AI in VFX: where automation is changing the pipeline

By Michael MacLaren, Michael Knott and Grace Gunawan

The debate about AI and visual effects (VFX) has often been framed around disruption and displacement. Yet the reality is more nuanced. AI is not removing VFX as a key pillar of the entertainment industry. It is reducing the time and labor required for specific types of execution work, especially where tasks are structured and repeatable. That shift has distinct implications for different VFX studio tiers. This article explains where AI is entering the workflow, what kind of automation is actually involved and what studios and investors should do now. This matters because the strategic value will not come from adopting more tools but from redesigning the operating model around the new shape of work.

Where AI is changing the pipeline

To identify where AI and automation are entering the VFX pipeline, we used a structured AI agent to analyze 35 sources – primarily vendor documentation – and map 313 vendor-documented capabilities against an 82-step VFX workflow taxonomy. The strongest signals appear in steps where the work is clearly defined and quality can be checked against a known reference. Roto, clean-up, plate preparation, matte generation, denoising, asset upscaling and texture generation all fit these criteria.

The parts of VFX that depend on supervision, art direction, client-facing decisions, hero-shot judgment and cross-shot continuity remain much less exposed to documented vendor AI investment. AI is reducing the time required for execution work more than it is replacing the creative and supervisory work that holds a production together. The impact on the industry? Not less creative work, but more capacity for it – if studios redirect the time AI gives back toward the parts of the pipeline where judgment matters most.

From core to periphery

Two distinct zones of change are emerging: a contested execution core and a newly visible pre-production periphery. The execution core includes areas such as compositing and rendering, alongside animation and texturing. On a simple feature count, these areas look busy: vendors of different types are documenting capability across them. But the pattern is not always deep convergence. Some steps show strong single-vendor depth rather than broad market agreement. Foundry’s CopyCat in Nuke, for example, trains on artist-supplied frames and propagates the result across a sequence. That is a substantive machine-learning capability, but it should not be confused with proof that every compositing problem has become a multi-vendor AI battleground.

The clearest area of deeper machine-learning convergence is character work, especially character deformation, deformation and the simulation of cloth, hair and muscle. Epic Games and SideFX have each documented ML approaches in this area, using models trained on artist-supplied outputs. The common feature is narrow, supervised learning, not generic model generation.

The periphery looks different. Pre-production has historically been less productized by established VFX software vendors. Concept art, visual development, previsualization (previs), storyboarding and shot planning were not deeply integrated into the VFX software stack. That has created room for AI-native and adjacent vendors to enter. Runway, Luma AI, Alibaba’s Wan and Unity are targeting these earlier workflow steps with documented capabilities. The strategic point is not that these tools have already replaced production workflows. It is that they are entering stages where incumbents were less entrenched, and where studios can move closer to the point where a show is being shaped, not just where shots are being delivered.

Making generative AI production-ready

A further challenge sits across the pipeline. Generative tools can produce striking individual outputs, but production requires consistency: characters that hold across shots, repeatable results and art direction rather than repeated re-rolling. Orchestration tools such as ComfyUI and Weavy address this by chaining models and editing tools into repeatable workflows. They do not map neatly to a single VFX pipeline step, but may determine how quickly generative AI becomes production-grade.

AI will effect each studio tier differently

This same compression will not affect every studio in the same way. Premium independent studios are in the most defensible position, provided they restructure around the opportunity. Their pricing already reflects judgment and creative trust more than execution throughput. AI compression can allow premium studios to redirect capacity into stronger supervision and deeper creative development. The more interesting move is forward extension into pre-production. A studio that helps shape the look and concept direction of a production becomes harder to replace later in post-production, and more likely to become the trusted creative partner.

The mid-market is where the pressure concentrates. Mid-market studios have often competed by absorbing execution scope at moderate pricing. AI compression of roto, clean-up, plate preparation and compositing attacks the line items that have supported that model. At the same time, premium studios become more cost-competitive as execution work becomes faster, while budget studios become more capacity-competitive as their tooling improves. The middle iseffectively squeezedfrom both sides. There are two possible responses: move upward into supervision-led creative partnership, or restructure around a more cost-sensitive market if lower production costs unlock new mid-budget work.

Budget studios face the opposite opportunity to the premium end. They compete on volume, throughput and pricing discipline. AI can expand the segment of work they can win. The risk is over-extension. Higher shot throughput still requires supervision and quality control. Without that discipline, AI can increase revenue while damaging delivery and margin.

Turning AI capability into advantage

The priority is not to adopt more AI tools in isolation. VFX studios must map which steps in the cost base are most exposed to vendor AI investment, and what share of revenue depends on those steps. Studios then need to decide where spare execution time should go: more iteration, earlier supervisor involvement, pre-production participation or a lower-cost delivery model.

For investors, the question is not whether a studio “uses AI.” The most attractive studios will be those with the operating model to convert AI into capability-based advantage. That means looking for shot-level cost data and supervision capacity that can be deployed earlier in the production cycle, supported by modern pipelines and disciplined tooling adoption. The most exposed studios will be those sitting in the mid-market without a clear route either upward into creative partnership or across into disciplined high-volume delivery

The answers to these questions will dictate whether AI is still improving isolated workflow steps or beginning to reshape the economics and structure of VFX production. The compression is already visible. The AI advantage comes from understanding where value is moving and redesigning the operating model.

Conclusion

AI is changing the economics of VFX, but not in the way many expected. The competitive advantage will not come from adopting the latest tools alone, but from redesigning operating models, reallocating creative capacity and positioning earlier in the production lifecycle. As the market evolves, studios will need to make strategic decisions about where to compete, how to capture value and how to build resilience in an increasingly differentiated landscape. This is where Roland Berger can helpcombining deep industry expertise with strategic and operational transformation capabilities to help entertainment businesses navigate technological change, strengthen competitive positioning and turn AI-driven disruption into sustainable growth.