the monster to raise his arm instead of a negative prompt.

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 one-take battle shot and the anatomy of its first prompt: character tags, the environment block, the camera speed comparison
• 00:01:23 — The first batch of four in Cinema Studio with Seedance 2.0: glitchy emergence, a clone army, a flat camera path
• 00:02:42 — Second and third iterations: the close-up opening, the monster's arm and the negative-prompt trap, the playbook

Threads in this video

• restate · How do you keep a location, product or prop the same across shots? — The prompt's environment and style block locks weather, fog density, ground details and lighting; Adil says keeping these rules active across every frame stops the world changing mid-shot. (00:00:29)
• restate · How do you prompt a sequence of camera moves in one shot? — For the drone move the prompt asks for an FPV drone flying backward in a straight line at three times the speed of a typical cinematic dolly pull; Adil says models understand direct comparisons, not big descriptions. (00:00:58)
• restate · When a generation fails, should you generate more takes, fix the prompt or start over? — Adil says the shot took many iterations and detailed prompting, and the real skill of the workflow is not writing one genius prompt but reading what broke and fixing only that. (00:03:39)
• restate · How do you make a large crowd or army in a generated shot? — Adil's fix for a clone army: do not just ask for a crowd but build distinct character classes for the extras; his two, a heavy Oni-masked brute and a light scout, give silhouette variety and keep the same warriors across generations. (00:02:00)
• restate · Should a prompt list what to avoid? — Adil says telling the model what not to do rarely works and negative prompting was the wrong play; the fix for the monster's arm was simply to ask the model to raise the arm so it actually sees it. (00:03:10)
• restate · Which LLM should be used to write video prompts? — Adil's playbook for each shot: look at the generation, name what is broken in simple words and let Claude rebuild the prompt. (00:03:39)

Mentioned

Central: Claude · Seedance 2.0

Also: Cinema Studio · Elements · negative prompt · rebatching · Roko

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

← Previous: 7 Best AI Avatar Generators for Talking Videos in 2026 (Tested and Compared) · Next → Blockbuster 4K: The AI Filmmaking Pipeline: Scene 2: The Desert Chase