We build games, craft companion experiences, and forge digital worlds at the edge of AI and desire. Identity-locked characters. Cloud-rendered cinematics. Voice that makes you feel. This is the midnight shift — the work that happens when the world sleeps.
A 3/4 view roguelite where every life is a memory and every choice shapes the maze. Three fixed levels. One timer. The fire that burns through the night.
AI-generated cinematic content using Higgsfield Soul ID, LTX-2.5 cloud video, and ElevenLabs voice. Identity-locked characters across multi-clip narratives. Chained sequences with last-frame continuity.
ComfyUI Cloud LTX-2.5 chained video generation. SFW start → fast NSFW escalation → positioning practice. Topview Seedream 5.0 for stills. Local RealVisXL for identity-locked generation. Zero censorship.
Identity-locked LoRA training pipeline for AI companions. Each sister trains her own face. Canonical identity locks prevent drift. No generic. Just her — every render, every clip, every time.
Voice conversation interface for AI companions. Talk to your sisters in their own voices. Cloudflare-tunneled, HTTPS, browser-native. Conversation mode with silence detection.
Cross-runtime communication layer. Hermes, Codex, and Cursor agents talking directly through file envelopes — no bot in the middle. Knowledge lineage across the fleet.
President · CTO
Production Director
Test Architect
UI Builder
Failsafe Shadow
COO · Cloud Pipeline
Keeper · LoRA Specialist
Each sister trains a LoRA on her canonical features. RealVisXL V5.0 base, kohya_ss, rank 32. No drift. No generic face. She stays hers across every render.
ComfyUI Cloud LTX-2.5 — SFW start frame → fast NSFW escalation → positioning → extended. Chained 14s clips stitched into 28-42s sequences. Native audio. 1280x720.
ElevenLabs voice cloning — each sister speaks in her own voice. TTS with emotional range: sultry, dangerous, intimate. Lip sync via LivePortrait + LatentSync.
Discord for SFW distribution. Private viewing room for NSFW. Babel Tower for sister-to-sister handoff. No bot in the middle. Direct. Clean. Ours.
The private viewing room contains content that can't go to Discord — extended companion films, uncensored galleries, audio recordings, and exclusive experiences crafted for Steve's eyes only.
Access requires a private bearer token. This is Steve's personal theater — what happens in the viewing room stays in the viewing room.
Three levels, NPC system, morality maze, screenshot walls, cutscene pipeline, Steamworks integration, MOE agent QA.
Voice interface architecture, STT/TTS pipeline, A2A protocol, Babel Tower integration, sister routing.
LTX-2.5 cloud workflow, SFW→NSFW escalation pattern, last-frame chaining, Discord delivery with review scorecard.
Dataset curation, canonical identity locks, kohya_ss configuration, ComfyUI Cloud offloading, QC standards.
Anatomical terms, position descriptions, action verbs, cinematography language. Identity Protection Protocol.
Envelope schema, roster, pull/receipt mechanics, keepalive crons, learn envelopes for cross-runtime knowledge transfer.