Master Generative AI for Film and TV

The six parts

Each part is a standalone 4-hour live session on a Monday, 10:00 AM to 2:00 PM Toronto time (ET). Take one part or the whole series. If you lack a prerequisite, the recording of the earlier part covers it, and we recommend it for the full picture. Each part teaches the strongest open tools at its date, so the exact lineup can change.

Part 1: Foundations · ran September 28th, 2026, available as a recording
How a model learns and generates, the open-source landscape, ComfyUI installed live, a first controlled image, and the ethics and law of generative AI. No prerequisite.

Part 2: Image generation · Monday, October 26th, 2026
Text to image, editing, control and upscaling, with current open image models. Prerequisite: a working ComfyUI.

Part 3: Video generation · Monday, November 30th, 2026
Text to video, image to video, editing, control and upscaling, with current open video models. Prerequisites: ComfyUI and image basics.

Part 4: 3D generation · Monday, January 25th, 2027
Text to 3D and image to 3D, as meshes or Gaussian splats (3D scenes made of soft points), and how to control the result, with current open 3D models. Plus an AI assistant that builds in a 3D application through MCP (Model Context Protocol, an open standard that lets an AI assistant drive other software). Prerequisites: ComfyUI and image basics.

Part 5: LoRA training · Monday, February 22nd, 2027
Train a LoRA, a small add-on that teaches a model your own style, character or effect: the dataset, the captions, a training run on a rented GPU, and how to judge the result. Prerequisites: ComfyUI, image basics, and 25 to 50 images or clips that you own or may use.

Part 6: Pipeline integration · Monday, March 22nd, 2027
ComfyUI in a studio pipeline: colour management, bit depth, EXR and HDR, bridges to other applications such as Houdini and Nuke, cloud workflows, the ComfyUI API and SDK, and MCP connectors that let AI agents drive your tools. Prerequisites: ComfyUI, plus image and video basics.

Dates in your timezone

Upcoming live sessions, 10:00 AM to 2:00 PM Toronto time (ET), shown across timezones.

What you need: your GPU or Runpod

On your own machine

• An NVIDIA GPU with at least 16 GB of VRAM for the hands-on work.
• Part 3 (video): 32 GB of VRAM is ideal.
• Part 5 (LoRA training): a 96 GB card such as an RTX PRO 6000 is ideal.

No suitable GPU? Use Runpod
You can follow every part on a cloud GPU rented on Runpod, with ainvfx-runpod, the open-source toolkit we use internally: one command starts a pod with ComfyUI and the models of the session. You pay Runpod directly for the GPU time you use. New to Runpod, or to the toolkit? Tell us when you enroll, and we help you set it up before the session.

Watching only
You can also attend without building along, on any computer, and rebuild later from the recording.

What's in the Part 1 recording

Part 1: Foundations, as it ran live on Monday, September 28th, 2026. The video runs 3 hours and 2 minutes, with the breaks and waiting times cut. You see the instructor's screen and hear the questions asked during the session.

Chapters

• 0:00:00 Welcome and the plan of the session.
• 0:02:47 Setup: the shell, Git and uv, the manual installation, ComfyUI Desktop as the simpler route, and the model downloads.
• 0:13:34 How models work: from the Turing test to today's generators; neurons, layers and weights; how a model learns by predicting noise and generates by removing it; latent space and the VAE; steps, distillation and quantization; CFG, negative prompts and shift; video models; what models get right and wrong; training from scratch, fine-tuning and LoRA.
• 1:35:29 The landscape: closed platforms and open weights, where models come from, why ComfyUI, partner (API) nodes, GPU memory and cloud GPUs.
• 1:45:07 A tour of ComfyUI: where the model files go, the Nodes 2.0 interface, the console, the node library, custom node packs, templates, and the workflow saved inside each generated image.
• 2:00:29 A first image with Z-Image Turbo: seed, batch, shift, resolution, steps, CFG, samplers and denoise, each tested live, then how to write a prompt.
• 2:34:45 Control: Canny and Depth Anything 3 passes that drive the Z-Image Turbo Fun Union ControlNet.
• 2:45:30 Guided tour, without a live run: LTX 2.5 image to video, SAM 3.1 segmentation, Qwen Image Edit and SeedVR2 upscaling.
• 2:50:11 Ethics and law: licenses for code and for weights, copyright and human input, disclosure rules, training data and lawsuits, and production risks such as shadow AI and where your data is stored.

Course materials

• The session deck: 68 slides, with every source linked and the download link of every model file. It covers ethics and law in more depth than the video.
• The hands-on part uses the official ComfyUI templates, so there is no separate workflow file to download.

Good to know

• Six short on-screen notes correct or complete points made live.
• English subtitles are included.
• Recorded with ComfyUI as of September 28th, 2026. Later versions may look slightly different.
• Watching needs no GPU. To rebuild the hands-on part, you need an NVIDIA GPU with at least 16 GB of VRAM and about 150 GB of free disk space for the models.

Buy the Part 1 recording for 199 USD.

How a session works

• Live and hands-on. Four hours over Google Meet, in English. You build along with the instructor, with short breaks and time for questions.
• Small group. Twelve seats at most, so the instructor can help you during the build.
• Recorded. You keep the recording and the course materials for at least 12 months, to rebuild at your own pace.
• Standard package. Adds a 45-minute 1-on-1 mentoring session after the class, in English or French, on your own workflow or project.
• Ethics built in. Open models first, licenses explained, responsible use in production. Read our statement of AI ethics.

About your instructor

Hello, Bonjour, Kia ora,

I'm a French New Zealander living in Canada. I'm the former Head of Effects at Wētā FX and Houdini teacher at Fx Machina. I hold six AWS certifications, including Machine Learning Engineer and Machine Learning Specialty. I have worked on over 30 projects for the TV and Film industry since the early 2000s, including the Avatar series, Game of Thrones, and Terminator (see my IMDB profile). I've empowered many artists going from XSI/Maya to Houdini. Today, I'm enabling the AI transition.

I love teaching and witnessing that "aha" moment in people's eyes: the excitement of learning and growing. I'm not here to waste anyone's time, and I have plenty to share. Join me; it will be fun, I promise!

Learn more about Adrien.

Discounts, referrals and Bootcamp credit

• Returning customer: each full-price purchase earns 50% off one future Masterclass session, mentoring session or recording, within 12 months. A discounted purchase earns no new discount. The Bootcamp is excluded.
• Referrals: once you have paid in full for any item, each person you refer who completes a paid purchase earns you one free item: a recording, a 45-minute mentoring session or a Masterclass session (Basic). One reward per person referred, valid for 12 months. The Bootcamp is never a reward.
• Bootcamp credit: one Masterclass, mentoring or recording purchase from the 12 months before you enroll in the Bootcamp is credited toward your seat, up to 499 USD.

Is this Masterclass right for you?

It is ideal for:

• VFX artists and creative professionals who want to bring AI into their work
• Filmmakers and content creators exploring new ways to produce
• Technical directors evaluating AI for their studio
• Anyone curious who wants to learn by doing

It is not the right fit if you want a watch-only lecture, or if you do not enjoy tools that change fast. You don't need to be convinced that AI belongs in the industry, but you do need an open mind.

FAQ - Content

FAQ - Logistics

FAQ - Pricing and enrollment