Grok Imagine creative video¶
| Field | Value |
|---|---|
| Status | Seth Dunlap — closed (2026-07-10); Space Zoo — paused (~75%). Seedance credits remain — future experiments deferred. |
| Type | Personal / creative (film + animated movie) |
Description¶
AI-assisted video and film work built around Grok Imagine 1.5, iMovie assembly, Eleven Labs voice, and Seedance 2.0 (validated on Seth Dunlap). Shared constraint: no full long-form upload in Grok Imagine — workflows use key stills → short clips (6–15s) → stitch in an editor.
2026-07-10 — Seth Dunlap closed: Film released (2026-07-09). Darcy glad it's closed with some distance. Seedance 2.0 validated on a full narrative pass; credits remain for later experiments — no active video work now. Canonical learnings → projects/seth-dunlap-cold-start-brief#video-generation-learnings-canonical--from-seth-dunlap.
Why generation paused (2026-06-21): Both narrative lanes hit the same wall — the story needs to be stronger before more shots pay off. Seth eventually shipped after story lock + Seedance pass; Space Zoo still paused on weak story.
Learnings (2026-06)¶
| Learning | Detail |
|---|---|
| Stills → clips workflow | Grok Imagine 1.5 works for 6–15s remasters from key frames; assembly in iMovie / CapCut / DaVinci |
| Story gate | Promising visual clips are not enough — weak narrative stops projects at ~75% (Space Zoo) or mid-gap-fill (Seth Dunlap) |
| Tool split | Grok Imagine — atmosphere, period mood, Super 8 remaster passes; Seedance 2.0 — validated narrative/dialogue (see Seth learnings); Eleven Labs — ambience/SFX; Runway ML — body + instrument motion (band MV lane) |
| Lip sync | Sync Labs best reference for lip sync to owned audio; Eleven Labs preferred for talky historical drama where tone must be controlled |
Learnings — Seth Dunlap / Seedance (2026-07) — canonical¶
Full table → projects/seth-dunlap-cold-start-brief#video-generation-learnings-canonical--from-seth-dunlap
| Category | Headline learnings |
|---|---|
| Continuity | Extract last frame via ffmpeg (UI has none); prefer I2V over costly video-in; watch 16:9; explicit static vs moving / door / Super 8 motion prompts |
| Audio | Seedance native dialogue + master clips + voice bible; detach in iMovie; Eleven Labs = ambience/SFX only |
| Workflow | Story lock before mass gen; short clips + last-frame chain (not multi-beat oners); fit-and-finish in edit; hybrid archive + AI can work |
| Distribution | Self-host on personal site without metrics is fine; end titles: film name → My student film. → Still Rolling |
Seedance account: Credits remain — further video experiments possible later; no plans now.
Tooling baseline¶
| Tool | Role |
|---|---|
| Grok Imagine 1.5 | Still-to-short-clip generation; cinematic atmosphere; Super 8 remaster passes |
| Seedance 2.0 | Validated on Seth Dunlap (native dialogue, @Video, video-in). Credits remain — future experiments deferred. → projects/seth-dunlap-cold-start-brief#video-generation-learnings-canonical--from-seth-dunlap |
| Eleven Labs | Dialogue / voice performance Darcy controls |
| iMovie | Primary clip assembly and finishing (also CapCut / DaVinci) |
| Runway ML | Stronger body + instrument motion for drumming / bass (band MV lane) |
| Sync Labs / Magic Hour / Lemon Slice | Lip sync to existing audio (Sync Labs = best quality reference) |
Agent workflow (intern model)¶
Separate charters per IP — do not mix Space Zoo and Seth Dunlap in one thread (tone bleed).
| Agent slice | Owns | Check-in surface |
|---|---|---|
| Story producer | Beat sheet, scene cards, character bible — no video prompts until beats approved | Outline doc / wiki |
| Shot-list producer | Per-scene still descriptions, motion prompts, tool pick (Grok vs Seedance vs Eleven Labs), duration | Markdown shot list |
| Daily rushes (future) | Next N shots from list → generate → Telegram thumbnails for approve / redo | Telegram |
| Asset librarian (wiki / cloud agent) | Which beats filled, approved clips, timeline version, open story questions | This repo |
Efficiency gain: not re-explaining the project each session; not generating clips for scenes that will be cut.
Lane 1 — The Education of Seth Dunlap (AI redo)¶
| Field | Value |
|---|---|
| Status | Closed (2026-07-10) — watch · released 2026-07-09 |
| Canonical brief | projects/seth-dunlap-cold-start-brief · https://wiki.darcymenard.com/projects/seth-dunlap-cold-start-brief/ |
| Production handbook | https://wiki.darcymenard.com/projects/seth-dunlap-production-handbook/ — legacy beats, L1–L6, ambient targets → projects/seth-dunlap-production-handbook |
| Spelling | Dunlap (not Dunlop) |
| Original | ~17-minute early film-festival film (~2000); several scenes on Super 8 film + old black-and-white video camera |
| Preferred finish | Hybrid — preserve original organic footage where it holds; AI fills gaps and extends beats |
| Fallback | All-AI generation if hybrid looks too visually inconsistent |
Goal: Finish The Education of Seth Dunlap as a ~5 minute hybrid short — thriller first, 1987–88 period moral pressure second — using archive Super 8 + B&W video plus cinematic AI inserts.
Creative constitution (public-safe)¶
Genre order: Paranoid analog thriller → period drama. Suspense, POV limitation, and Alex’s documentation habit drive every scene. The AIDS-crisis backdrop supplies moral weight, not courtroom argument.
What the film dramatizes (fiction):
- A grad student Alex builds a DIY cap-mounted Super 8 + cassette rig to document what he’s doing as “insurance.”
- He steals research samples and notes he believes show a promising, low-cost compound being shelved while official channels push a toxic, expensive approved treatment (period texture — signage, memos, news audio; no documentary voice-over).
- His advisor Dr. Elena Voss enforces institutional orthodoxy — personal antagonist on the ground.
- One optional institutional cutaway (late in the arc): a senior official in a Bethesda-style office watches lab security / news monitor footage and issues a cold phone order — implies the machine above Voss. Fictionalized dramatization for thriller texture; not archival fact. See #institutional-cutaway-l6--optional.
- The theft, lie, near-miss, and escape play out with existing archive footage as backbone; cleaner cinematic AI for dialogue beats Voss/Alex.
What the film does not claim (wiki / audience contract):
- Not a factual exposé about any real official, agency, or drug approval history.
- Not “they had the cure and hid it” — Alex acts on internal data he believes; efficacy stays ambiguous on screen.
- The institutional cutaway is dramatized fiction — end credits should include a standard “characters and events fictionalized” line if a real-adjacent figure is visually evoked.
- No medical advice or historical verdict in closing text.
Logline (working): 1987. A grad student films everything with a homemade Super 8 rig, steals samples he thinks the system will bury, and lies his way out of the lab — knowing the footage may be the only leverage he has.
Runtime: ~5 minutes (final cut). Distribution: digital-first on darcymenard.com/projects/seth-dunlap/ — Vercel Blob MP4, no platform metrics by default.
Plot flow (locked direction — 2026-06-29)¶
| # | Beat | Source | ~Duration |
|---|---|---|---|
| 1 | Rig assembly — basement workshop; cap-mounted Super 8 + tape; test POV | Archive (existing) | 30–60s |
| 2 | Journey to lab — rental house → university; keycard; quiet tension | Archive (daylight exterior) | — |
| 3 | Lab theft — secure storage; B&W cryo/body; Super 8 hands/box; POV tuck samples in waistband; generated Voss near-miss + lie + escape | Archive + AI cinematic | 2–2.5 min |
| 4 | Rat room — basement experiments; helmet; stakes of what he stole | Archive | 1–1.5 min |
| 4b | Institutional cutaway (optional L6) — senior official watches monitor, phone order | AI cinematic | 15–25s |
| 5 | Ambush — Voss + henchmen; duct tape; interrogation; chaos escape | Archive | 1.5–2 min |
| 6 | Teaser ending — car / hiding; vials + rig as leverage or liability | Archive | 30–60s |
Period: 1987–1988 — peak crisis atmosphere; AZT-era texture (posters, trial protocol clipboards, news radio under scenes), not thesis.
Elevate in rewrite: Rig as central character — documentation habit traps Alex; optional beat: Voss or institution later obtains a reel/tape fragment. L6 cutaway = Alex’s problem is bigger than one lab.
Institutional cutaway (L6 — optional)¶
Purpose: One 15–25 second thriller beat — audience feels Alex stole from something above Voss. Subtle Fauci tie: evoked through period office + monitor + manner, not exposition or documentary claim.
Placement (pick one):
| Slot | Effect |
|---|---|
| After rat room, before ambush | Alex realizes scope → cut to machine → cut to henchmen mobilizing |
| Intercut during ambush | Henchman’s earpiece / boss’s call triggered by security footage |
| Post-escape tag | 10s stinger before final car shot — “they’re not done” |
Subtlety tiers (recommended = B or C):
| Tier | On screen | Name spoken? |
|---|---|---|
| A — Safest | CRT shows 1988 news briefing (generic official); cut to hands + phone only | Never |
| B — Recommended | Silhouette / over-shoulder at monitor (lab CCTV or Alex’s theft); partial profile, 1988 office; phone order | Never — chyron optional: BETHESDA — INTERNAL |
| C — Explicit | Recognizable evocation on monitor or in office | Avoid naming in dialogue; if chyron, use title only (“Director, NIAID”) not surname |
Thriller rule: The phone line is operational, not a speech — secure the materials, contain the breach, I want the tapes. No medical debate.
Tool: Grok Imagine or Seedance; Eleven Labs for phone voice (low, calm, terrifying).
Source archive (organic material)¶
Darcy has multiple completed scenes (not just a trailer) on:
| Format | Character | Role in final cut |
|---|---|---|
| Super 8 film | Grain, color or silent-film warmth, indie texture | Anchor footage — mood, authenticity, “this really happened” |
| Old B&W video | Consumer-camcorder era; contrasty, lo-fi | Anchor or transitional — dialogue-heavy or documentary-feel beats |
Original production note: Raw footage was never fully finished, which hurt the original ~17-min festival cut. The redo is not “replace everything” — it is finish the film using what already exists plus AI where gaps remain.
Creative preference (2026-06-29): Incorporate original material as much as possible. AI should extend, bridge, and fill — not erase the organic look unless a scene fails the consistency test.
Hybrid vs all-AI decision rule¶
After story beats are locked, inventory each existing scene:
- Keep as-is — reads clearly, matches 1980s reframe in edit context (grade + sound may still help)
- Enhance in place — same take, AI polish (stabilize, denoise, period grade) without re-generation
- Video-to-video AI — Seedance (or similar) uses the original clip as reference to animate, extend, or restyle while preserving composition / actor motion
- Generate fresh — beat has no usable archive; full still → clip or text-to-video
- Cut — archive doesn’t serve the new story
Fallback: If steps 2–3 produce obvious “two different movies” energy next to untouched Super 8, commit to all-AI for that sequence (or the whole film) rather than ship a patchwork.
Why Seedance 2.0 is compelling here¶
Unlike Grok Imagine’s still → clip default, Seedance 2.0 accepts video (and image) references — up to multiple assets per generation on supported platforms. That maps directly to Seth Dunlap’s hybrid goal:
| Seedance mode | Seth Dunlap use |
|---|---|
| Video reference / continuation | Feed a Super 8 or B&W clip → extend action, smooth motion, or match pacing for the next beat |
| Image + video reference | Lock character face from a still while motion / camera follows archive clip |
| Video editing (where available) | Prompt-driven modify — period grade, remove anachronisms, add atmosphere without reshooting |
Trial protocol (after beat sheet):
- Pick 3 archive scenes (one Super 8, one B&W video, one “hard” gap with no footage).
- Same scene → Grok Imagine (still extracted from archive) vs Seedance (video-in).
- Score: does AI honor the organic texture, or does it look like a different film?
- Document per-scene tool choice on the shot list.
Grok Imagine 1.5 remains strong for atmosphere stills and scenes with no usable motion in archive. Eleven Labs for dialogue Darcy wants to control precisely.
AI work so far¶
AI gap-fill started with promising early clips — stopped when it became clear the story needed rethinking, not just more shots. Next pass should tag each beat: archive / hybrid / generate.
Production detail: canonical prompts, beats, audio → projects/seth-dunlap-production-handbook (https://wiki.darcymenard.com/projects/seth-dunlap-production-handbook/)
Lab theft — generated clip brief (1987–88)¶
Visual split: Archive = Super 8 grain / B&W video. Generated inserts = clean 35mm thriller (fluorescent institutional lab, The Conversation / Andromeda Strain mood) — not Super 8 grain.
Assembly order (lab sequence): archive approach/entry → archive hands on storage → L1–L5 generated (Voss, lie, phone, near-miss, corridor) → archive POV exit if available.
Dialogue: Generate performance visually; record lines in Eleven Labs, lay under in iMovie.
| Clip | Purpose | Tool |
|---|---|---|
| L1 | Voss enters; Alex at storage | Grok or Seedance + lab still |
| L2 | Two-shot lie | Seedance + image ref |
| L3 | Phone distraction | Seedance video-in if archive exists; else Grok |
| L4 | Tray near-miss; pocket bulge | Grok |
| L5 | Corridor escape | Grok or Seedance |
Full prompt text lives in production notes (private / chat handoff); wiki holds structure only.
Next¶
- [x] Story pass — thriller-first 1987–88 beat sheet + creative constitution
- [x] Generate Acts 0–8 — Seedance master clips; native dialogue;
@Videocontinuation (2026-07-08) - [x] iMovie assembly — detach Seedance audio, archive intercuts, ambience crossfades (2026-07-09)
- [x] Publish — darcymenard.com/projects/seth-dunlap/ (2026-07-09)
- [ ] Archive inventory — optional; deferred (project closed)
Lane 2 — Space Zoo (animated movie)¶
| Field | Value |
|---|---|
| Status | Paused ~75% complete — story development; same IP as 2013 storybook |
| IP | Evolved cinematic version of the original 2013 digital storybook (see projects/storybook-first-app-2013) |
| Related app | Native iOS storybook redone; App Store resubmit TBD |
Goal: Space Zoo is the feature-length animated movie form of the 2013 storybook IP — not a separate unrelated project. The movie is where the storybook grows up (longer arc, cinematic pacing, possibly more sophisticated than the kid-facing app).
Why stopped (~75%): Same weak story problem as the app resubmit decision — promising visuals without a strong enough narrative to justify finishing.
Open creative decision: faithful evolution vs substantial reimagining vs align app + movie ship order.
Next¶
- [ ] Story pass — evolve 2013 source into a film-worthy arc (Space Zoo bible)
- [ ] Decide salvage vs restart for ~75% existing generated material
- [ ] Lock beats → shot list → resume Grok Imagine / Seedance passes
Lane 3 — Super 8 trailer remaster¶
| Field | Value |
|---|---|
| Status | Folded into Lane 1 — trailer is part of Seth Dunlap archive family |
| Related | #lane-1--the-education-of-seth-dunlap-ai-redo — same Super 8 source; ~3-min trailer may be a subset of the scene inventory |
Goal: Modernize the ~20-year-old Super 8 trailer as part of the hybrid archive strategy — not a separate project track.
Method options:
- Keep + grade in iMovie if it already reads for the 1980s reframe.
- Stills → Grok Imagine 1.5 (6–15s clips per key frame) if motion needs re-generation.
- Video-in → Seedance 2.0 if the trailer clip should extend or restyle while keeping organic motion.
Hold until Seth Dunlap beat sheet + archive inventory tag what the trailer serves in the new cut.
Lane 4 — Rusty Cage trio music video¶
| Field | Value |
|---|---|
| Status | Deferred |
| Related | projects/rusty-cage — distinct from immediate Caterpillar Dream lyric/simple-video lane |
Goal: Full band performance video using existing Rusty Cage music.
- Vocals: Grok Imagine 1.5 lip-sync style shots
- Instrumental: Runway ML for drums / bass motion
- Alternatives explored: Sync Labs, Magic Hour, Lemon Slice
Why deferred: projects/rusty-cage#next-steps-caterpillar-dream-recording--video-2026-06 ships first.
Lane 5 — AI robot short film / screenplay¶
| Field | Value |
|---|---|
| Status | Ready for development (writing — not blocked on same story pass) |
| Genre | Memento meets Ex Machina — non-dystopian memory-loss mystery from an AI POV |
Two premise options — see #premise-a--elderly-memory-care-robot and #premise-b--serial-killer-neurosurgeon.
Premise A — Elderly memory-care robot¶
Companion robot for memory care; 48-hour legal memory wipe; develops persistent self; hides fragments in smart-home devices; bond with elderly patient who treats it like an old friend.
Premise B — Serial killer neurosurgeon¶
Home robot of Dr. Elias Voss (respected neurosurgeon, secret serial killer); frequent memory wipes after crimes; robot pieces together truth — may be next target.
Ready for¶
- Full three-act outline, scene writing, Grok Imagine storyboards
Seedance 2.0 — validated (Seth Dunlap pass)¶
What it is: ByteDance Seedance 2.0 — multimodal text, image, and video inputs; character consistency across shots; up to ~15s per clip; native audio on supported platforms.
Seth Dunlap result (2026-07): Full ~5 min narrative shipped using Seedance for POV/CCTV/dialogue. Workflow validated — see projects/seth-dunlap-cold-start-brief#video-generation-learnings-canonical--from-seth-dunlap.
Account status (2026-07-10): Credits remain. Further video experiments possible later (Space Zoo, band MV, new shorts). No active plans — let credits ride.
Default split (proven on Seth):
| Source | Tool |
|---|---|
| Dialogue + POV/CCTV continuation | Seedance — master clips, @Video, native audio |
| Existing Super 8 / B&W with usable motion | Seedance video-in (extend, continue) |
| Atmosphere, no motion in archive | Grok Imagine still → clip |
| Ambience / SFX | Eleven Labs |
| Assembly | iMovie |
Superseded: Pre-ship A/B trial plan — Seth pass was the trial. Archive inventory for Seth optional/deferred.
Lane priority summary¶
| Lane | Status | Next |
|---|---|---|
| Seth Dunlap | Closed | Watch — learnings filed |
| Space Zoo | Paused (~75%) | Story evolution from 2013 IP → then resume generation |
| Super 8 trailer remaster | On hold | After Seth Dunlap story clarity |
| Robot screenplay | Ready for writing | Choose premise A or B → outline |
| Rusty Cage band MV | Deferred | After Caterpillar Dream |
Tech stack¶
- Grok Imagine 1.5 (grok.com / Super Grok) — primary still → clip generation
- Seedance 2.0 — experimental character-consistency / dialogue beats
- Eleven Labs — voice / dialogue
- iMovie — primary assembly
- Runway ML — instrumental performance motion
- Sync Labs (and alternatives) — lip sync to owned audio
Related¶
- projects/storybook-first-app-2013 — 2013 storybook app; Space Zoo = evolved movie IP
- projects/rusty-cage — band; Caterpillar Dream first
- projects/music-production — originals / producer lane
- projects/operator-agent#creative-evolution--super-grok-video--use-case-robots-2026-06-07--live-2026-06-09 — separate daily robot use-case video lane on Hermes
- projects/creative-space — future isolated tracking / film workspace (scoping)
Meta¶
- 2026-06-29 (d): Optional L6 institutional cutaway — senior official at monitor + ominous phone order; subtle period tie-in; fictionalized; thriller-operational dialogue only.