Best Practices

Workflow patterns, AI writing guidance, and publishing preparation for getting the most out of LoreForge.

World Structure & Organization

Start with the skeleton, then flesh out

Create your major characters, locations, and factions as draft entries first. Give each a one-paragraph summary. Connect them with edges. Then go back and expand each one. This approach lets you see your world's shape before committing details, and it gives the AI much better context when you do ask it to expand entries.

Use status deliberately

Key rule: Only promote entries to Canon when you're confident in the content. Canon entries shape every AI generation. Contradictory canon entries will confuse the AI and produce inconsistent output.

Choose entry types with purpose

LoreForge has 11 built-in types plus custom types. Pick the type that best describes what the entry is, not what it relates to. A "The Great War" entry should be type Event, not Lore — even though it contains lore. The type affects how AI interprets and writes about the entry.

Tag consistently

Tags are free-form, but establish conventions early:

• Use lowercase, hyphenated tags: northern-realm, magic-user, book-1
• Use tags for cross-cutting themes that don't fit as entry types: villain, deceased, unreliable
• Use tags for story scope: book-1, arc-2, side-quest
• Use the Tags filter on the entries page to see all entries with a given tag — a fast way to check coverage

Use eras and regions

Eras and regions are optional metadata on entries, but they unlock powerful filtering on the timeline and map views. Define your eras early (even rough ones like "Ancient", "Modern", "Future") and assign entries to them. You can always rename or split eras later.

Writing Effective Entries

Use the body template sections

When you create an entry, LoreForge pre-fills section headings based on the entry type. Keep these headings — they give structure to the AI's context window and help it find relevant information faster. Fill in what you know and leave blank sections as placeholders.

Use wiki-links liberally

Type [[ in any entry body to create a wiki-link to another entry. Wiki-links do three things:

1. Create clickable navigation between entries
2. Appear as hover-preview cards in the markdown view
3. Help the Codex Sidebar in the chapter editor find referenced entries

Good practice: when you mention another entry by name in the body, always wiki-link it. This builds your knowledge graph organically.

Connect entries with edges

Edges are typed relationships between entries. Use the Edge Manager or the whiteboard to create them. Common patterns:

• Characters: ally_of, enemy_of, parent_of, mentor_of, member_of
• Locations: located_in, rules_over, borders
• Factions: allied_with, opposes, member_of
• Items: created_by, owned_by, located_in

Connected entries are automatically included as context when the AI generates content for any entry in the group. More connections = better AI output.

Keep entries focused

One entry per concept. If a city entry is growing to 3,000+ words, consider splitting it: the city itself, its government (faction), its notable locations, and its history (event or lore). Shorter, focused entries produce more targeted AI generations than sprawling ones.

Getting Better AI Generations

Write good user instructions

The "Additional instruction" field on every AI action is your most powerful lever. Specific instructions produce specific output. Compare:

Build context before generating

The AI's output quality is directly proportional to the quality of its context. Before running an Expand or Dialogue action on an entry:

1. Make sure the entry has edges connecting it to relevant entries
2. Promote related entries to Canon status
3. Fill in at least a paragraph of body text — don't run Expand on an empty entry

Use style samples

In Settings > World, paste 500–1,000 words of your own writing as "Style Samples." This is the single most effective way to make AI output match your voice. The AI uses these samples as few-shot examples for every creative action.

Iterate, don't accept first drafts

Treat every AI generation as a first draft, not a final product. The workflow should be:

1. Generate → Review → Accept or Reject
2. If accepted, rewrite and edit the output in your own voice
3. Run the quality checker (it scores every generation) — aim for 80+
4. Regenerate with different instructions if the output isn't what you wanted

Choose the right action

Writing Quality & Voice

Recognizing AI writing patterns

LoreForge includes a built-in writing quality system that catches common AI prose patterns. Every generation is scored 0–100. Here are the patterns it detects and why they matter:

Overused words

Words like delve, tapestry, palpable, visceral, nuanced, and multifaceted appear in AI writing at 10–100x the rate of human writing. Readers and editors have learned to spot them. Replace with simpler, more specific words.

Cliché phrases

Phrases like "a tapestry of emotions," "let out a breath they didn't know they were holding," and "tears pricked at her eyes" are near-universal in AI output and nearly absent in published fiction. If you see one, rewrite the entire sentence.

Named emotions

AI writes: "A wave of sadness washed over her." Good fiction shows the emotion through action: "She set the mug down wrong and it sloshed. She stared at the spill and didn't wipe it up." Never name an emotion when you can show it.

Fancy dialogue tags

AI loves murmured, breathed, intoned, mused, and quipped. Professional fiction uses said for 80% of dialogue tags, or drops the tag entirely and uses action beats instead. Let the dialogue itself convey tone.

Structural uniformity

AI writes paragraphs of nearly identical length (3–5 sentences each). Vary dramatically. Use one-word paragraphs. Use long, winding ones. The rhythm of your prose is part of your voice.

Using the quality checker

Every AI generation returns a quality score. You can also check any text manually via the POST /actions/quality-check endpoint. Scores mean:

Developing your world's voice

The most effective anti-AI-slop measure is having a strong authorial voice. Here's how to build one in LoreForge:

1. Write your style samples first. Before using any AI features, write 500–1,000 words of prose in your voice. Paste this into Settings > World > Style Samples.
2. Set your narrator persona. In Settings > World, define your narrator's identity: their perspective, limitations, biases, and vocabulary. A cynical bard writes differently than an imperial historian.
3. Choose tone and prose style. These shape every generation. "Gritty + sparse" produces very different output than "mythic + dense."
4. Write custom instructions. Add permanent guidance like: "Never use the word 'journey.' Characters in this world swear in Dwarvish. All magic has a physical cost."

Novel Planning Workflow

Recommended workflow

1. Build your world first. Create entries for major characters, locations, and factions. Connect them. Promote to canon.
2. Create a Novel Plan. Outline your story structure: acts, major beats, character arcs. The AI chat can help brainstorm.
3. Define Story Threads. Create threads for each subplot, character arc, or thematic line. Link thread items to entries.
4. Plan chapters. Use the Chapter editor to outline each chapter with beat lists. Assign threads to chapters.
5. Write in the Chapter editor. The Codex Sidebar auto-detects wiki-links and entry references in your text, giving you instant access to canon context while writing.
6. Track progress. Set writing goals (daily/weekly/monthly word counts) and log sessions. The Writing Stats page shows streaks and progress charts.

Writing mode

Use the distraction-free writing mode (toggle in the chapter editor) for focused drafting. It hides the sidebar and reduces UI chrome. Combined with typewriter scroll mode, it keeps your current line at a comfortable screen position.

Collaboration Patterns

Role-based editing

Using Lore Notes for feedback

Viewers can't edit entries, but they can submit Lore Notes — tagged comments with types like "question," "suggestion," and "contradiction." This is ideal for beta reader feedback: they read your world and leave structured notes without risk of accidentally changing canon.

Real-time collaboration

LoreForge supports real-time collaborative editing on entry bodies. When multiple people edit the same entry simultaneously, changes sync via WebSocket. A green "Live" indicator appears in the editor showing how many people are connected. Save regularly — the WebSocket syncs edits in real-time, but the database save is still manual (Ctrl+S).

Maintaining Consistency

Run consistency checks regularly

Use the Contradictions action on entries that have been heavily edited or expanded. Run the Continuity Report (available from the sidebar) to scan your entire world for relationship gaps and contradictions.

Canon review workflow

1. Write and expand entries in Draft status
2. When satisfied, review the entry for internal consistency
3. Run Contradictions check against existing canon
4. Fix any issues found
5. Promote to Canon

This prevents half-formed ideas from contaminating your canon context and producing confused AI output.

Edge label conventions

Be consistent with edge labels. If you use "ally_of" in one place and "allied_with" in another, the AI treats them as different relationship types. Stick to the built-in labels when possible. Create custom labels only for relationships that don't fit any existing category.

Working With Large Worlds (500+ Entries)

Use eras and regions as organizational layers

When your world grows beyond a few hundred entries, eras and regions become essential filters. Assign every entry to at least one. This lets you scope the timeline view, map view, and search results to relevant subsets.

Tag for scope

Use tags like book-1, arc-2, main-cast to filter entries by story relevance. When working on a specific section, use search and tag filters to focus the entries list.

Archive liberally

Entries you've outgrown but don't want to delete should be archived. Archived entries don't appear in default views or AI context, reducing noise without losing the content.

Whiteboard for visual mapping

Use the whiteboard to lay out relationship clusters visually. The board templates (Hub & Spoke, Timeline, Hierarchy, Clusters, Circle) help you see different structural views of the same entries. Export as PNG for reference.

Publishing Preparation

Pre-publication checklist

LoreForge includes a built-in Publication Checklist (accessible from the sidebar under "Publish") that walks you through 16 items across four categories: Legal & Copyright, Platform Requirements, Writing Quality, and Publication Prep. Use it before every release.

Export workflow

1. Run the quality checker on all chapters — aim for 80+ on every one
2. Run a full consistency check across the world
3. Export in your target format (PDF, EPUB, or DOCX)
4. Review the exported file in a reader/viewer — formatting can shift
5. Export a JSON backup of the entire world as your archive

Copyright page template

The Publication Checklist includes a copyright page template with an anti-AI-training notice. Copy it and customize it for your book. The key clause is:

"Without in any way limiting the author's exclusive rights under copyright, any use of this publication to 'train' generative artificial intelligence (AI) technologies to generate text is expressly prohibited."

Legal Considerations for AI-Assisted Writing

Can you copyright AI-assisted writing?

Yes, with conditions. The US Copyright Office (January 2025, Part 2 Report) confirmed:

• Works generated solely by AI are not copyrightable
• AI prompts alone are insufficient for authorship
• AI as an assistive tool IS copyrightable — when the human controls the expressive elements (writing, editing, selecting, arranging)
• Human modifications to AI output are copyrightable

The Supreme Court declined to review the Thaler v. Perlmutter ruling in March 2026, confirming that human authorship is required for copyright protection under current US law.

How LoreForge fits

LoreForge is a writing tool where the human author creates characters, plots, lore, and prose. AI assists with brainstorming, expanding, editing, and consistency checking. This falls squarely in the "AI as assistive tool" category that the Copyright Office treats favorably. Your strongest protection comes from:

1. Writing your own drafts and using AI for expansion and editing
2. Substantially rewriting every AI-generated passage in your own voice
3. Documenting your creative process — LoreForge's revision history and generation records serve as evidence of your human authorship

Protecting your work

1. Register copyright with the US Copyright Office ($65 online at copyright.gov). Describe your human authorship contribution.
2. Keep process documentation. Your LoreForge project — with its revision history, generation records, and editing trail — is exactly the kind of evidence that strengthens copyright claims.
3. Add the anti-AI-training notice to your copyright page.
4. Never list AI as a co-author. You are the author. AI is a tool, like a spell-checker or thesaurus.

Platform Policies

Amazon KDP

• Disclosure required: Select "AI-generated" if AI produced any final text or images (even if edited). Select "AI-assisted" if AI was only used for brainstorming or grammar (no disclosure shown to buyers).
• 3 titles/day limit for self-published authors
• Penalty for non-disclosure: Book removal and potential account suspension

Barnes & Noble Press

• No specific AI disclosure checkbox, but all content must comply with their Content Policy and IP requirements
• B&N has conducted purges of low-quality AI-generated titles

Kobo, Apple Books, Smashwords

• Less explicit AI-specific policies than Amazon
• General content quality and IP compliance terms apply
• Apple requires disclosure for AI-narrated audiobooks

European Union (August 2026+)

• The EU AI Act requires AI-generated content to be labeled when distributed to inform the public
• A Code of Practice on AI labeling is being finalized (expected June 2026)
• If distributing in the EU, prepare to label AI-generated content per the forthcoming standard

Self-Hosting Tips

Model selection

For the best writing quality with self-hosted Ollama:

• Writer model: qwen3:30b-a3b for a good balance of quality and speed. Upgrade to qwen2.5:32b if your GPU supports it (16GB+ VRAM).
• Utility model: qwen3:14b for fast operations (summarize, contradictions, connections)
• Cloud providers (Anthropic, OpenAI) generally produce higher quality than local models for creative writing, but cost money per generation

Bring Your Own Key (BYOK)

Writer and Studio tier users can enter their own API keys in Settings > AI Provider. This gives unlimited generations at your own API cost, which is often cheaper than per-generation pricing for heavy users.

Backup strategy

• Self-hosted: Export your world as JSON regularly (Export page). The SQLite database file is at data/loreforge.db.
• Cloud: Automatic backups via the backup cron. You can also export JSON anytime.
• Use the scripts/backup.sh script for automated PostgreSQL backups if running the cloud stack.