An Instagram roundup describes a workflow pairing MiniMax's new H3-Image for stills with Lightricks' LTX-2.5 for motion and sound, both on a single RTX 5090. The stills stage is one ALFW runs itself.
In ai-muninn's documented local test, a Blender blockout supplied camera movement and staging while separate reference images supplied character appearance.
In a November 9, 2026 demonstration, educator Jerrod Lew uses character and style references to generate a 3×3 storyboard in MiniMax H3 Image, then supplies that grid to H3 Video to create a multi-shot clip.
A creator-demonstrated workflow uses Magnific's Assistant to turn a finished graphic into a motion prompt, then sends both the prompt and original artwork to MiniMax H3.
A creator's object-transformation tutorial offers a useful way to control AI video: film a real opening frame, design the destination as a still image, and let MiniMax H3 generate the transition.
The standard way to stretch an AI video shot is to start each new clip from the last frame of the one before it. An independent test with MiniMax H3 measured what each of those links costs, frame by frame.
The usual local route to MiniMax H3 runs through ComfyUI, custom nodes and a terminal. Wan2GP offers a browser interface for the same model, and Pinokio will install it for you.
A creator showed how to turn ordinary footage into hand-drawn-illustration video with MiniMax H3 by uploading the clip as a reference and prompting only the style change.
Vidu (Shengshu Technology) launched Q4 Preview on October 7, 2026, with up to 15 image references, up to 3 audio references, 2K/4K 10-bit output, and launch pricing from $0.014 per second.
Hailuo's MiniMax 3 (the H3 line) is the most capable video model the company has shipped: up to 15 seconds, 2K resolution, native stereo sound, and a reference system that takes text, images, video and audio as one…
Hailuo's MiniMax 3 is the most capable video model the company has shipped: up to 15 seconds, 2K resolution, native stereo sound, and an omni-reference system that accepts text, images, video and audio as one combined…
The clean route is ComfyUI 0.30.0 or later, its built-in MiniMax H3 template, and the smallest official four-file set: a 21GB pruned INT8 diffusion model, a 15.7GB Qwen3-VL text encoder, a 5.21GB video VAE, and a 605MB…
Forget one prompt that spits out a finished game. The workflow that actually works splits the job into three roles: a designer that turns your idea into rules, a programmer that builds and tests it, and an artist that…
Hailuo MiniMax 3 takes reference images, movement references, audio, and written direction in a single generation. Throwing everything at it at once wastes credits.