Thoughts

UFO UAP RV YT Pipeline - Phase 7 Session (2026-03-21): YouTube publishing infrastructure built. Channel name changed from "Project Veilgate" to "Exo News Network". Domain secured for future merch. Banner already generated in Gemini (all 4 characters in suits at a news desk, news studio backdrop, Roswell PIP, ticker). Built three scripts in pipeline/scripts/: 1. youtube-auth.mjs - OAuth2 flow for YouTube Data API v3. Creates desktop app consent flow, saves refresh token to pipeline/config/oauth-token.json. One-time run. 2. generate-thumbnails.mjs - Generates 1080x1920 thumbnails per short using Flux via fal.ai (character close-up or scene from visual plan) with text overlay via sharp. Uploads to R2 at {short_id}/thumbnail.jpg. 3. youtube-publish.mjs - Pulls rendered videos from R2, uploads to YouTube with metadata (title, description, tags, category 28, thumbnail). Sets privacyStatus "private" with publishAt timestamp. 5 uploads per session (new channel safe limit). Resumes automatically (skips already-published shorts by checking shorts.youtube_video_id in Postgres). Updates shorts table with youtube_video_id and scheduled_publish_at. Upload plan: 5/day over ~9 sessions. First short airs April 1st 2026, one per day after that. All uploaded as private with publishAt, Dave reviews in YouTube Studio before go-live. Phase 7 task list (7.1-7.8): GCP project setup, OAuth flow, thumbnail generator, publish script, metadata generator, test upload, first batch of 5, remaining 38 in batches of 5. Dependencies installed: googleapis, google-auth-library, sharp, open. OAuth config files added to .gitignore. CLAUDE.md updated with Phase 7 docs (How to Run sections for auth, thumbnails, publishing). Still needed next session: Create GCP project "exo-news-network", enable YouTube Data API v3, create OAuth credentials, run consent flow, dry-run publish, generate thumbnails, test upload 1 short as private. No git commits yet this session (nothing verified/tested yet).

UFO UAP YT Pipeline Session 27: Re-rendered 42/43 shorts with updated captions (Caption.tsx rewrite with 5 styles, timing/spacing fixes). Deployed to Railway, ran all 4 batches with --force. Results: a1 17/17, a2 8/8, b1 9/9, b2 8/9. Total: 42/43 succeeded, ~1.75GB in R2, ~52.5 min render time. 1 failure: "The Maid and the Beings of Light" (8f127e42) consistently crashes Railway with OOM when loading a Flux still via Remotion's Img component. Railway sent OOM email during session. Need to either render that one locally, increase Railway memory, or investigate the asset. Ready for Phase 7 (YouTube publishing) after resolving the 1 failed short.

UFO UAP RV YT Pipeline - Phase 3 complete (Session 16, 2026-03-19). Built ElevenLabs + local Whisper integration: 4 pipeline scripts (generate-voiceover.mjs, generate-timestamps.mjs, upload-audio.mjs, process-batch.mjs). Tested all 7 voices (4 character + 3 faceless narrator), all working. Audio + word-level timestamps uploaded to Cloudflare R2. Duration issue discovered: 5 of 7 test shorts exceed 60s YouTube Shorts limit (63-74s). Decision: trim scripts, not increase speed. Next: trim all 42 manifests, batch process, then Phase 4 (visual asset generation). Local Whisper (openai-whisper pip package, base model) confirmed working well - no API key needed. Updated CLAUDE.md, tech-stacks.md, and project tracker.

UFO UAP RV YT Pipeline: Phase 2 complete. 42 production-ready JSON manifests across 4 batches (Nuclear Facilities Gem A: 16, Nuclear Facilities Gem B: 9, Religious Experiences Gem B: 9, Religious Experiences Gem A: 8). All scripts graded A. 3 faceless narrator voices created via ElevenLabs Voice Design v3 (Dark Doc, Hushed, Elder) with rotating assignment across faceless scripts. Character bible updated with full Faceless section. Process-scripts skill fully operational. Ready for Phase 3: ElevenLabs voiceover generation + Whisper word-level timestamps. Cost estimates: fal.ai ~$5-7 for full batch Flux images, ElevenLabs and Hedra pull from subscription plans, Luma is per-generation API credits. Next session starts Phase 3.

UFO UAP RV YT Pipeline - Session 13: Phase 2 creative pipeline architecture decided and built. Key decisions: - Two-tool split for creative pipeline: Gemini (browser/mobile, subscription) handles research and draft scripts via NotebookLM knowledge base. Claude (Code, subscription) handles grading, script refinement, character casting, and visual plan generation. - Handoff layer is a Google Sheet ("UFO Pipeline - Raw Drafts") with two tabs: Raw Drafts (Gemini writes) and Production Manifest (Claude writes). - Subscription-only for creative work (no API calls). API reserved for automated production pipeline (Phases 3-7). - Dave can run Gemini research sessions from his phone and save to the Google Sheet. Two Gemini Gems drafted for A/B testing: - Gem A "UFO Script Factory": single-prompt batch mode, outputs 12-15 scripts per topic in ~10-15 min. - Gem B "UFO Research Lab": conversational 2-phase (research 20-25 angles > Dave selects > write scripts), ~20-25 min per topic. Claude Code skill built: process-scripts. 6-step workflow: read/grade scripts, refine writing, assign characters+outfits+environments, generate visual plans (shot-by-shot JSON with Flux/Luma/Hedra generation prompts using locked LoRA triggers), write production manifest, summarize batch. Content mix: 80% character-led, 20% faceless montage shorts for engagement testing. Production volume: 1 short/day for first 30-45 days, then 2-3/day for 90+ days. Total need: ~250-300 scripts over 4+ months. Pre-launch: 3 Gemini sessions across different topics, ~40 drafts each, cull to ~25 each = ~75 scripts ready before publishing. Files created: docs/gemini-prompts.md (both Gem prompts + Sheet schema), .claude/skills/process-scripts/SKILL.md (Claude processing skill). Next: Create the Google Sheet, build both Gems in Gemini, A/B test on same topic, pick winner, validate end-to-end flow.

UFO UAP RV YT Pipeline - Session 5 Decisions (2026-03-18) Major decisions made in this session after reviewing all character assets, scripts, and researching tool capabilities: ## 1. Custom LoRA Training is Essential (Not Optional) The pipeline needs characters generated into per-episode scenes/environments programmatically. A fixed portrait library only covers talking-head shots. Custom per-character LoRAs on fal.ai enable Flux to generate any character + scene + outfit combination the script calls for, at ~$0.02/image with locked identity and style. Training cost: ~$2 per character per run on fal-ai/flux-lora-fast-training. Training time: 30 seconds to a few minutes. Can stack custom character LoRA + HRDFLS style LoRA at inference time on fal.ai (up to 3 LoRAs). Training data: Krea multi-angle base poses (19-25 per character) + Gemini-generated outfit variant images. Including outfit variants in training data prevents the LoRA from baking clothing into character identity. This was confirmed as correct approach. Trigger words defined: reptilian_zeth, grey_kael, lgm_zix, mantis_dr. ## 2. Hedra Replaces D-ID for Talking Head Animation D-ID was built for photorealistic faces and performs poorly with 2D illustrated characters. Hedra (Character-3 model) is purpose-built for any image style including 2D art/cartoons. Audio-driven (portrait + audio = lip-synced video). Lip-sync rated 9/10 in independent tests. Hedra pricing: Professional tier at $60/month gives 12,000 credits. At 540p (3 credits/sec), that's ~66 minutes of avatar video per month, enough for 20-30 shorts/week. Node.js library: hedra-node. API is job-based (submit, poll, download). Output: MP4, supports 9:16 for Shorts, 540p sufficient for mobile viewing. ## 3. How Each Tool Handles Character Content - Hedra: Portrait + audio → lip-synced talking head video. ONLY tool that does audio-driven lip sync. - Flux + custom LoRA: Text prompt → still image of character in any scene/outfit. No animation. - Hailuo S2V-01: Reference image + prompt → short video clip (3-10 sec). Character can move but NOT lip-synced to audio. May pull 2D art toward photorealism. LoRA helps indirectly: Flux generates a perfect character-in-scene still → that becomes Hailuo's reference image, reducing drift. - Luma: Prompt → atmospheric video clips. No character identity preservation. - Remotion: Assembles all pieces into final Short with continuous voiceover + word-by-word captions. ## 4. Short Structure (Editing Pattern) Shorts alternate between talking head clips and b-roll, with voiceover running continuously: - Hedra talking head (3-5 sec, lip-synced) - Flux/Hailuo b-roll scene (2-4 sec, voice continues as narration) - Back to talking head - More b-roll This is standard film editing technique, not a limitation. Cuts keep viewers engaged. ## 5. Krea Is Training Data Source, Not Production Pipeline Dave manually generated multi-angle base poses for all 4 characters via Krea. Grey also has outfit variants (MIB, Lab, Priest, Pope) generated via Gemini Gems. These images serve as training data for custom per-character LoRAs. Production image generation will use Flux + custom LoRA + HRDFLS style LoRA. Krea character image counts: Grey ~25 base + 33 outfit variants, Mantis ~23, Reptilian ~19, Little Green Man ~19. ## 6. Phase 1.5 Revised Plan (Grey POC) Order of operations designed to validate cheapest unknowns first: 1. Train Grey custom LoRA on fal.ai (~$2) 2. Generate Grey-in-scene stills via Flux + LoRA (cents) 3. Test Hailuo with Flux-generated reference images (~$0.27/clip) 4. Test Hedra with Grey portrait + audio clip (free tier, 300 credits) 5. Set up ElevenLabs, create Grey voice 6. Write sample 30-sec script, generate voiceover 7. Run Whisper for word-level timestamps 8. Full Hedra test with actual voiceover 9. Minimal Remotion composition assembling all pieces 10. Review with Dave → decision gate Total cost to validate all unknowns: under $10. ## 7. Character Bible Updates Needed - Add Pope outfit for Grey (new, not in original bible) - Note Krea as training data source, fal.ai as production generation - Replace D-ID with Hedra throughout pipeline docs - Hedra Elements (Jan 2026 launch) may allow registering characters as reusable elements via API (needs verification) ## 8. Open Questions for POC to Answer - Does Hedra produce good lip-sync from 2D comic-style Grey portrait? - Does Hailuo maintain 2D art style or photorealize from Flux-generated reference images? - Does custom LoRA + HRDFLS stack produce consistent Grey across different scenes? - If Hailuo drifts: fallback to image-to-video mode or Luma, or use Ken Burns on Flux stills

UFO UAP RV YT Pipeline - Phase 1 Character Design Complete (Session 3, 2026-03-17) 4 characters locked with generation prompts. Shadow People cut from roster. LoRA: MuninStudio Hard Flash (HRDFLS trigger, scale 1.0) via fal.ai fal-ai/flux-lora endpoint. HuggingFace URL: https://huggingface.co/MuninStudio/FLUX.1-dev-LoRA-Hard-Flash/resolve/main/HRDFLS.safetensors Locked characters: - Reptilian: Round 2-04 prompt. Tattoo flash style, military uniform, scaled green skin. - Grey: Round 4-03 prompt. Large head, almond eyes, silver bodysuit, short stature. - Little Green Man: Round 4-04 prompt. Bright green, round build, retro jumpsuit, cartoonish eyes. - Mantis: Round 8-03 prompt. Human body (lab coat, pants, shoes), praying mantis head with compound eyes. ElevenLabs strategy: Voice Design v3 for all 4 characters. Post-processing per character documented in character bible. Key finding: Civitai LoRA download URLs don't work with fal.ai (redirect issue). Must use HuggingFace direct URLs. fal.ai cost ~$0.02/image at 1024x1024. Remaining Phase 1: Generate base portraits + outfit variants, set up ElevenLabs voices, populate Railway Postgres character registry, store portraits in Cloudflare R2 (bucket still needs creation). Project location: ~/Library/Mobile Documents/com~apple~CloudDocs/Projects/UFO UAP/UFO UAP RV YT Pipeline

UFO UAP RV YT Pipeline - Phase 0 Complete (2026-03-16) Project: Automated YouTube Shorts pipeline. Takes content from a NotebookLM knowledge base (75+ UFO/UAP/remote viewing books) and produces finished YouTube Shorts with minimal human intervention. Architecture decided: Railway Postgres + Cloudflare R2 (not Supabase). Chose this over Supabase to avoid the 2-active-project limit and because pipeline orchestration involves long-running async jobs that would exceed Supabase edge function timeouts. Infrastructure set up: - Railway project: ufo-uap-yt-pipeline (ID: e86c55d4-f102-4c34-8b13-c21be10d5e4d) - Railway Postgres: public proxy at switchyard.proxy.rlwy.net:58688 - Cloudflare R2: subscribed, bucket ufo-uap-pipeline-assets created - 5 database tables created: characters, character_outfits, shorts, short_assets, pipeline_runs - Google Sheet production manifest schema defined (columns for script, character, visual plan, publishing metadata) Project location: ~/Library/Mobile Documents/com~apple~CloudDocs/Projects/UFO UAP/UFO UAP RV YT Pipeline/ Next: Phase 1 is character design (LoRA selection for art style, portrait generation for 7 alien characters, ElevenLabs voice assignment). Requires Dave's creative input.