Film Konnections Botswana
Two-Week Progress Report
A creative-diplomacy pilot bridging Denmark and Botswana through filmmaking and AI technology. Facilitated by the University of Botswana and a specialized Danish creative team led by Hollywood actor Roland Møller.
A rigorous three-stage selection pipeline was designed to evaluate 22 treatment submissions with fairness, cultural sensitivity, and production viability.
Stage 1: NotebookLM Narrative Slate Review
AI-Assisted Initial Analysis
All 22 submissions (later consolidated to 17 unique viable concepts) were fed into Google NotebookLM to produce the Botswana Narrative Slate Review — a 21-page visual portfolio (Botswana_Narrative_Slate.pdf) organizing the submissions into three narrative pillars:
Pillar I — Heritage & History: Salt in the Sand, The Queen Mothers, Polelo Yaga Serai, Morero, Bokspits, Motho le Motho Sengwe
Pillar II — Social Realism & Crime: Ecominus, Tea for the Ward, Moxaxa, Unseen Sacrifice, The Science Guru
Pillar III — Allegory & Speculative: Not Yet Death, Man in the Mirrors, Signal in the Reeds, The Morning the School Bell Bled, Light in the Forest, Skeletons in Our Closets
Each film received an AI-generated one-sheet with logline, key elements, and producer's note assessing production complexity and viability. The Slate Portfolio Matrix then ranked all 17 concepts by genre pillar, complexity, and script readiness.
Stage 2: AI Writing Detection (Authenticity Screening)
Two-Pass Detection Methodology
To ensure we were selecting genuine Batswana voices — not AI-generated scripts — we ran a comprehensive authenticity analysis on all 17 treatments.
Pass 1 — Automated lexical scan (ai_detect_analysis.py): A custom Python tool scanning for em-dash density, AI-typical vocabulary ("delve", "leverage", "navigate", "robust"), filler words, sentence uniformity, academic tells, and structural patterns. Each treatment received a composite risk score.
Pass 2 — Qualitative deep-dive: Four parallel subagent batches performed close-reading of every treatment, evaluating AI signals vs. human signals. Specific markers of human authorship included: Setswana code-switching, authentic typos ("armbush", "becasue"), culturally specific names and locations, unpolished transitions, and idiosyncratic sensory details. AI signals included: triple-adjective patterns, clean binary oppositions, em-dash parentheticals, and "pitch document" prose.
Key decision rule: "Setswana absence = AI signal" — every genuine Batswana filmmaker code-switches naturally between English and Setswana. Complete absence of Setswana was flagged as a red flag.
Reports: fw-ai-detection.pages.dev · Raw notes: reports/batch1–4_results.md
Stage 3: Production Complexity Scoring & Final Shortlist
From 17 to 8
Each surviving treatment was scored on production complexity (1–10) assessing: number of locations, cast size, period/costume requirements, VFX needs, and cultural sensitivity requirements.
Cut: Bokspits (thinnest treatment, 320 words, AI bio, overlapped with other water stories), Science Guru (flagged HIGHLY LIKELY AI-generated), and 7 others that either scored too high on AI detection or exceeded production capacity.
Added: The Last Water Day (late submission with strong authentic writing and unique dystopia genre) replaced Science Guru in the final 8.
Report: fw-film-shortlist.pages.dev
The Final 8
| # | Film | Genre | Writer | Director | Tier | Complexity |
|---|---|---|---|---|---|---|
| 01 | Not Yet, Death | Allegorical Drama | Abale | Katamelo (Trippi) | 1 — Micro | 3.0 |
| 02 | Motho le Motho Sengwe | Historical Comedy | Ruby Ramabu | Mégrangé | 1 — Micro | 3.5 |
| 03 | Skeletons in Our Closets | Psych. Thriller | Setso Siane | Lebetha Kutlo | 1 — Micro | 3.0 |
| 04 | The Last Water Day ★ | Dystopia | Nthebe Peloewetse | Pesalema | 2 — Short | 4.5 |
| 05 | Moxaxa | Crime | Molefi Moses Molefi | Larona | 2 — Short | 5.0 |
| 06 | Morero | Historical Drama | Tshepiso Monnawaletsatsi | Pius Cezuka | 2 — Short | 5.5 |
| 07 | Unseen Sacrifice | Crime/Thriller | Lorna Bonno | Tshepo | 3 — Dev | 6.5 |
| 08 | Polelo Yaga Serai | Historical Drama | Kago Mochotlhi | Lone | 4 — Epic | 8.0 |
54 actors signed up through in-person casting sessions at UB and digital self-signup. 34 have signed consent forms; 32 have photos on file.
Casting Progress by Film
Total: 39/88 roles in callback or cast status (44%). Film 07 (Unseen Sacrifice) is fully cast. Film 04 (Last Water Day) leads Tier 2 at 70%. Film 03 (Skeletons) has the most open roles — casting sessions pending.
Casting System
A custom real-time casting tracker (fw-tracker.pages.dev) was built and deployed on Cloudflare Workers + KV storage. Directors cast actors directly via the web interface. 245 state edits logged across 14 unique editors including directors Pesalema, Lone Saakane, Setso, Bontle K, and others.
18 crew candidates interviewed at UB. 9 selected, 9 declined. The team covers cinematography, sound, editing, VFX, production, set design, costume, and BTS documentation.
| Name | Rating | Roles | Key Strength |
|---|---|---|---|
| Mohau | 10/10 | Lead DOP | Lead cinematographer since 2013. TV & film. |
| Lame | 10/10 | Sound, Editor, Music | Full-stack audio + editing. Premiere, DaVinci, Pro Tools, Logic. |
| Constantine | 10/10 | Editor, VFX | Lead post: Blender, Unreal, Maya, AE. 15yr experience. |
| Malebogo Setuke | 10/10 | Composer | Score & soundtrack. Professional notation (Sibelius). Age 48. |
| Aobakwe | 9/10 | Set Design | Production design. Film school. Horror background. |
| Dimpho | 9/10 | B-Cam DOP | 2nd unit camera. Young, high-rated. |
| Faith Dibakoane | 8.5/10 | Producer, Costume | Production swiss-knife. Reliable logistics lead. |
| Benedict | 8.5/10 | Producer, C-Cam | 3rd camera + production. 10yr music background. |
| Losika | 8/10 | BTS / Social | Social media, documentary, support editor. |
Crew Allocation Plan
Film 01 (Not Yet, Death) — A-team: Mohau (DOP), Dimpho (B-cam), Lame (sound), Aobakwe (set), Constantine (VFX/edit), Losika (BTS). The intimate chamber piece demands precise lighting and sound in a concrete room.
Film 04 (Last Water Day) — Lean crew: Danish DOP leads, Benedict (B-cam), Faith (production/costume). Documentary-style, 8+ locations, run-and-gun. Post-production compensates with news overlays and AI color grading.
Shared post-production: Constantine (edit + VFX), Lame (sound design), Malebogo (score), Losika (BTS edit) work both films sequentially.
Two core masterclass tracks delivered at UB campus, combining Hollywood craft with AI-assisted filmmaking techniques.
🎬 Reality Rules — Roland Møller
Hollywood actor Roland Møller (Papillon, Atomic Blonde) delivered the pilot masterclass on screen acting fundamentals. Core teaching: "Reality Rules" — the invisible but ruthless logic behind every believable scene. Focus on authentic emotional truth, improvisation grounded in character, and the difference between theater and camera acting.
Presentation: filmkonnections.com/acting-masterclass/ · PDF
🤖 The AI Operator — Gustav & Team
Interactive masterclass on AI-assisted filmmaking workflows. Covered: AI-assisted VFX (environment extension, digital doubles), AI color grading, atmospheric generation, storyboard generation using text-to-image, and how emerging economies can leapfrog expensive traditional production models. Delivered at UB with VAPA (Visual and Performing Arts) department. Slides: Reveal.js presentation with 8 film case studies.
Presentation: code/masterclass/ · Live version
Additional Sessions
Victor's Social Media for Film Production — scheduled masterclass on optimizing social media for independent film promotion.
Director mentorship sessions — ongoing 1-on-1 mentoring between Danish crew and 8 assigned Batswana directors, covering shot composition, actor direction, and script development.
Two films have finalized shooting scripts. Storyboards, location scouts, concept art, costume lists, and prop breakdowns are in progress across the slate.
📝 Scripts Finalized
Film 01 — Not Yet, Death: Full 10-scene shooting script by Abale Nyatshane & Rikkie Molatlhegi. Director-ready.
Film 04 — The Last Water Day: 12-minute shooting script by Nthebe Eric Peloewetse, co-written with Palesa Rapoo. Scene-by-scene with stage directions.
🎨 Storyboards
Film 01: 20 AI-generated cinematic keyframes + 4 grid compositions. Full scene breakdown with lighting, camera, and blocking notes.
Film 04: 34 storyboard frames across 2 versions. Shot bible, DOP quick-reference, and per-crew-member checklists deployed at filmkonnections.com/films/04/storyboard/
📍 Location Scouting
Film 04: 10 location photos scouted by Kimberly. Manase Fences (industrial depot), Osanta Complex, Mochudi Strip, Red Yard homestead. Interactive GPS map deployed.
Film 06: 10 reference photos for kgotla/cultural settings. Molepolole ward research.
🎭 Concept Art & Props
Film 01: Costume list, makeup notes, props list (Death's chair, Book of Life, Soul Catcher, dice set). Production breakdown document complete.
Film 06: 19 concept art pieces for Morero (kgotla, cattle, lobola ceremony).
Film 03: Ending enhancement brainstorm with 12 creative options for the ghost reveal.
🎥 Equipment Strategy
Production equipment shortlist finalized with a "backpack cinema" philosophy: one camera, one good lens, AI in post. Emphasis on sound (2× wireless lavs, shotgun + boom, field recorder) because "AI can upscale video but cannot convincingly rebuild dialogue." Minimal lighting kit supplemented by African natural light. Full document: reports/production-equipment-shortlist.md
Audition Sides Prepared (11 films)
Not Yet, Death (dice game scene) · Motho le Motho (bedroom scene) · Skeletons (park breakup) · Last Water Day (candlelit confrontation) · Moxaxa (morning pressure scene) · Morero (missing cow scene) · Unseen Sacrifice (investigation scene) · Polelo Yaga Serai (regiment rally) — plus 3 additional audition docs. All in source/auditions/
A field-tested rescue workflow for difficult FX6 production dialogue. Built from real Last Water Day clips, not theory. The rule is simple: protect speech first, remove noise second, and only use AI when it actually wins in A/B comparison.
Standalone page
The full dedicated reference now lives at filmkonnections.com/hub/audio/. The summary below stays in the report for context, but the standalone page is the main home for the audio workflow.
Core Principle
Good dialogue starts on set. Clean lavs, disciplined boom placement, proper gain staging, and room-tone capture still beat any post tool. AI can help rescue ugly production audio, but it cannot magically rebuild a destroyed performance without tradeoffs.
Working philosophy: capture the cleanest production track possible, then use post-production like a scalpel — not a fantasy machine. Every cleanup pass must be judged against the raw source for intelligibility, consonant survival, and emotional naturalness.
What Actually Worked on Real Clips
| Clip | Best Winner | What Failed | Main Lesson |
|---|---|---|---|
| INTERFILM_3002 | Custom local spectral-gate + polish chain | DeepFilterNet over-suppressed speech · ElevenLabs full clip collapsed badly | Local cleanup won clearly; AI rescue only worth considering on tiny phrases. |
| INTERFILM_3079 | Local full-clip cleanup + ElevenLabs on first 18 seconds only | Full-clip ElevenLabs underperformed local pass | Phrase-level AI can work brilliantly, but only on the exact sections where it wins. |
| INTERFILM_3084 | Track-2-led anlmdn rescue chain | Track 1 was not useful enough for subtraction · ElevenLabs opening phrase did not beat local | Sometimes the strongest result is a disciplined local denoise workflow, not a fancy model. |
Primary Workflow — Full Clip Rescue
1. Extract and profile all 8 audio channels.
2. Choose the real dialogue track (usually A2 on these FX6 clips).
3. Find long low-speech windows inside the same take.
4. Run conservative denoise / spectral gate.
5. Add light intelligibility EQ, de-ess, compression, and limiting.
6. Export full review video and compare against raw in sync with picture.
Secondary Workflow — Phrase-Level AI Rescue
Use only when needed. Extract a 5–15 second dialogue-heavy problem phrase, run cloud isolation or enhancement on that phrase only, and compare against both the raw phrase and the best local phrase. If the model wins, patch only that repaired segment back into the local master with a short crossfade.
Recommended Tool Order
Tier 1 — Start Here
Local signal processing: real noise-print denoise, spectral gating, EQ, de-ess, compression, limiting. This remains the most reliable first pass because it is controllable and often sounds more natural than generic AI isolation.
Tier 2 — Best High-End Pro Option
iZotope RX 11 Advanced remains the strongest untried premium route for these clips: Dialogue Isolate, Spectral De-noise, Dialogue De-reverb, Spectral Repair, and manual phrase surgery.
Tier 3 — Cloud / Model Assist
ElevenLabs Voice Isolator is useful for targeted phrase rescue, not as a blind full-clip master. Adobe Podcast Enhance and Resemble Enhance are also worth phrase-level testing. These tools should earn their place with A/B listening, not hype.
What We Learned About AI Models
DeepFilterNet3: technically workable with compatibility patches, but on these clips it often dulled or over-suppressed speech.
ElevenLabs full-clip isolation: technically functional, but not the winning full-length master for the tested clips.
ElevenLabs phrase rescue: legitimately useful in selective cases (especially the opening of 3079).
fal.ai: only worth trying if it exposes a genuinely strong hosted speech enhancement, dereverb, or source-separation model. fal itself is not the magic — the underlying model is.
Hybrid Workflow We Recommend
Local Master + AI Patches
Best practical workflow for Film Konnections:
1. Build the best possible full-length local master.
2. Mark the ugliest 3–5 phrases only.
3. Test AI isolation/enhancement on those phrases.
4. Reinsert only the phrases that objectively improve clarity.
5. Deliver both a strongest cleanup version and a more natural alternate for director/editor preference.
What to avoid
Do not trust full-clip AI isolation blindly. Do not replace production dialogue with synthetic voice unless the team explicitly wants ADR-style intervention. Do not over-denoise until consonants vanish. And never judge audio from waveform screenshots alone — always review in sync with picture and performance.
Future tests still worth doing
Still worth testing: iZotope RX 11, Adobe Podcast on short phrases, Resemble Enhance on short phrases, dereverb-first workflows, Accentize/Waves/Acon dialogue tools, and any fal.ai model that clearly offers speech enhancement or dereverb rather than generic audio generation.
A full digital production infrastructure was built and deployed on Cloudflare, providing the team with casting tools, storyboards, checklists, and production dashboards accessible from any phone.
| Platform | URL | Purpose |
|---|---|---|
| filmkonnections.com | filmkonnections.com | Main site: hub, crew page, film subsites, masterclass |
| Production Hub | filmkonnections.com/hub/ | Central dashboard for all 8 films |
| Casting Tracker | fw-tracker.pages.dev | Real-time casting with role assignments, actor photos, consent tracking |
| Tracker API | fw-tracker-api (Worker) | KV-backed REST API for state & actor data. 245 versioned edits. |
| Film Shortlist | fw-film-shortlist.pages.dev | Final 8 with complexity scores, AI VFX map, timeline |
| AI Detection | fw-ai-detection.pages.dev | Writing authenticity assessment for all 17 treatments |
| Late Submissions | fw-late-submissions.pages.dev | 3 late treatments evaluated separately |
| Film 04 Storyboard | films/04/storyboard/ | 21-shot visual bible with character refs, DOP checklists, GPS map |
| Film 01 Checklist | films/01/checklist/ | Roland's personal prep checklist for Not Yet, Death |
29 HTML pages deployed across Cloudflare Pages. All sites are mobile-optimized — crew and directors access storyboards and casting data from phones on set.
Social media presence established at @film.konnectionz
🎬 This Week (4–8 Mar) — Active Production
• Tue 4 Mar — Film 04 (Last Water Day) interiors at Botswood ✅ ~30% complete
• Wed 5 Mar — Film 01 (Not Yet, Death) interiors at Botswood + Film 04 outdoor scenes
• Thu 6 Mar — Film 01 outdoor scenes (coffin, park) + Film 04 depot at Raber's fence compound
• Fri 7 Mar — Film 04 city montage (Mochudi dawn shoot) + remaining depot/road scenes
• Sat 8 Mar — Final pickup shots + golden-hour riverbed scene (Film 04 SC12)
🛫 Sun 9 Mar — Wrap & Depart
• Danish team departs Botswana
• All Film 01 & Film 04 footage backed up and transferred
• Dailies review & selects handed off to local editing team
• Location releases and permit paperwork finalized
🎯 Post-Production & Next Phase
• AI VFX pipeline begins on Film 01 & Film 04 footage (remote)
• Color grading, sound design, atmospheric generation
• Local directors continue prep for Tier 2 films (Moxaxa, Morero, Unseen Sacrifice)
• Campus screening of completed micro-dramas at UB
Film Konnections · Botswana–Denmark Creative Diplomacy Pilot
University of Botswana · DVCRE Research & Innovation · March 2026