Follower-first content machine

Growth lane

AI domestic serials, comment-court posts, Photo Mode quizzes.

Trust lane

Realistic proof demos: product visible, one tactile payoff.

Identity

Mara/persona posts: “I test cheap home gadgets so you don’t buy junk.”

Make these first

Warm but not spammy

• Days 1–3: 1/day. Safest broad hook, watch for suppression.
• Days 4–7: 2/day. One serial + one proof/comment-court.
• Days 8–14: 2–3/day only if distribution is healthy.
• Scale: comments asking for next part or strong follows per 1k views.

Mara Vale identity lock

Promise: I test cheap home gadgets so you don’t buy junk.

Use as PFP/reference face. Keep dark wavy hair, small left-cheek beauty mark, cream sweater + denim overshirt, skeptical-friendly expression.

PFP PNG5s clip

Caption / bio line

• Bio: “Testing cheap home gadgets so you don’t buy junk.”
• CTA: “comment COURT for the next audit.”

One-job proof demo

The steamer TikTok works because frame 0 already shows problem + product + action. No intro. Proof appears by second 3, then the video repeats the result across multiple fabrics/use cases.

• Template: product touches problem immediately → visible before/after stripe → product close-up → second/third use case → verdict.
• HGT copy: mini vacuum already touching crumbs in cupholder; clean stripe by 3s; dust cup/nozzle close-up; floor mat + fabric seam tests; PASS/MID/TRASH.
• Use for: 20% product-proof lane. Pair with Household Court/Mara so the account still builds follows.

Hard QA

• Product + problem + action in frame 0.
• First proof by second 3.
• At least two use cases.
• No fake AI proof.
• Native short overlays only.

First 3 sec gate

• Need 2+: movement, contrast, face/hand/animal anchor, pattern break.
• Reject static packaging, slow unboxing, text-only hook screens.

Maxfusion rule

• Keep variant-factory thinking.
• Reject copied ads, AI-swapped third-party actors, 20-account spam.

Proof rule

• AI serials can be weird; product demos cannot fake proof.
• No baked poster typography; native overlays only.

Current board state

Next upload packet

iPhone realism layer

Machina's AI-influencer post is useful as a phone-camera realism recipe, not as an account identity. Use iPhone computational-photography language to make Mara/HGT references feel like casual TikTok source footage.

• Steal: front-camera/handheld rear-camera perspective, phone HDR, imperfect crop, mundane car/apartment settings, ambient window light, real hair/skin texture.
• Reject: thirst-trap influencer framing, fake testimonials, celebrity resemblance, provenance/detection evasion.
• Use: Mara car snack archive selfie, junk-drawer confession, kitchen proof stills.

Prompt stance

“You are specialized in computational photography and iPhone Pro camera optics…” then anchor Mara, product, mess, and ordinary environment. Product/problem must be visible in-frame.

Character sheet lock

MeAsifAI's character-sheet prompt is directly useful for Mara: one PFP is not enough. Before scaling recurring UGC clips, generate a multi-view sheet with expressions, poses, wardrobe details, and hands/product callouts.

• Panels: front, 3/4, side, back, expressions, pose variations, detail callouts.
• HGT add-ons: gadget in hand, kitchen counter pose, phone/selfie setup, skeptical test expression.
• Use: crop sheet into reference assets for future Runway/Nous shots.

Reject if Mara loses

• dark wavy shoulder-length hair
• small left-cheek beauty mark
• cream sweater / denim overshirt family
• buyer-advocate, not glam influencer energy

Compact cinematic montage prompt

Rovvmut's Seedance example proves a short prompt can work if it encodes the whole production spine: duration, genre, recurring character, journey, motif, wardrobe family, and final payoff.

• Template: duration + genre + one recurring character + 3-location journey + prop/action motif + wardrobe/color variation + explicit ending.
• HGT version: Mara moves kitchen → junk drawer → car snack archive, voice-memo/earphone motif, PASS / MID / TRASH payoff.
• Use for: portfolio reels and domestic microdrama showpieces, not factual product proof without raw proof footage.

Why it works

• One human anchor prevents random montage drift.
• Three locations create a story arc.
• Repeated prop/motif binds shots.
• Final action gives closure.

Runway cinematic narrative quality

Julia Li's hiring post is the clearest signal: people will pay for AI videos that feel like coherent short films, not random generated clips. The target is worldbuilding + recurring characters + edit taste.

• Pipeline: world bible → character/location refs → 8-12 short Runway takes → contact-sheet QC → 30-45s reel.
• HGT translation: Mara + Household Court + recurring domestic locations, not copied bodega/music-video IP.
• Service angle: sell AI cinematic shorts/tool demos once we have a reel that proves consistency.

Quality gates

• Same world every shot.
• 2+ recurring anchors/characters.
• Short takes; cut around face/hand drift.
• Warm saturation/contrast pass.
• No fake product proof.

Video Use rough-cut layer

Saidstetic's post points to Video Use: drop raw footage in a folder, tell a coding agent what edit you want, get a rough final.mp4. Treat it as editing automation, not proof that CapCut is obsolete.

• Use for: dead-air cuts, caption draft, warm/contrast grade, quick proof-demo recuts.
• Stack: Runway/Nous raw clips → Video Use rough cut → Diffusion Studio/OpenCut/CapCut final QA.
• Guardrail: no unsupported product claims, no over-graded SKU colors, no baked text when native overlays are required.

Smoke test

• One Snack Archive source folder.
• One 12–20s rough cut.
• Captions + color pass.
• First-3-sec/proof QA before posting.

Code-native edit recipes

Konstantin Paulus / Diffusion Studio is a stronger editor-automation signal than generic hype: every edit can become TSX/code, so agents can modify, diff, review, and rerender cuts.

• Adopt now: store source assets + edit recipe/code + render + QA result for serious videos.
• Use for: Household Court templates, Snack Archive proof recuts, Mara intro/scorecard components.
• Gate: smoke-test Diffusion Studio/dapi before making it production; keep CapCut/OpenCut fallback.

Why this matters

• No more opaque final-MP4-only workflow.
• One-variable variants become auditable.
• First-3-sec and proof claims can be checked before render.
• Agents can edit code instead of guessing a GUI timeline.

Claude Code web taste loop

Nerea's Claude Code tutorial is useful less for the “$50k website” claim and more for the build loop: reference board, named aesthetic, live preview, targeted critique, and screenshot QA.

• Loop: moodboard → named direction → Claude Code build → browser preview → specific critique → responsive QA.
• HGT direction: domestic evidence board, friendly lab notebook, transparent commerce command center.
• Guardrail: polish is not strategy; label real/manual/verified data clearly.

Use next rebuild

Ask for concrete changes: reduce generic SaaS card feel, make first fold explain the operating decision faster, improve mobile CTA, and make the board feel live rather than decorative.

Level 1 skip-rate loop

Grant/Alex Garcia's useful lesson for us: at the 1k stage, every post is a zero-budget experiment to reduce skip rate and earn follows. Do not buy scale until a format wins organically.

• Measure: first-3-sec hold, completion, rewatches, follows per 1k views, comments/saves/shares, profile visits.
• Bad openers: hey-guys, logo intro, slow unboxing, product box only.
• Good openers: product already attacking the mess, visible before/after stripe, skeptical binary question, immediate conflict.

Per-concept loop

same product/problem → 3 hooks → 2 formats → 1 follow-up if comments appear

• This mini vacuum has one job.
• Snack archive vs $12 vacuum.
• The ad skipped the worst part, so I tested it.

Bridgeable viral slideshows

OnlineDopamine's hook examples are real follower-growth psychology: confessional rules, identity sorting, authority secrets, and aspirational reassurance. Use the mechanics, but keep them in domestic chaos / household judgment so the account can still monetize later.

• New sprint mix: 60% bridgeable viral slideshows, 25% Household Court/domestic microdrama, 15% proof demos + lab identity.
• Rule: viral enough to grow, coherent enough to monetize later.
• Reject: generic relationship/IQ/founder bait unless it has a household bridge.

Hook translations

• Rules we made after our kitchen almost broke us.
• What your junk drawer says about your mental state.
• A cleaner showed me the one test every cheap gadget has to pass.
• When your apartment is tiny but storage is mostly geometry.

Lingya AI-video methodology scout

LOOPY's Korean thread points to Tencent Lingya as a Chinese AI creator community with AI films, short dramas, animations, prompt/model notes, storyboard breakdowns, and free classes. Use it as private methodology research, not public source footage.

• Use for: Household Court story beats, character consistency references, AI music-video/directing examples, storyboard-first Runway/Nous packets.
• Scout cap: inspect up to 6 examples, deep-analyze at most 2, convert one pattern into a packet/skill.
• Guardrail: no copying/reuploading Lingya clips; adapt structure only.

Source

Lingya: 灵芽社区 - AI创作者的聚集地
URL: lingya.qq.com
Companion: AirTranslate macOS app for Chinese/Korean workflow reading.

Directed AI short method

ZenTalksAI's thread is the clearest process rule yet: most of the work is direction, not generation. Character dynamics, expression scripting, shot storyboards, and attention/color design come before Runway/Nous.

• Ratio: about 75% creative decisions, 25% generation/edit.
• Packet: premise, characters, conflict, wardrobe/prop meaning, expression beats, storyboard, attention/color plan.
• HGT use: Household Court / Animated Case Files; Mara + gadget/problem in full color, background clutter/crowd flatter/cooler.

Pre-gen checklist

• Character before face.
• Script expressions, not just dialogue.
• Storyboard every camera/focus/background beat.
• Use color to direct the eye.
• Generate only after the packet is locked.

Vercel vgpu / WebGPU demo

Archie Auburn's reel points to Vercel Labs vgpu: an agent-first WebGPU library with CLI docs, examples API, MCP/docs surfaces, and headless Node paths. I verified the repo locally, built it, ran fast tests, pulled an official example, and made a small HGT evidence-field demo.

• Use: game-dev shader prototypes, interactive evidence maps, attention/heat-field dashboard panels, premium AI-tool demos.
• Guardrail: critical dashboard content stays HTML first; WebGPU is progressive enhancement with fallback.
• Local verification: vgpu build passed; 101 test files passed; 864 tests passed; doctor healthy on CPU llvmpipe renderer.

What this changes

Claude in the reel means Hermes/me: choose tools that expose docs, examples, CLI checks, and testable surfaces so agents can actually build with them.

• Prefer agent-ready libraries.
• Pull verified examples before inventing.
• Run shader/build/test gates.
• Demo with fallback, then promote to dashboard only when reliable.

Photo Mode assets

Runway video assets

Mara identity video