AI Scene Choreographer (Bible) v4.3 - Akki

If you've run the full AkkiNodes pipeline, this is the payoff: AI Scene Choreographer (Bible) v4.3 is the node that finally turns all that accumulated production data into the actual text prompts you'll render. Pick a scene number, feed it your production CSV and the lookdev prompts for your sets and characters, and it generates ready-to-render image prompts for every shot in that scene - consistent with the art direction you established in Phase 3.

How it works

Two LLM stages, which the README calls the "Director + Promptsmith pipeline":

1. Director (stage 1) - reasons about the scene and figures out what each shot needs: framing, action, which characters and sets are involved.
2. Promptsmith (stage 2) - assembles the final prompts, weaving in the lookdev bible data so a shot of the tavern actually matches the tavern you art-directed, and the character actually matches the character bible.

The v4.3 changelog matters here: it made asset matching case-insensitive to fix a "context bleed bug" where set/character data leaked between scenes. If you've ever had a desert character show up in your snow scene, that's the class of bug this version was fighting.

Inputs that matter

• llm_model - the LLM_MODEL from a loader.
• csv_report - the production shot list CSV (from the Phase 2 parser).
• scene_number - which scene to choreograph (default 1).
• set_names_STRING, set_prompts_STRING, character_names_STRING, character_prompts_STRING - these come straight from the Lookdev Bible Loader, which has STRING outputs built specifically for this node.
• prompt_director and prompt_promptsmith - prompt-file selectors, one file each in the pack.
• temperature 0.5, max_tokens 4096.

Outputs and wiring

• shot_names_LIST and final_shot_prompts_LIST - parallel lists of names and prompts. The shot prompts are the money output; feed them into a text encoder or loop over them with a batching node.
• scene_name, shot_count - handy for loops and file naming.
• choreography_text_DEBUG - the raw director/choreography text, for when a prompt looks wrong and you want to see the reasoning.

If you're using the Generic File Saver to write each prompt to disk, pair it with the shot names and a loop.

Install and gotchas

Standard for the pack: 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; drop a GGUF into ComfyUI/models/llms/; restart.

The trap beginners hit: this node expects the string lookdev outputs, not the lists, and expects the names and prompts to be in matching order. If your prompts come out with the wrong character attached, check that set_names_STRING/set_prompts_STRING come from the same loader run. And as always with this pack, your GGUF is the ceiling - a small quant will produce prompts that ignore the bibles and drift toward generic. It's the last node before rendering, so it's worth feeding it a decent model.