Methodology

Sourcing

Editorial copy is drawn from Mythopedia, Encyclopedia Britannica, the Oxford Dictionary of World Mythology, the British Museum ancient-civilizations catalog, the Hindu Scripture Archive, the Yoruba Cultural Heritage Center, the Poetic Edda, and seven internal editorial reports covering the canonical-100 deities, 200 names of the One God, 50 biblical theophanies, 200 Native American spirits, and the Ancient Near East pantheon adjacent to early Israel. Sources are listed on the sources page.

Image generation

Each prompt is authored under a fixed template — "<name>, <visual description>, bold crosshatching, aged parchment texture, museum-quality illustration" — and rendered by Higgsfield Seedream v4.5. Reroll batches (`_b`) are kept; canonical selection is the most-recent file by mtime. Every visitor sees the better-quality reroll where one exists.

Video generation

Five-second 720p loops are rendered by Kling against the same prompt as the still. Loops are then re-encoded to 480p H.264 (~250 KB) and VP9 (~280 KB) plus a JPEG poster (~80 KB) for the deity-card hover affordance. Loops are silent; audio narration is layered separately.

Audio narration

Each entry's museum-card narration ("<name>. <cleaned description>. From the <organization>; <time period>; <location>.") is synthesized by Microsoft's neural voice models via edge-tts. Default voice: en-GB-RyanNeural at rate −5 %. The voice can be changed at any time without regenerating images or videos.

Editorial-data integrity

The corpus passes through eight discrete pipeline stages:

1. build_asset_manifest.py — filter the Higgsfield job manifest to "hand-drawn ink" prompts, slugify, group by source-series.
2. triage_corpus.py — reclassify 549 natural-history prompts as Animism (era I) so the chronology starts where it should.
3. parse_named_deities.py — extract 1,050 named figures from seven editorial reports; match 943 (90 %) to image entries.
4. narrative_order_and_synonyms.py — recover narrative ordering from the prompt source files; surface 35 canonical-100 deities that exist in editorial copy but were never authored as prompts.
5. parse_theophany_scenes.py — match 48 of 50 biblical theophanies to image groups in canonical narrative order.
6. dedup_canonical.py — across 1,123 prompts with multiple variants, re-pick the canonical to be the newest reroll.
7. enrich_deity_metadata.py — clean prompt-craft tail, split name from description, attach organization + time period + location across ~50 culture buckets.
8. build_taxonomy.py — produce the tag namespace index for filter UI.

Current corpus

1,947 visual depictions, distributed:

• Animism — 549 entries (animal spirits, sacred landscapes, cosmic forces)
• Polytheism — 962 entries (named deities across global pantheons)
• Henotheism → Monotheism — 0 entries (theophany scenes)
• Monotheism — 388 entries (200 names of the One God + biblical scenes)
• Deified Humans — 48 entries (pharaohs, emperors, modern god-kings)

What's missing

93 authored prompts did not generate. 35 deities from the canonical-100 list (Atum, Ymir, Pangu, Chaos, Brahman, Ometeotl, Khonsu, Ranginui, …) have no illustration. The /contribute page is seeded with these gaps.

Open repository

Source code, build scripts, editorial JSON manifests, prompt templates: github.com/rocitreal/thehistoryofgod. Generated assets (images, videos, audio) are regenerable from sources — the repository is small by design, the archive is heavy.