![October 7: AI video news roundup. Reka announces Rho-1, bringing video generation and editing into one model. Higgsfield updates film festival schedule, with audience voting planned for October 8. [Reference] Video DeltaNet speeds up H3 video generation. A character for the October 7 edition appears at the top right, and the 1 Minute Video Lab icon at the bottom right.](../../../assets/news/news-2026-10-07-mascot-07.webp)
Reka announces Rho-1, bringing video generation and editing into one model
On October 5, Reka announced the 19-billion-parameter Rho-1 as a research preview. One model handles understanding and generation of text, images, and video, as well as robot actions. The official demo shows a conversation that turns an image into video, changes the weather, and explains the changes.
The company reports that a distilled version generated 5.3 seconds of video in approximately one second. This is an internal measurement. Resolution is up to 672×384, and the company explicitly notes issues with spatial structure breaking down in longer generations and with the stability of local edits. This is not an announcement of a finished commercial product.
Sources, images, and examples
Higgsfield updates film festival schedule, with audience voting planned for October 8
In its October 5 official changelog, Higgsfield announced an extension to the Global Film Festival schedule. The change allows more time to review submitted works.
The revised schedule lists the shortlist and Audience Choice voting for October 8, jury review for October 9–14, and winners in the final week of October. This is the published schedule, not an announcement that the submission deadline has been extended again.
Sources, images, and examples
[Reference] Video DeltaNet speeds up H3 video generation
We checked the official model card for Video DeltaNet, which appeared in the original news for the October 7 edition. Contrary to information giving October 5 as its release date, the official history dates the release of weights and code to September 6 and the introduction of FreeVideo for consumer GPUs to October 2. This is reference information on an acceleration method, not a new release.
Built on MiniMax H3, it combines an efficient attention mechanism with a small adapter for additional training. In official measurements using eight B200 GPUs, it generated 14.4 seconds of video in approximately nine seconds, including all processing. This does not mean an ordinary PC can achieve the same speed.
Sources, images, and examples
Information as of October 7, 2026. Production ideas are the editorial team's analysis. Images and videos can be viewed in the linked original articles.
![October 12: AI video news roundup. [Reference] Runway Dev adds Grok Imagine Lite, with prices varying by resolution. Protoface API expands to more than 40 models, with multiple providers under one key. [Reference] Runway MCP uses Brand Kits assets as generation references. A character for the October 12 edition appears at the top right, and the 1 Minute Video Lab icon at the bottom right.](../../../assets/news/news-2026-10-12-mascot-12.webp)

