ai-lab-workshop-20251008-initial-planning

# ai-lab-workshop-20251008-initial-planning Perfect — tomorrow’s session (Oct 8) is **Media Production with AI: Prep (Session 1 of 3)**, so this is the one focused on **pre-production**: planning, research, storyboarding, scheduling, and concept visualization. Below is a **long brainstorm list** of things you could do, organized by type so you can mix and match depending on the group’s energy and what tech setup you have ready in the Lab. --- ## 🧭 Framing and Setup * **Quick framing talk (5 min):** how pre-production determines the success of any media project, and how AI lets us compress the “thinking → making” loop. * **Prompt scaffolding mini-demo:** show how structured prompts can replace traditional shot lists, treatments, or mood boards. * **The “producer’s triangle”:** fast ↔ cheap ↔ good—then ask how AI might change that constraint. --- ## 🧠 Concept Development * **Idea generation sprint:** give everyone a shared theme (“migration,” “time,” “thresholds”) and 10 min to co-develop a concept outline with ChatGPT or Claude. * **Genre pivot exercise:** have AI rewrite a concept across genres (documentary → horror → rom-com) to illustrate tonal flexibility. * **AI as creative partner:** prompt comparison—same idea pitched to 3 models, discuss differences in narrative logic and aesthetic taste. --- ## 🎬 Storyboarding & Visualization * **Text-to-image storyboards:** * Use DALL·E 3 or Midjourney to generate visual sequences from scripts. * Compare coherence across models. * Discuss prompt tokens like *“wide-angle cinematic lighting”* vs *“flat storyboard sketch.”* * **Camera grammar exercise:** have Claude or Gemini output a JSON shot list (`scene`, `shot type`, `motion`, `duration`), then visualize it. * **AI animatic:** feed the storyboard frames to Runway or Pika to get quick motion tests. * **Prompt translation:** show how a screenplay paragraph → image prompt → generated frame → refined prompt chain can evolve. --- ## 📅 Planning, Scheduling, & Logistics * **AI producer assistant:** demonstrate prompting ChatGPT Edu to generate call sheets, budgets, and production timelines. * **Constraint play:** “We only have one camera, 2 hours, and a single hallway.” Let AI propose creative shooting plans. * **Cast/crew matrix:** generate hypothetical credits and role descriptions to see how AI handles interpersonal logistics. * **Notion/Sheets automation:** build an AI-generated production plan imported into Google Sheets or Airtable. --- ## 🎭 Script, Voice, and Tone * **Script style translation:** same scene written by AI in three tonal registers (academic, comedic, noir). * **Table-read with voice models:** use ElevenLabs or ChatGPT Voice to audition AI voice actors. * **Dialogue polishing:** paste human-written dialogue, have the model suggest pacing/beat adjustments, then critique its taste. --- ## 🎨 Visual Moodboarding * **Collective moodboard wall:** project a stream of generated images; participants upvote or sketch alternatives. * **Prompt dissection:** show how adding art-historical or cinematic references changes style. * **Cross-model remix:** take one prompt and pass it through DALL·E → Stable Diffusion → Ideogram to illustrate model bias and “house look.” * **Style library exercise:** build a shared “Harvard AI Visual Lexicon” board—participants tag prompts and results for future projects. --- ## 🧩 Workflow Design * **From chat to pipeline:** show how a text conversation can become a structured workflow (e.g., generate shot list → export → Storyboarder → Premiere). * **API preview:** brief glimpse of Realtime API or Whisper for logging interviews in later sessions. * **Version control for ideas:** demonstrate naming conventions, file versioning, or Git for media projects. --- ## 🧍 Hands-On Team Activities * **3-person mini-crews:** writer + designer + producer each prompt different stages; recombine outputs. * **“AI Pitch Meeting”:** teams develop a one-minute AI-generated pitch deck (3 slides, 1 image, 1 tagline, 1 budget line). * **Lightning showcase:** end with quick 60-second presentations of each project’s AI-assisted pre-production materials. --- ## 🧩 Reflection & Critical Layer * **Bias and taste discussion:** how model training data affects visual language (e.g., gendered camera gaze). * **Ethical reflection:** synthetic casting, likeness rights, and authorship. * **Meta-prompting:** have participants ask the AI to critique its own suggestions (“Why did you choose this shot?”). * **Archivability:** how to store and cite AI-generated pre-production material in academic or creative workflows. --- ## 🧰 Possible Tools to Feature | Category | Tools | | ----------------- | ----------------------------------------------------------------------- | | Chat interfaces | ChatGPT Edu (4o mini), Claude 3.5 Sonnet, Gemini 1.5 Pro | | Visual generation | DALL·E 3 in ChatGPT, Midjourney v6, Ideogram, Pika Labs video | | Storyboarding | Storyboarder (WonderUnit), Notion AI whiteboards, Figma AI plugin | | Planning | ChatGPT Advanced Data Analysis, Google Sheets + GPT add-on, Airtable AI | | Scriptwriting | Sudowrite, ChatGPT Edu Custom GPT “Script Doctor” | | Voice/Audio | ElevenLabs, ChatGPT Voice, Whisper Live | | Scheduling | Notion Calendar, AI Google Calendar Assistant | --- Would you like me to turn this brainstorm into a **structured 90-minute run-of-show** (with time blocks, goals, and materials) for tomorrow’s session? Perfect — those new directions push this session into a **richer hybrid of analog + digital media prototyping**, which aligns beautifully with the Learning Lab’s ethos. Here’s an expanded brainstorm structured around your three threads — **Nano-Banana & ingredients**, **line fidelity & AI visual precision**, and **pitch deck / lookbook assembly** — plus a few linking ideas that can make the whole workshop feel like one coherent experiment in *AI-supported creative pre-production*. --- ## 🍌 **Nano-Banana + Ingredients: AI Thinking with Tangible Media** *(for context: Nano-Banana = an LLM “sketching” playground that can integrate “ingredients” like text snippets, prompts, and small files)* ### 1. Ingredient-Based Prompt Crafting * **Physical Inputs Table:** lay out a “mise en scène” of analog items — postcards, markers, fragments of text, paper textures, screenshots, printed photos. Participants choose 3–5 and feed their short descriptions into Nano-Banana as “ingredients.” → Output: a storyboard seed, a concept treatment, or a lookbook moodline. → Meta-point: showing how *prompting as collage* parallels analog montage. ### 2. Analog–Digital Roundtrip * Participants **sketch an idea on paper**, photograph it, and upload it as an “ingredient.” Nano-Banana then expands the sketch into a concept or storyboard panel. → Demonstrates the creative feedback loop between analog thinking and LLM remixing. ### 3. Ingredient Archetypes * Introduce ingredient “types”: * 📄 *Textual (script excerpts, poetic fragments)* * 🎨 *Visual (color palette, texture, sketch)* * 🔉 *Aural (vibe descriptor, song lyric)* * 🧭 *Conceptual (theme, constraint, or emotion)* Participants label their ingredients, and Nano-Banana combines them algorithmically. → This can evolve into a structured creative taxonomy for future workshops. ### 4. Ingredient Remix Game * Teams exchange one ingredient with another group (physical or digital) and re-run Nano-Banana with the altered set. → The exercise shows how even small prompt perturbations shift the aesthetic dramatically. --- ## ✏️ **Line Fidelity & AI Visual Precision** ### 5. Line-Driven Visual Prompts * Show how to *condition AI image generation* on specific line drawings: * **DALL·E 3 “style reference” uploads** — use a scanned sketch or pencil drawing to constrain composition. * **ControlNet / Scribble mode** — demonstrate how edge maps guide Stable Diffusion output. * **Ideogram “match drawing style” prompts** — test prompt weighting for line vs. fill fidelity. ### 6. Progressive Constraint Pipeline 1. Start with a hand-drawn storyboard panel. 2. Generate a high-fidelity AI image that respects contours. 3. Regenerate with style modifiers (engraving, lithograph, blueprint). 4. Export variations to build a visual language library for the project. → You could display these on the projector as a “from hand to machine” timeline. ### 7. Physical–Digital Hybrid Output * Use tracing paper, light tables, or overhead projectors: * Trace AI outputs to reassert human gesture. * Rescan the traced version → feed back into AI → show how human touch reconditions the model. → Creates an embodied understanding of “co-authorship” between artist and AI. ### 8. Error Aesthetics * Deliberately push models into misalignment: * Over-contrast sketches. * Ask for “edge hallucinations” or “dream in ink.” → Opens discussion about productive failure and ambiguity as design tools. --- ## 📘 **Pitch Decks, Lookbooks, and AI-Generated Presentations** ### 9. Auto-Deck Builder * Use ChatGPT’s **Canvas or File Upload** to turn a concept document + 3 images into a **slide deck outline**. * Then refine slide-by-slide: * *Slide 1:* Title & tag line * *Slide 2:* Concept summary * *Slide 3:* Moodboard / visual references * *Slide 4:* Technical approach * *Slide 5:* Schedule / resources ### 10. Dynamic Lookbooks * Use **Notion**, **Figma**, or **Pitch.com** to live-assemble lookbooks with: * AI-generated visuals from earlier exercises * color palettes extracted by Claude or ChatGPT’s Vision tools * text snippets (loglines, tone statements, keywords) * Bonus: generate “tone comparison spreads” (e.g., *how it looks if Wes Anderson directed it vs. Wong Kar-Wai*). ### 11. StoryWorld Decks * Show participants how to create a “World Bible” deck — maps, mood shots, key props, and style cues. * ChatGPT or Claude can write world summaries in production-binder format. * Combine text + visuals in Google Slides or Figma live. * Encourage participants to include a “feeling palette”: emotions or temporal rhythms. ### 12. AI Deck Persona * Let Nano-Banana or ChatGPT adopt the persona of a **creative producer or studio exec**, and have participants “pitch” their deck to it. * The model gives structured feedback (“logline clarity,” “visual coherence,” “market pitch strength”). * Turns AI into a rehearsal partner for real-world pitches. --- ## ⚗️ **Hybrid and Meta Experiments** ### 13. “Reverse Ingredient” Challenge * Feed Nano-Banana a finished AI image or deck and ask it to infer what ingredients might have produced it. → Highlights the interpretive nature of AI “reading” media. ### 14. Concept → Script → Board Chain * Chain models: * ChatGPT: concept synopsis → Claude: script draft → DALL·E: storyboard panels → Nano-Banana: lookbook assembly. * End with discussion on interoperability and “pipeline literacy.” ### 15. Tangible Pitch Wall * Print AI-generated visuals on cardstock, pin them up, and let participants arrange decks physically. * Photograph final arrangements as analog lookbooks. * Optionally, use a webcam feed + CLIP model to auto-tag themes (“urban decay,” “ritual,” “thresholds”). --- ## 🧩 **Linkages to Broader Themes** * **Human–AI Co-authorship:** gesture, materiality, and iteration. * **AI as Translation Layer:** moving between modalities (sketch → prompt → deck). * **Studio Pedagogy:** how embodied and social practices can be preserved in AI workflows. * **Context Engineering:** designing metadata, tone cues, and “ingredient ontologies” for media projects. --- Would you like me to draft a **90-minute plan** that integrates these (e.g., Nano-Banana ingredients → sketch-to-AI → deck assembly → reflection), with suggested timings, prompts, and materials list (markers, cameras, scanning stations, etc.)? That’s *exactly* the right instinct — to use the studio’s equipment **symbolically and atmospherically** in Session 1 (“Prep”) without crossing into actual **production workflows** that you’ll reserve for “On Set.” The key move is to make the physical studio act as a **thinking instrument**, not a recording environment yet. Below are layered ways to bring in cameras, lights, screens, and projectors for *ideation, framing, and visualization* — rather than shooting. --- ## 🎥 **Cameras as Idea Machines (Not Capture Devices)** ### 1. Camera-as-Viewfinder * Mount a camera or phone on a tripod but keep it **unplugged from recording**. Let participants look through the lens to *frame compositions* for AI storyboards. → Then describe or photograph the framing for prompt input (“medium close-up from low angle, tungsten key, shallow DOF”). * This turns the camera into a *prompt-building reference tool*. ### 2. Shot Grammar Exercise * Place 2–3 cameras around the room aimed at blank backdrops. Participants walk between them to *feel* what “coverage” means. → They then script or prompt the same micro-scene from each camera’s POV. → Connect to later AI shot-listing in Session 2. ### 3. Lens-as-Metaphor * Offer lenses (or simulated focal lengths on phones) as *creative constraints*: “What’s the 24 mm version of your concept? What’s the 85 mm version?” → Participants translate those physical optics into AI prompt parameters (wide, telephoto, portrait compression). --- ## 🟩 **Green Screen as Concept Canvas** ### 4. Projection Surface for Imagination * Instead of compositing, use the green screen as a **live projection wall** for evolving AI imagery (storyboards, color palettes). * Run a looping slide deck of participants’ generated frames. * Lights low, slow fade transitions → transforms the studio into a “thinking cinema.” ### 5. “Standing in the Scene” * Invite participants to stand in front of the projected AI scene, *not for filming*, but to **embody scale, light, and composition**. * Have them describe how being “in” their AI image changes their prompt choices. * Take stills (optionally) for reflection, not production assets. ### 6. Green-Screen Palette Wall * Tape printed color swatches or keywords to the green surface. * These serve as analog “tags” for AI generation (“tone: elegiac,” “light: sodium vapor”). * Photograph the wall → upload to Nano-Banana as an “ingredient board.” --- ## 🖥️ **Screens and Projectors as Collaborative Instruments** ### 7. Multi-Screen Moodboard Loop * Set up the Lab’s projectors or screens to cycle through: * Participant prompts * Generated images * Live updates from Nano-Banana * Treat it like a *media aquarium*: ideas swimming across screens while teams work. * Great ambient energy; reinforces AI as part of the studio ecology. ### 8. Split-Screen Comparisons * Side-by-side projection: hand-drawn sketches ↔ AI renderings ↔ modified re-prompts. * Invite micro-critiques (“where does the machine misread the hand?”). * Builds visual literacy before Session 2’s technical compositing. ### 9. “Projection Mapping Lite” * Use one projector to cast simple geometric frames or grids onto paper or props on a table. * Participants sketch within those light contours — a pre-visualization of how light and frame interact. * Photograph the sketch under projection and feed it to the model for iteration. --- ## 💡 **Studio as a Pre-Production Playground** ### 10. Lighting as Metaphor * Set up 3-point lighting rigs but use them theatrically: * Toggle between warm/cool scenes as participants discuss mood. * Let the “lighting state” correspond to AI tone cues (“noir,” “pastoral,” “clinical”). * No recording — just embodied understanding of light language. ### 11. Spatial Storyboarding * Use masking tape on the floor to block out a scene. * Participants mark camera, subject, prop positions. * Label with sticky notes: “Close-up,” “Cutaway,” “Insert.” * Photograph layout and upload to AI for automatic diagram interpretation. → Serves as bridge to Session 2’s real blocking exercises. --- ## 🧩 **Integrative Nano-Banana + Studio Idea** ### 12. Ingredient Stations in Physical Space * Place QR codes or NFC tags around the room: * “Camera,” “Lighting,” “Set,” “Sound,” “Mood.” * Scanning one sends an “ingredient” to Nano-Banana. * Participants physically *collect* ingredients by moving through the studio. → Turns spatial exploration into prompt-building. --- ## 🧠 **Save for Session 2 (“On Set”)** To maintain a clear pedagogical progression: * **Avoid:** live recording, chroma key compositing, real-time transcription, or AI camera control. * **Do:** concept visualization, framing practice, projection feedback, embodied composition. This way, Session 2 can feel like a natural escalation from *imagination space → capture space.* --- Would you like me to sketch a **hybrid floor plan + time flow** showing how these activities could unfold across the Lab’s physical zones (tables, projection wall, green screen, lighting corner)? It’d give you a visual map of how the pre-production “studio ecology” could operate tomorrow. --- That’s a great and very strategic question. Editors, technical directors, and post-production people are actually *perfect* participants for a pre-production workshop—if you design it to foreground the *structural thinking*, metadata, and pipeline logic that make a production succeed later. Below is a brainstorm of **pre-production workflows that resonate deeply with editors, post supervisors, and technical crew**, grouped by mindset. All of them stay comfortably in the “Prep” phase—no cameras rolling yet—but tap into the habits and logics editors already live by. --- ## 🧩 1. **Thinking Like a Postperson in Pre-Production** Editors and tech leads are naturally *reverse-engineers*. You can frame pre-production as “designing the edit before you shoot.” ### a. Edit-First Storyboarding * Build a **“paper timeline”** using stills or AI frames: sequence them left-to-right on a wall or in Figma/Storyboarder. * Editors annotate: *What will this cut feel like? Where would you want coverage?* * Then feed that markup into ChatGPT or Claude to generate a **shot list** automatically from editorial intention. ### b. Data Management & Naming Logic * Simulate file-naming schema early: `PROJ_SC01_SH03_V01.mov` → show how LLMs can enforce or generate consistent naming conventions and metadata templates. * Editors immediately recognize this as sanity insurance for future workflows. ### c. Proxy Thinking * Generate **AI proxy assets** (temporary dailies): low-res, text-to-image stand-ins for scenes. → Editors can cut with these in mind to test pacing before footage exists. ### d. Conform & Metadata Design * Have participants define **metadata schemas** for assets *before production*: `scene, location, take_quality, emotion, sound_notes`. → Show how these schemas can later feed into automated transcription, tagging, or RAG workflows. --- ## 🧮 2. **Technical Planning & Pipeline Design** ### a. Folder Architecture Exercise * Groups design a **directory tree** that anticipates post workflows. → Then prompt an LLM to generate automation scripts that would create that tree on a drive. * Example: “Create a bash script that builds our media project folder structure.” ### b. File Round-Trip Simulation * “Fake ingest”: drop a few images/audio clips into a shared folder. * Ask AI to generate a *pre-flight report* (“missing metadata,” “duplicate names”). → Editors love this—it’s pre-production QA. ### c. LUT and Color Style Planning * Instead of grading, generate **AI lookbook stills** that express color space intention: “Teal-orange grade,” “bleach bypass,” “Fuji Eterna.” → Editors can imagine how the final look will cut together; colorists appreciate being consulted early. ### d. Audio Tone Previz * Use text-to-sound models or libraries to create **mood beds**—placeholder ambiences or emotional arcs. → Editors recognize this as temp music or reference tone that informs pacing. --- ## 📖 3. **Script-to-Edit Continuity Planning** ### a. Script Breakdown Automation * Run scripts through Claude or ChatGPT to extract: * Locations * Characters * Props * Shot tags → Feeds directly into scheduling and editing prep (what will need continuity checks). ### b. Continuity & Coverage Maps * Generate heatmaps or diagrams of scene transitions, coverage density, or expected cutting rhythm. * For example: “Where do we expect jump cuts vs. match cuts?” → Editors enjoy visualizing editorial energy before footage exists. ### c. Pre-Labeling for Transcripts * Build **tag taxonomies** editors will later apply to transcripts (topics, tone, characters). → Use AI to auto-suggest tags, then refine as a team. --- ## 🎞️ 4. **Editorial Aesthetics in Pre-Vis** ### a. “Edit Intent” Annotation * Take AI storyboards or Nano-Banana frames and have editors label: * “Cut on motion” * “Hold for 3 sec” * “J-cut / L-cut opportunity.” → Those notes can become structured prompt inputs (“Generate a storyboard anticipating J-cuts between dialogue lines.”). ### b. Montage Exercises * Give everyone a shared text corpus (say, student interviews or public-domain footage descriptions). * Have AI assemble *written montages*—juxtapositions, repetitions, visual rhymes. → Editors will instinctively recognize Eisensteinian principles encoded in language. ### c. Pre-Visualized Rhythms * Use **pacing simulations**: LLM generates timestamps or beats per minute for scenes, visualized as waveform-like timelines. → Editors discuss how this anticipates rhythm and emotion. --- ## 🖥️ 5. **Tools & Cross-Department Handshakes** ### a. AI-Assisted Paper Edit * Give them raw transcript text (from a hypothetical doc shoot). * Let ChatGPT condense it into a *paper edit outline* (soundbite order). * Editors evaluate and critique the AI’s editorial decisions. ### b. Pre-Production as Schema Design * Frame pre-production as **schema thinking**—what structured data will make post easy? * Shot logs * Audio notes * Scene metadata * AI-assisted ingest logs ### c. ChatGPT + Airtable + Resolve API demo * Show how you could, in theory, export an Airtable shot list → CSV → Resolve metadata panel. * They’ll instantly see the utility for post workflows. --- ## ⚙️ 6. **Integrating the Studio Hardware Lightly** Editors and techs will appreciate the **signal flow** aspects: * Use the green screen stage as a **metadata playground**: tag imaginary shots (“wide green setup,” “close interview setup”) and simulate ingest forms. * Project AI-generated look references on the screens—then label them with LUT intentions or exposure notes. * Run a camera tethered to a monitor purely to simulate *ingest path*, not capture. → “What metadata could travel with this file?” becomes a creative question. --- ## 💬 7. **Meta-Conversation Prompts** * *“If you could redesign pre-production to eliminate one post headache, what would it be?”* * *“How could AI help you get dailies that cut together better?”* * *“What metadata do you wish directors understood?”* These prompts let technical participants surface process wisdom while still playing creatively. --- If you want, I can sketch a **version of tomorrow’s run-of-show oriented toward editors/tech participants**, weaving in a few Nano-Banana “ingredient” stations and light studio use. Would you like that?