ComfyUI Video Training: LTX 2.5 and MiniMax H3

Turning a still image into a moving, directed shot is now something you can do on your own computer. Our ComfyUI video training teaches adults, company teams and enterprise creative departments image to video using the two strongest open models of 2026, LTX 2.5 from Lightricks and MiniMax H3, the model behind Hailuo 3. Taught by a creative director with over 20 years of advertising experience in motion and 3D, the course focuses on what working creatives actually need: directing performance, protecting a product, and getting a usable result without endless retries.

Below are the exercises we use in class, with the real clips our attendees learn to recreate. Every clip was made in ComfyUI on local hardware, which is the same setup you will have on your own laptop after the course.

Lesson Exercise: Same Image, Two Models

Attendees start with one photograph of a model holding a skincare tube and give both models the same direction: look at the product, smile, then look back at the camera, with a locked-off camera. Comparing the two results side by side is the fastest way to understand what each model is good at.

LTX 2.5 stays very close to the original frame. The face stays consistent, the camera does not move, and the tube stays upright with its label readable, which is exactly what a product shot needs. The trade-off is that the first beat of the direction, looking down at the product, is barely acted.

MiniMax H3 performs every beat in order: the glance down, the second hand coming up, the smile, the look back to camera. It feels like directing a real model on set. The label turns away as she handles the tube, so for packshots LTX 2.5 is the safer choice, and for performance MiniMax H3 wins.

Mind Theory is Singapore’s pioneering AI and creative-tech education provider, established March 2023, with hundreds of students across primary, secondary, and adult programmes. Read about Us.

Writing Prompts Like an Animator

The single most useful skill we teach for image to video is writing direction as timed beats. Instead of describing a mood, attendees write what happens from 0 to 2 seconds, what holds at the peak, and how the shot settles. These exercises show how far that technique goes.

A smile that builds

A short prompt: the expression builds from a gentle smile to a full laugh, while the hair, choker and pendant stay consistent with the original character.

A sneeze in four timed stages

The sneeze prompt was written like a timing sheet, and the model hit every stage, including the hair following through after the head snaps forward. This is squash and stretch, the same principle traditional animators use, applied through a prompt.

1990s cel anime comedy

Held poses, a lens flash on the glasses, a sweat drop and a slow realisation. Attendees learn how to keep a hand-drawn style from drifting into 3D, and how comedy timing is described in words.

A one-line prompt

Sometimes short is enough. A single sentence gives a full turn with flowing hair, and the character is still recognisable when her face comes back round. Attendees learn when a simple prompt works and when a scene needs full direction.

Want to direct shots like these on your own machine? Email us or WhatsApp us about the next ComfyUI class.

What the Course Covers

Installing ComfyUI and the video models on your own laptop, loading official image to video workflows, choosing resolution and duration, running quick low-resolution drafts before a full render, and knowing which model to use for which brief. LTX 2.5 outputs up to 4K and can create several connected shots in one generation, while MiniMax H3 leads on acting and follows timed direction closely.

Because everything runs locally, your source images and client work never have to be uploaded to a cloud service. For the full lesson plan, see our ComfyUI course. Companies can also book ComfyUI corporate training for their in-house teams.

More from this ComfyUI course series: AI image generation and image editing.

Established in March 2023, Mind Theory is Singapore’s pioneering AI education provider, offering Gen AI courses and corporate training for adults, working professionals and marketers.

Attendees build web apps through vibe coding and create production-ready AI visuals using prompt frameworks and AI-powered tools. To find out more, visit our Courses page.

Contact us to book a class by email or Whatsapp..