Generative AI is rapidly changing VFX, but moving beyond hype requires tools and techniques that give artists more control. This course offers a deep, technical look at the Stable Diffusion ecosystem, taught by Emmy and VES Award–winning VFX Supervisor Eran Dinur (Hereditary, The Wolf of Wall Street, Boardwalk Empire). This 10-class course (4h 27m) is for VFX artists, generalists, matte painters, previs/vis teams, and technical directors who want to control and understand generative AI.
A Practical, Production-Focused Approach to Generative AI
Unlike prompt-based tutorials, this course covers the building blocks of Stable Diffusion and how to assemble, guide, constrain, and enhance them using node-based tools. Working inside ComfyUI and InvokeAI, Eran walks through:
- Constructing custom workflows from scratch
- Understanding checkpoint models, samplers, latent space, CLIP, and VAE decoders
- Building tileable texture generation tools for 3D workflows
- Using controlNets, preprocessors, and IP Adaptors to guide imagery
- Driving AI output with CG passes (beauty, Z-depth, normals, cryptomattes)
- Generating fully-textured 3D assets using recent diffusion-to-mesh techniques
- Leveraging InvokeAI’s layer-based inpainting/outpainting for matte painting & comp
- Applying segmentation and upscaling tools for high-quality outputs
This practical toolkit fits modern pipelines. Use it to enhance matte assets, augment comps, develop lookdev elements, or generate textured 3D models for previs and environments.
Why ComfyUI & InvokeAI?
Stable Diffusion’s open-source ecosystem evolves rapidly. ComfyUI and InvokeAI are two of the most capable and production-friendly SD platforms today. This course helps you:
- Install and configure complex ComfyUI environments.
- Work confidently with third-party custom nodes and dependencies.
- Understand InvokeAI’s strengths for matte painting and interactive inpainting.
- Evaluate SDXL and other high-fidelity photoreal models.
- Apply multi-modal guidance: image → depth → normals → masks → style
If you are seeking repeatable, documented, and studio-ready workflows rather than one-off AI experiments, this course is designed for you.
Course Breakdown
Below is a snapshot of what Eran covers throughout the 10 classes:
Class 1: Introduction & Setup
Comparing ComfyUI and InvokeAI, environment setup, and managing dependencies.
Class 2: Essential Building Blocks
Constructing the first SD workflow, understanding models, samplers, latent noise, and core parameters such as CFG, seed, and step count.
Classes 3–4: Building a Seamless Texture Tool
Developing a robust texture-generation workflow in InvokeAI and ComfyUI, integrating controlNets, preprocessors, and SDXL features.
Classes 5–6: Guiding AI with CG Elements
Using Maya, Nuke, and ComfyUI together to drive layout, lighting, and composition via 3D passes, cryptomattes, and IP adaptors.
Classes 7–8: Generating 3D Assets
End-to-end AI asset creation: installation of HunYuan-based tools, image-to-mesh workflows, de-lighting, UVs, multi-camera projection, and final export.
Class 9: Inpainting with InvokeAI
A focused look at InvokeAI’s layer-based tools for matte painting and compositing.
Class 10: Segmentation & Upscaling
Using the Florence model for image-to-language segmentation and comparing advanced upscaling methods.
About the Instructor: Eran Dinur
Eran Dinur has more than 40 feature and television credits, including Boardwalk Empire, Hereditary, The Wolf of Wall Street, and Uncut Gems. His experience spans on-set supervision, studio leadership as Head of 2D at FuseFX, and industry-wide training initiatives. He is the author of The Filmmaker’s Guide to Visual Effects and The Complete Guide to Photorealism, and an adjunct professor at the School of Visual Arts (NYC). His technical depth, clarity, and production-oriented thinking make him one of the most respected educators in the VFX field.
Who This Course Is For
- VFX and compositing artists adopting AI-assisted workflows
- 3D generalists and environment artists
- Matte painters and concept artists
- Technical directors and pipeline developers
- Anyone wanting to build structured, controllable generative AI tools should not rely solely on prompting.
No prior experience with generative AI, ComfyUI, or InvokeAI is required. However, participants should have a basic familiarity with Nuke, Maya, and Photoshop to follow the technical workflows presented.
Start Building Production-Ready AI Workflows
AIF201: Generative AI for VFX with ComfyUI & InvokeAI is available now for fxphd members. This course is more than an AI feature tour. It provides a technical foundation for the use of generative methods in VFX pipelines. If you want to understand the tools and use them precisely, this is an essential training resource. View the course on fxphd