THRONE OF AURAThe Open Playbook
Chapter 04

Video Generation

Judge what comes out of the black box.

  • Video generation providerHiggsfield
  • Video generation modelSeedance 2.5
Video Generation · 04.1

Preparing the prompt and references

Creator’s processPrompt framework
  • Attach exactly the references the prompt names, and check each label matches its file.
  • Each part of a split scene is its own generation: repeat the reference roles and style section in each.
  • Follow the tool’s current length limit.
  • Drop any reference-video wording when no video is attached.
Setup

Scenes were generated with Seedance 2.5 on Higgsfield, and the prompt framework in chapter 03 is written for it.

In Higgsfield, the prompt refers to each upload by slot: <<<image_1>>> for the first, and so on. A role reads like “Match <<<image_1>>> rendering across the entire frame.” See the full Higgsfield example.

Creator’s process
  • Aspect ratio: 4:3. Matches the PS3/Xbox 360-era cutscene look.
  • Resolution: 480p. Saves credits and still looks great.
  • Duration: flexible; shorter is better. Split longer scenes at meaningful beats (see timed shots).
  • Audio: Seedance generates dialogue and sound with the video, so the prompt carries exact dialogue and voice references.
Video Generation · 04.2

Reviewing results

Creator’s process

Each video was watched for pacing and coherence:

  • Did it make sense?
  • Did the camerawork flow well?
  • Was anything off, confusing or absurd?

Generations rarely land on the first pass. With several takes, the best parts can be stitched together.

Common failures

  • Characters in the wrong place
  • Garbled text
  • Incorrect motions
  • Wrong camera movement or placement
  • The goal of the scene misunderstood
  • Words mispronounced
Detail checks from the prompt frameworkA closer checklist for when something feels off.
  • Does the whole frame keep the game render, including the background, in every cut?
  • Is it the same character throughout?
  • Are screen sides and eyelines consistent?
  • Do hands stay on props, and do props stay where they were?
  • Is the right person speaking, with the exact words? Are internal-voice lips still?
  • Are concealed faces still concealed until their reveal?
  • Does the camera end where the shot said it would?
Video Generation · 04.3

Refining a take

Creator’s process

The retry process: match the produced video against the prompt.

  1. What went wrong, and where?Find the exact moment and the prompt line behind it.
  2. Tighten the direction, or cut it.Make the instruction clearer and more specific, or drop it if the scene works better without it.
  3. Generate again and keep the best parts.The strongest pieces of several takes can be stitched together in the edit.
Suggested

Where to look in the prompt for each failure:

  • Characters in the wrong place → world and staging.
  • Wrong camera movement or placement → camera language and that timed shot.
  • Incorrect motions → animation and performance.
  • Misunderstood goal → the opening sentence and the shot’s action line.
  • Look drifting towards photoreal → reference roles and rendering style.