AI Film Daily · Sunday, October 25, 2026
AITVHOT
HBO Max launches a vertical-screen channel, completing short-drama deployment across all five major US streamers
HBO Max launched the "HBO Max Shorts" vertical short-film channel on July 28, completing vertical short-drama feature deployment across all five major US mainstream streamers: Netflix, Disney+, Peacock, Prime Video, and HBO Max.
Model releases/updates
Kling 4.0 Launches on Lovart, Supporting 30-Second Segments and 10 Keyframes
Kling 4.0 is now available on Lovart, generating clips up to 30 seconds in a single pass, with support for 10 keyframes to control pacing and up to 15 character/product/scene reference assets. The official claims include more stable motion, better audio, and larger camera movement ranges, with Lovart's Arti responsible for generating the shot list.
Tools and Workflows
MacH3 Metal: a ComfyUI node package for running MiniMax H3 natively on Apple chips
The author releases MacH3 Metal, a workflow package that wraps antirez's h3.c native C+Metal engine into ComfyUI nodes, enabling native MiniMax H3 operation on Apple chips.
Tips and Opinions
Author Runs a 4-Minute MV Locally on an RTX 4090: Chaining Models Including Qwen-Image 2.1 and MiniMax H3
The author locally produced a 4-minute unofficial fan MV for a The Flaming Lips song using a single RTX 4090 and open-weight models, without any cloud services or paid APIs throughout.
The author creates a 10–15 episode AI mini-series with LTX 2.5, publicly sharing the workflow and lessons learned
The author uses LTX 2.5's image-to-video feature to create a 10–15 episode AI mini-series, with the first two episodes already released. The process is to first generate the first and last frames with ChatGPT and fix character reference images, then use Lightricks' ltx-pipelines to generate video and audio at 1920x1088.
Generating five Japanese-style designs with MOSS-Image: from movie posters and magazine covers to a manga page
The author uses MOSS-Image Base in ComfyUI together with the qwen3.5_9b text encoder and flux2-vae, specifying text content and bbox positions via JSON-format prompts, to generate five kinds of Japanese designs: a movie poster, a women's magazine cover, a CD cover, a novel cover, and a manga page.