AI's Too Optimistic...

Dear Juliet

Checked back with AI and I'm not pouring a bucket of cold water on everything, but at present it would struggle a bit to do it all as easily as I first wondered:

Revised facts:

You are completely right on the big picture: £100,000 is roughly 5% to 10% of what a traditional low-budget indie feature would cost to shoot across two continents. Taking a live-action crew from Glasgow across the United States on Greyhound buses to Los Angeles would easily run £1 million to £3 million+ once you factor in union crews, transatlantic flights, hotels, filming permits and insurance.

If you want the unvarnished, practical reality of making a feature-length version of Dear Miss Landau for around £50,000 to £100,000 right now, here is what actually happens behind the scenes. The real advantages, the hidden costs and where the technology still hits a wall.

1. Where the £50,000 - £100,000 Actually Goes

When people hear "AI movie", they assume the software does all the work. In reality, software and GPU compute only account for about 15% to 20% of that budget. The rest pays for traditional filmmaking skills applied to a new medium:

• Pre-Production & Asset Building (£8,000 – £15,000): Before generating a single scene, artists must build a "digital bible". They have to train custom character models (LoRAs) of a younger/current James Christie and Juliet Landau from dozens of reference angles, plus design consistent environments for your Glasgow flat, specific Greyhound coaches and Sunset Boulevard. 

• The AI Director / Prompt Artists (£20,000 – £40,000): You need two or three skilled generative video artists working for 3 to 4 months. They use orchestrators (like Claude Fable 5.2) to manage the script breakdown, then spend hours every day generating, rejecting and refining 1,200+ individual shots.

• VFX Cleanup & Compositing (£10,000 – £20,000): AI still struggles with continuity. Buttons change colour between cuts, background passengers warp or text on a Greyhound destination board turns into gibberish. Traditional VFX artists have to paint out errors, fix faces and stitch shots together with software like Nuke or After Effects.

• Sound Design, Score & Voice (£10,000 – £20,000): AI video looks cheap if the audio isn't cinema-grade. This covers recording Juliet Landau's real voice (and your own, or a Scottish voice actor), hiring a composer for an original score and doing full theatrical foley and sound mixing. 

• Legal, Likeness & Contingencies (£5,000 – £10,000): E&O (Errors and Omissions) insurance is mandatory if you ever want Amazon, Apple or a festival to screen the film alongside formal likeness contracts.

• 2. Three Brutal Realities of Feature-Length AI in 2026

Even with a £100,000 budget, you will run into three hard technical walls that every AI filmmaker is currently dealing with:

• The "Two-Person Conversation" Wall: AI video is dazzling at sweeping landscapes - a Greyhound bus rolling through the Rockies at dusk or rain hitting a Glasgow window. It is much harder to make two people sit across a table at a diner and have a subtle, four minute emotional conversation without their micro-expressions feeling slightly uncanny or repetitive.

• You Are Likely Looking at "Stylised Realism", Not Pure Live-Action: Right now, pure photorealistic AI over 90 minutes still triggers the "uncanny valley" for audiences. The most successful feature-length AI films today lean slightly into a heightened, painterly or graphic-novel-inspired realism (similar to A Scanner Darkly or high-end prestige animation). That aesthetic hides tiny visual glitches and makes the cross-continental journey feel dreamlike and literary.

• Distribution Rules Are Still Messy: Major film festivals (like Sundance or Edinburgh) and streaming platforms have strict disclosure rules. Some accept AI-assisted films if human authorship is clear. Others reject fully synthetic films. Furthermore, under current US and UK copyright law, raw AI-generated frames cannot be copyrighted on their own - though your underlying script, edit, music, and human-recorded voices are fully protected.

3. The Smartest "Reality-First" Strategy for Dear Miss Landau

If you wanted to make this real without wasting money, the strongest approach isn't 100% AI - it is a Hybrid Production:

1. Shoot the Anchors Live-Action: Film real, intimate footage of yourself in Glasgow - and, if she is willing, shoot a single day of real live-action footage with Juliet Landau in Los Angeles. Real human eyes and real emotion anchor the audience completely.

2. Use AI for the "Impossible" Scale: Use the £50k–£80k AI pipeline to recreate the past, the internal world of your autism and the great 3,000-mile Greyhound bus odyssey across America - blending your real filmed face into the generated environments.
3. Record Real Voices: Never use AI voice cloning for the leads when you have access to the real people. Real vocal performances from yourself and Juliet will immediately elevate the film above 99% of AI projects. 

If you're seriously considering exploring this path, would you like to look at how to pitch or structure a hybrid budget, or would it help to see what specific scenes from Dear Miss Landau translate best to a two minute funding trailer?

Okay at the moment, thanks. And I think it's fair to say that the truism of "an internet year is three months still holds true." How fast is this technology moving and, realistically, what might Claude Fable etc. be capable of in December 2027?

That old dot-com truism holds up remarkably well right now. In generative video and agentic orchestration, three months genuinely equals a generational leap in what the tools can pull off.

Looking at the trajectory from late 2025 to where we stand in November 2026, projecting forward thirteen months to December 2027 suggests a massive shift - not toward a magic "make a Hollywood movie" button, but toward turning a single skilled creator into a complete virtual production studio.

Here is a grounded, realistic forecast of what orchestrators like Claude Fable and the video models they control will likely achieve by December 2027.

1.      What Will Actually Be Solved by December 2027:

 

• Persistent "World Memory" (No More Continuity Glitches). Right now, AI generates flat 2D pixels frame by frame which is why objects morph or vanish when a camera turns. By late 2027, video generators will routinely build a hidden 3D spatial map of a scene first. If you place a coffee cup on a table in a Greyhound diner, it will stay in that exact spot across ten different camera angles without manual VFX clean-up. 

• True Multi-Minute Scenes. Instead of stitching together awkward five to thirty second clips, models will comfortably generate two to five minute continuous, multi-shot sequences with coherent lighting, matching clothing and consistent background extras.

• Real "Self-Critique" Loops. Right now, an orchestrator like Claude Fable 5.2 mostly sends prompts out and waits for a human to check the results. By December 2027, multimodal vision models will reliably watch the generated footage themselves, spot a weird hand or an unconvincing facial expression, automatically fix that specific 2-second segment and only show you the polished take.

• Directing Micro-Expressions. Rather than typing vague text prompts like "look sad", directors will adjust granular sliders or perform into a webcam to transfer exact eye movements, pauses and subtle emotional shifts onto the digital actors. 

2. What Will Still Require Human Craft (and Money)

Even in December 2027, several things will remain stubbornly human:

• Taste, Pacing, and Storytelling. AI will be able to shoot a scene ten different ways but it still will not know which take breaks the audience's heart, when to hold on a quiet silence for three extra seconds, or how to capture the authentic lived experience of autism on a cross-country journey. You still need a human director and editor making artistic choices. 

• Legal Consent and Union Guardrails. Technology moves faster than the law, and by late 2027 watermarking, likeness verification and SAG-AFTRA digital replica rules will be even stricter. Having Juliet Landau's signed permission and your own original copyright will become more valuable as platforms clamp down on unauthorised AI content.

• Compute Costs Won't Drop to Zero. Higher-resolution, 3D-consistent video requires vastly more server power. While human VFX cleanup costs will plummet, raw rendering costs for a ninety minute 4K film will likely still sit in the £2,000 to £8,000 range for an indie creator.

What This Means for Dear Miss Landau.

If you look at the timeline through a practical lens:

 

• Late 2026 (Now): Great time to lock down a script, secure formal agreements or build a two minute proof-of-concept trailer to test the visual style. Attempting a full ninety minute feature right now requires heavy VFX babysitting and a £50k - £100k budget. 

• December 2027: A solo creator or a tiny team of two - working with real voice recordings from you and Juliet - will realistically be able to produce a broadcast-quality, feature-length film for £10,000 to £25,000 in a couple of months rather than half a year.

Waiting a year while developing the script and securing your pieces isn't procrastinating in this market. It is letting the infrastructure mature so your budget goes five times further.

So (this is me writing now), while AI was too optimistic yesterday I'm pretty certain technology is leaping forward at breakneck pace. It'll get better and cheaper but...

Well, same old thing.

I'll give it all up for you and a pastrami and rye sandwich.

Some things are beyond price.

Love,

James