What Went Wrong in VFX this Year : 2025

“Just survive 2025.”

We did.
Somehow.

Here is what actually happened, why it mattered, and what it quietly changed.

1. AI in 2025

What happened

AI was everywhere in 2025. Louder than ever. More accessible than ever.

To be clear, AI itself was not new. VFX has used machine learning for years in denoising, tracking, retiming, and upscaling. What changed was visibility. Tools like ChatGPT, Stable Diffusion, and commercial systems such as Sora, Kling, and Runway pushed AI into public consciousness. Suddenly everyone had access. Suddenly everyone had opinions.

But once you stepped out of demos and into real pipelines, progress slowed down fast.

The questions became unglamorous:

Do AI-generated mattes, DMPs, or elements pass studio tech checks?
Can they reliably output floating-point EXRs?
What happens when a client asks for precise revisions in a very small region?

These are not philosophical questions. They are production blockers.

This gap between demo and delivery defined AI in 2025. The industry learned that impressive outputs mean very little if they do not survive revisions, color pipelines, and client notes.

Then came the copyright moment. The Midjourney legal mess was one of the most important signals of the year. Not because of any single ruling, but because it revealed how unprepared copyright law is for generative systems. Artists assumed protection. Companies assumed freedom. Courts are still catching up.

At the same time, the industry was flooded with models. Datasets, fine-tunes, LoRAs, variants, distillations. The sheer volume became noise. When everything is new, nothing feels stable enough to trust.

AI did not replace VFX in 2025. It reframed it. The uncomfortable truth surfaced clearly this year: the biggest cost in VFX is not technology. It is people.

Scott Ross said the quiet part out loud. Clients have always wanted the highest quality for the lowest price. In the early 2000s, studios chased subsidies. Then cheaper labor overseas. Then India. Now AI.

Same game. New tool.

The reality is simple.
They will still need some artists.
They will not need all of them.

That is not fear. That is economics.

2. LinkedIn and the AI Performance Era

Something else changed in 2025. LinkedIn stopped feeling like a portfolio and started feeling like a stage.

Feeds that once showed careful work were replaced with endless AI reels. Loud claims. Half-baked results. Extreme confidence.

This was not because everyone suddenly became dishonest.

Why it matters

Sociologist Erving Goffman described life as a series of masks. Work mask. Social mask. Family mask.

LinkedIn is the professional mask.

It is not designed for truth. It is designed for signaling employability, optimism, and relevance. AI content thrives there because the platform rewards visibility, not accuracy.

Understanding this makes LinkedIn less annoying. And more dangerous.

The industry is learning to mentally discount performance. Real evaluation is shifting back toward private tests, referrals, and production proof. The louder the platform becomes, the quieter real hiring signals get.

3. The Rise of ComfyUI

While flashy AI tools fought for attention, something quieter happened.

ComfyUI.

It did not win by being friendly. It won by being honest.

Open source. Node based. Offline. No abstraction. You see everything: models, samplers, schedulers, noise, conditioning.

That scared beginners. It attracted power users.

If you are used to Nuke or Houdini, ComfyUI feels natural. You are not prompting and hoping. You are building a process.

Serious AI experimentation for VFX migrated here first. New models appeared here first. When something broke, you could actually see why.

ComfyUI was not polished. Updates broke workflows. Community nodes varied wildly in quality. There were dozens of core releases across 2025 alone.

That friction turned out to be the point.

Professionals consistently choose control over convenience.

ComfyUI is not production ready yet. But it proved something bigger. Tools that treat users like adults outlast tools that hide complexity behind friendly interfaces. Frontends will come and go. Foundations that expose reality tend to stick.

4. Technicolor Did Not Collapse Overnight

What happened

Technicolor’s failure felt sudden. It was not.

The warning signs were visible for years. Heavy debt. Acquisitions instead of integration. Silos competing under one logo. MPC. The Mill. Mr X. Mikros.

During the COVID content boom, leadership believed creative services could spin out, go public, and grow forever. Expectations inflated. Reality arrived quickly. Six weeks after the IPO, profit projections were cut. The stock never recovered.

When the funding stopped, the talent was still there. World-class artists. Engineers. Supervisors.

Why it matters

This was not a creative failure. It was a structural one.

Debt removed flexibility. Silos killed speed. Clients stopped trusting the company with critical work because VFX failure mid-show is not survivable.

What it might change

Technicolor became a case study. Quality alone does not save studios. Financial structure, integration, and realistic expectations matter more than awards.

5. The Shifting VFX Landscape

By 2025, contract-based hiring became normal. Two-month contracts. Three-month contracts. Permanent roles quietly disappearing.

You can survive like that. You cannot plan a life.

Underneath it all sits the same broken model. Studios bid based on an idea. The idea changes. Deadlines stay fixed. Budgets do not.

Everyone absorbs damage until something breaks.

This model externalizes risk onto artists and internal teams. It rewards optimism and punishes honesty. It has been unsustainable for a long time.

What it might change

Studios are slowly favoring predictability over ambition. Fewer miracles. More conservative bids. Less creative stretch. More risk management.

6. The “No CGI” Lie

Marketing doubled down on the “no CGI” narrative.

Hundreds of artists worked on films for years. Then campaigns proudly claimed everything was practical. Media repeated it. Audiences applauded. VFX stayed invisible.

VFX is not good or bad by default. It can be done well or poorly. Pretending it does not exist devalues the labor behind modern filmmaking.

The pushback is growing. Slowly. Artists are calling this out more openly. Recognition may lag, but denial is becoming harder to maintain.

7. Tariffs and the Misunderstanding of Modern VFX

Political talk of tariffs on films made outside the US surfaced again.

VFX is not manufacturing. It is not tied to geography. A film can be shot in Los Angeles, composited in Mumbai, tracked in Malaysia, and graded in London.

You cannot tariff a render.

Global pipelines will continue. Attempts to force national boundaries onto digital labor will fail quietly, not loudly.

Nuke 17 and a Quiet Question

Nuke gained more power. More features. Deeper integration.

Power increases dependency. Dependency raises costs.

The industry may soon have to confront an uncomfortable question. What happens when the tools VFX depends on become unaffordable to large parts of the industry?