Nodes/AkkiNodes LLM Suite: Your Personal AI Film Studio/AI ScriptCrafter 01 (Bible) v7.2 - Akki
ComfyUI Node

AI ScriptCrafter 01 (Bible) v7.2 - Akki

One call, and you have a character bible and a world bible

By routhakash·Created about a year ago·Updated 9 months ago· 8
AI ScriptCrafter 01 (Bible) v7.2 - Akki
  • llm_model
  • logline
  • theme
  • world_bible
  • character_bible
  • full_analysis_text
story_text
protagonist_name
protagonist_type
protagonist_age
antagonist_name
antagonist_type
antagonist_age
genre
period
max_tokens4096
temperature0.70
top_p0.95
top_k40
seed1235

The AI ScriptCrafter 01 (Bible) v7.2 is the foundation of the whole AkkiNodes pipeline. You give it a prose story plus a handful of "orthogonal" creative parameters, and it returns the two documents everything downstream depends on: the character bible and the world bible. Every later node - the beat sheet, the screenplay, the lookdev, the scene choreographer - reads these. Get this stage right and the rest of the pipeline has a solid ground truth to stay consistent with.

Why the "orthogonal" inputs matter

This is the pack's "Analyze & Refine" architecture, and the input design is deliberate. Instead of letting the model invent everything, you hand it the story plus explicit facts: protagonist_name, protagonist_type, protagonist_age, antagonist_name, antagonist_type, antagonist_age, genre, and period. The author's own bug log shows why: early versions had the LLM contradict itself on period and antagonist identity, and the fix was to feed those facts in separately rather than trust the story parser to get them right. So treat these fields as ground truth you're locking in - the model works around them, it doesn't decide them.

There's no prompt-file selector here; it's a single comprehensive LLM call followed by a deterministic Python post-processor that parses the output and merges duplicate or contradictory bible entries. The "refine" half of "Analyze & Refine" is pure code.

Inputs and outputs

Required: llm_model, story_text, the four protagonist/antagonist fields, genre, period, plus max_tokens (default 4096) and the sampling knobs (temperature 0.7).

Outputs:

  • logline and theme - the hook and the through-line. Useful for later stages and for keeping the story on-message.
  • world_bible - canon for the setting, rules, and locations.
  • character_bible - canon for who everyone is.
  • full_analysis_text - the complete raw analysis, for checking what the model actually said.

The two bibles are the ones that matter. Wire them into ScriptCrafter 02, and later into Character/Set Lookdev and the Scene Choreographer.

Install and practical notes

Same pack install as everything else: ComfyUI Manager search "AkkiNodes LLM Suite", or git clone https://github.com/routhakash/AkkiNodes-LLM-Suite-for-ComfyUI into custom_nodes; run install.bat as admin on Windows; GGUF into ComfyUI/models/llms/; restart.

Two practical notes. First, the bibles are only as canonical as your story_text and your model - if you feed in a half-formed story, you get a half-formed bible, and every downstream node inherits it. Second, a bigger max_tokens is your friend here; a truncated bible will quietly wreck character consistency two phases later. And if the antagonist keeps getting conflated with the protagonist, the first thing to check is that you actually filled in the separate antagonist fields rather than leaving them to the model.

CategoryAkkiNodes/ScriptCraft

Inputs (15)

NameTypeDefaultDescription
llm_modelLLM_MODEL
story_textSTRING
protagonist_nameSTRING
protagonist_typeSTRING
protagonist_ageSTRING
antagonist_nameSTRING
antagonist_typeSTRING
antagonist_ageSTRING
genreSTRING
periodSTRING
max_tokensINT4096256–16384
temperatureFLOAT0.70
top_pFLOAT0.95
top_kINT40
seedINT1235

Outputs (5)

NameTypeDescription
loglineSTRING
themeSTRING
world_bibleSTRING
character_bibleSTRING
full_analysis_textSTRING