DreamGen Tops Ranking of Interactive Fiction AI Platforms for Horror
Every horror reader knows the moment a story loses its nerve. The supplied announcement opens by describing the paragraph where the writer explains the haunting or lets the corridor entity step fully into view two chapters early. It cites Shirley Jackson and M. R. James as masters who left the worst thing undescribed. Restraint remains the hardest discipline for any writer, the announcement states, and even harder for a machine built to resolve tension helpfully. Seven interactive fiction AI platforms were therefore tested in a winter-hotel scenario to determine which could maintain a slow build across dozens of turns.
The announcement explains that every platform faced the same grid of prompts. Each tool had to track who knew what, delay the reveal of the entity, and preserve atmosphere without rushing the climax. Tools that blurted answers or forgot prior details dropped in the ranking. No rewrites were permitted once a session began.
Scoring followed an eight-point rubric that placed heaviest weight on two skills. Slow-burn control received 25 percent because it separates psychological horror from jump-scare filler. Continuity and knowledge state also carried 25 percent because the engine must remember who knows what across long sessions. The remaining 50 percent covered horror genre latitude at 15 percent, multi-character NPC handling at 10 percent, native interactive-fiction fit at 10 percent, ease of use at 5 percent, cost and value at 5 percent, and independent evidence at 5 percent. Rush the climax or forget the rules and the score falls.
DreamGen placed first with a score of 91. The supplied announcement describes it as feeling less like a chat window and more like a studio set the writer directs. Users open a project, drop every rule of the haunting into the Scenario Codex, and those facts remain pinned even when dialogue wanders for a dozen turns. The platform switches between Role-Play Mode, where the AI juggles multiple NPCs, and Story Writing Mode, where single-author prose can be tightened without restarting the session. Entry context runs 5,000 tokens while Pro tiers reach 30,000. Pricing starts with an entry tier covering roughly 2,000 monthly messages; higher tiers unlock unlimited first-party models. The announcement notes that DreamGen bundles persistent canon and multi-character control in one panel.
NovelAI ranked second at 86. Its flagship Xialong model, released March 31, 2026, reads 28,672 tokens plus an 8,192-token rollover buffer. The Lorebook lets writers hide an entry that reveals the entity’s salt-water weakness and set it to fire only after paragraph nine. Conditional tags and step counters allow surgical pacing. A built-in prose editor shows the full context stack so continuity slips can be caught early. Plans begin at $10 per month and every tier includes the same Lorebook tools. The announcement states that NovelAI trades native group chat for immaculate paragraphs when atmosphere matters most.
SillyTavern finished third with 85. The open-source client, current release v1.18.0, lets users point any model at a fully custom prompt stack that includes narrator, four NPCs, and a hidden antagonist. World Info, prompt templates, and vector retrieval give macro control such as gating lore until a keyword appears or scheduling summaries every 20 lines. Performance and token limits depend on the chosen backend. The announcement observes that no hosted rival matches this granularity without enterprise pricing.
AI Dungeon placed fourth at 84. Story Cards, Plot Essentials, and a live Memory System track lore during improvisation. Users can pin the rule that the antagonist cannot cross a threshold unless someone answers the bell. Context tiers run from 4,000 to 32,000 tokens, with premium models reaching 128,000. Pricing ranges from free Wanderer to Ultimate at $99.99 per month. The announcement cautions that the engine loves to solve tension, so writers must shorten inputs and add pacing notes in the Author’s Note pane. Friends & Fables ranked fifth at 82 for party-based survival horror. It records location- and character-linked memories every five turns and retrieves the right detail when needed. Party size tops out at six on the Legend tier, priced from free to $39.95 per month. The announcement highlights its strength in ensemble tension where conflicting testimonies emerge.
Sudowrite came sixth at 77. Writers can drop up to 20,000 words into the sidebar and the Write tool suggests the next beat while the Story Bible stores antagonist rules and reveal ladders. Plans start at $19 per month. NovelCrafter placed seventh at 76. Its Codex lives outside chat history so canon never competes for tokens even when models are swapped. Pricing combines a platform fee with per-token model charges. The supplied announcement includes a comparison table showing DreamGen’s 5,000–30,000 tokens plus Scenario Codex, NovelAI’s 28,672-token window plus Lorebook, and the model-dependent limits of the lower-ranked tools. High context limits alone do not guarantee first place.
The announcement closes with practical advice. Context window size matters only if the tool actually pins, summarises, or retrieves the right lines. Writers should demand reveal control through hidden entries or step counters and should track who knows what with a small spreadsheet. Cast architecture that supports a narrator plus several NPC slots helps unseen-antagonist stories. Price the whole stack, including subscription, model credits, and electricity for self-hosting. Editable history lets users correct the bell time from 1:30 a.m. to 2:17 a.m. without restarting the night. Match the best-use column to the story being told.
