SoulScape Day 2 — AI Production, Soul, and the Emerging Studio

Day two went deeper into the craft and business of AI filmmaking — from directing workflows and copyright to interactive narrative and what it means to produce with intention.

Things Learned From Making AIGC Projects

A practical session on the gap between prompting and quality — and how to close it.

The core problem: prompting alone makes it hard to consistently hit a target quality. Most people treat each generation as a lottery (what was called "gacha iteration") — running dozens of attempts and picking the least bad result. This produces mediocre 60–70/100 outcomes at high cost.

The better approach is layer iteration: treat each output not as a final result but as an intermediate state in a continuous construction. Focus each iteration on exactly one problem — composition, lighting, materials, or character details — and build up in stages. This requires only 3–5 trials per layer and consistently yields 90–120/100 outcomes.

Approach	Outcome Quality	Cost
Gacha iteration	60–70 / 100	High (many random attempts)
Layer iteration	90–120 / 100	Low (3–5 targeted trials)

A graph-based layout system helps here — you can organize all options at the same stage side by side, making it easy to compare and select before moving forward. A typical layer sequence:

Face appearance
Makeup refinement
Outfit and wardrobe

Panel: The Director's New Toolkit — Directing AI

Panelists: Jesse Z (MC), Philip Metschan, David Imbed, Maria Haras

How do you maintain quality control over AI-generated content?

David: Build a system with clear team roles. Use AI as you'd use CGI — as a production tool within a professional film structure, not a replacement for the director's vision. The key is establishing communication standards within the AI workflow.
Philip: Better team-oriented tooling is still needed. Right now most tools are built for individuals.
Jesse: Creators shouldn't have to manually check every frame. Automated quality and style evaluation pipelines are the right answer — freeing the director to focus on creative decisions, not quality assurance.

Idea worth pursuing: a persistent "style prompt" or visual style definition that governs all parallel work on a project — the equivalent of a visual bible, but enforced programmatically so collaborators don't drift from the intended aesthetic.

AI can produce visually impressive but emotionally hollow content. How do you inject soul?

Philip: Break the story into small, controllable pieces. Smaller units give you tighter creative control over each emotional beat.
David: Film language matters. The feeling of a shot — timing, intensity, the specific weight of a moment — has to be intentionally engineered. Motion capture is one path to getting real human performance into the system.
Jesse: The writing is the soul. Polish the story, not the prompts. If you can express what you feel with precision, the model will reflect that feeling back. Stop treating frames as the primary creative object.

How is AI changing storyboarding and pre-production?

David: Significantly more efficient. AI is genuinely useful here.
Maria: AI can generate concept art early enough to pitch tone and visual direction to buyers — before you have anything real. Traditional Hollywood still weighs story quality above all, but AI helps you show rather than tell at the pitch stage.
Jesse: AI works as a pre-visualization layer. You can mock up CG effects or entire sequences to pressure-test ideas before committing to real production. Some producers are now building full rough-cut mockups of films to exchange ideas with studios before greenlight.
Philip: Decisions are the most valuable thing on a production, and traditionally they get made under pressure on set without real thinking time. AI gives you a buffer to make better decisions earlier.

Where do AI tools still fall short?

Maria: Current models don't have a strong grasp of film language. Too many shots come out technically competent but cinematically mediocre.
Jesse: The root cause is in training: models are trained on whole films without scene-level labels or professional cinematographic annotation. The models don't understand shot grammar, pacing, or visual storytelling the way a DP does. The fix is refined, expert-labeled training data — and agents built specifically for film production contexts.

What will AI never take over?

David: AI is a tool. It doesn't replace authorship.
Philip: AGI could theoretically generate original creative content one day — but that's a different conversation.
Jesse: The market will shift toward short-form. And eventually: everyone will have their own studio.

Where does originality come from?

Maria: The reason to use AI is because you have a story to tell. In this era, intention drives everything.
Philip: You have an image in your head. That image needs to come out. The tools just change how it gets there.

Talk: Build World from Scratch

A session on the challenge of creating coherent, persistent fictional worlds in AI production.

The core thesis: A world is not a backdrop — it's a character. Place has personality. Space carries memory and history.

Current 2D pipeline pain points:

No spatial coherence: Images generated separately have no shared spatial logic.
No memory or rules: There is no persistent state. Each generation starts from zero.
Spaghetti pipelines: Fragmented outputs from multiple tools force unpredictable manual assembly.

What we actually need:

Persistence and editing: Worlds that can be revisited and modified without starting over.
Spatial reasoning: Understanding of how spaces relate to each other.
Scalable coherence: Consistency that holds across a full feature, not just a single shot.

Panel: The Studio Perspective — Integrating AI

Panelists: Evan Bailey, Renard T. Jenkins, Koh Terai

What actually defines a studio?

Two things: financing and distribution. A studio functions more like a bank than a creative entity — it packages and bets on projects.

Renard: The business model needs to shift. Studios may need to reposition as distribution-first companies rather than production-first ones. Independent filmmakers enabled by AI will force that change.
Koh: Came up through CS → Technical Director, now focused on reducing friction in getting production data AI-ready. The cleanest wins are backend changes that don't require changing the user interface — tools that slot into existing workflows invisibly.

What will AI actually change at the studio level?

Renard: There's a lot of experimentation happening, but the real key isn't cost-cutting — it's leveraging creator value more effectively. Some C-suite roles that exist to make high-cost judgment calls may be replaced as data gets better at predicting outcomes. The current gap is people who understand both film and technology. Creative technologists — people who can actually work in ComfyUI and speak the language of both worlds — are the most valuable hire right now.

Copyright: the unresolved problem

Studios require copyrightable content. AI-generated video, as it stands, is not clearly copyrightable in the US. This creates a structural tension at the core of any AI-native studio strategy.

Renard: The likely resolution involves quantifying the percentage of human creative contribution. The CGI precedent matters — CGI-assisted work is still considered human-authored because humans made the expressive decisions. The problem with generative AI is that the training data itself includes copyrighted material, which complicates the entire data flow.
Koh: Every asset in the pipeline needs a traceable lineage. Full copyright accountability across the data used to train, fine-tune, and prompt a model is table stakes for professional production. (Worth exploring: systems that automatically log and output copyright attribution weights from generated content — a provenance layer baked into generation.)

Panel: Beyond the Play Button — Interactive Narratives

Panelists: Coco Chen (MC), Stephan V. Bugaj, Cindy Fabian, Patrick Riley

What does an interactive AI film actually look like?

Stephan: Interactive film lets you change the path of a story and get different outcomes. It contains deliberate interaction points — but the structure is still authored.
Cindy: "Interactive experience" is a better label than interactive film. The goal is creating a genuinely new form, not replicating what games or film already do.
Patrick: High confidence this stays a moderated experience — someone designs the interaction space. Full autonomy isn't the right model. Human curation stays in the loop.

Real-time video generation — what does it unlock?

Stephan: Once real-time generation is viable, you can change style, storyline, or character state dynamically. But most people don't want full creative autonomy — they want a playable world with meaningful constraints. Inhabit a character, have some agency, but not be the author of everything. The distinction between people who want to create worlds and people who want to inhabit them matters enormously.
Cindy: Real-time conversation with characters — genuine interaction rather than branching menus — is the compelling near-term experience.
Patrick: Characters that respond authentically to real-world situations. That's the test.

Will small teams make great interactive content?

Stephan: Talent still determines quality. The bar hasn't dropped — the tools have gotten better. Skilled artists still need to train the next generation of creators.
Patrick: Teams will get smaller. But the craft requirement goes up, not down.

What devices host this new medium?

Stephan: TV and phone are the realistic near-term surfaces. VR headsets face too many friction points for mass adoption.
Cindy: Theatrical film is still growing — family moviegoing is one of the highest-growth segments.
Patrick: VR has real potential for the kind of presence that makes interaction feel meaningful — but hardware adoption remains the constraint.

How do independent creators finance interactive projects?

Stephan: He made a 23-minute film for $2,000. The honest advice: just start. Investment in story is low — build the thing first.
Patrick: Study how others have gotten financed. Understand what buyers and platform holders are actually looking for. Build a clear picture of the business model before you're deep in production.

Talk: Human-Cyborg Relations — Riff With Robots to Achieve Your Vision

Speaker: Stephan V. Bugaj | stephan@bugaj.com | Discord: @lhooqtius

A manifesto framing for working with generative AI as a creative partner.

GenAI is a dispassionate note giver.
GenAI is a tireless editor.
GenAI makes spectacle affordable.
GenAI removes gatekeepers and lets you speak directly to audiences.
GenAI does nothing creative without your artistic guidance.
GenAI expands your capabilities.

AI gives more voices a chance to be heard. The artists who will stand out are the ones who bring the most of themselves to the collaboration.

Practical takeaways:

Use LLMs as an early quality-check judge before expensive production steps.
Interactive pipeline systems are more powerful than one-shot prompting.
In one project: 3,500 generated shots were used to produce the final cut. AI makes that kind of exploratory volume feasible.
The more you bring to the process, the more creative control and ownership you retain.

On the future:

AI trained on the full breadth of human creativity can multiply individual human creativity — but only if a human is driving. Without AI, your commercial ceiling as an artist is lower. Without people, AI produces nothing but banality. The soul of the creative engine is still human.

Panel: Early Adopters — Surviving the Wave in Real Time

What do people get wrong about AI filmmaking?

There's still enormous work required to produce a finished film. AI handles some of it, not all of it.
AI hallucinates. Vigilant direction and review are non-negotiable.
Every component of traditional filmmaking still exists: you're just executing each one with new tools.

What traps do AI filmmakers fall into?

Chasing virality over craft. Algorithmic attention is not the same as an audience that cares.
Not showing work. In this industry, getting hired still depends on a visible body of work. Build a portfolio aggressively.

Hollywood is genuinely changing — the consensus in the room was that this is the filmmaker's year to move.

How do you price yourself as an AI filmmaker?

Think through three questions:

What is the actual compute cost to deliver this?
Am I learning something that builds long-term capability?
Is there backend participation (equity, residuals) that offsets a lower upfront fee?

And the honest filter: Do I actually care about this project? Misaligned taste is a real cost.

Reference numbers shared:

High-end commercial: $225K pre-production + $150K post + talent ≈ $500K total
Average 4–5 minute / 30-second spot: $78,000–$80,000

Talk: Producing with AI + Soul

Speaker: Evan Bailey

How AI is changing the producer role

Producing is structurally similar to running a startup. The classic "package" that unlocks a project has four components: IP, talent, financing, and distribution. Secure any one of them and you have a conversation.

In the past five years, engagement has become a fifth component. Streaming and YouTube changed buyer expectations — projects without a pre-existing audience or a name with a fanbase face an uphill battle. 100% of the greenlight pressure flows toward projects with demonstrated demand.

Two ways to secure IP and talent

Option and purchase: Pay an option fee for exclusive development rights, with a pre-negotiated purchase price on exercise. Expensive, but clean.
Shopping collaboration / if-come: Develop together with the rights holder, sell together, neither party can move forward independently. Lower cost, more aligned incentives.

How AI changes the packaging conversation

Have an explicit, open conversation with rights holders about AI involvement upfront.
Understand where buyers and key stakeholders sit on AI before you're mid-development.
Assumptions made silently become deal-breakers discovered late.

Four ways to finance a project

Studio/streamer setup — they pay, they own
Co-production — multiple industry partners share cost and risk
Crowdfunding — Kickstarter, Reg CF
Self-financing

Copyright is the foundation

If you create original IP, involve a human creator in a documented, registrable way. Video generated purely by AI may not be copyrightable — but the character, story, and script created by humans are protected. The safest approach: ensure every layer of the creative work has documented human authorship at the expressive level.

Trust is the operating system of every team

Every project develops its own culture. That culture either enables or destroys the work.

You can recover from a mistake. You cannot recover from broken trust.
Decide how AI will be used before people join the project. Document it. Be explicit.

Panel: AI Cinema as a New Art Form

Panelists: Chandan Perla, Maria Haras, Dorothy Pang, Ben Chang

Does AI have a voice of its own?

Dorothy: A true artist's voice comes from lived individual experience. AI doesn't have that — it doesn't have a soul. The useful historical parallel: when photography arrived in the 1800s, painters rejected it. Photography eventually found what it could do that painting could not, and became its own art form. AI cinema's job is the same — find what it can express that nothing before it could.

If a film makes you feel something, it does the job.

Ben: Art is the intent to make something that creates feeling. The medium is irrelevant to whether that intent can be achieved. What makes something art is not the tool — it's the decision to make someone feel.