Thanks and Suggestions
Thank you for the model. It turned out great.
I have a few suggestions that I believe many users would agree with. Overall, these are fairly obvious points and apply more to the AI industry as a whole than to this specific model, but I'd still like to mention them.
- Image-to-Image
The model is already quite multimodal. If it's possible to build on the existing work and fine-tune an image-to-image version, that would be fantastic.
At the moment, the only open-source options are the rather outdated Qwen 2511 and Klein 2, so I think such a release would make a huge impact. Even this video model already surpasses them in prompt adherence, and I've already seen workflows that use it for similar purposes.
Of course, it's currently limited in resolution and fine detail because it's optimized for video. But if there were a way to further improve it for image-to-image tasks, it would be extremely valuable.
- Speed
The model is noticeably slower than LTX. That said, the quality is also significantly higher, so I understand this isn't really a fair criticism. Training a high-quality model—or making it substantially faster without sacrificing quality—is obviously a very difficult problem.
- Long-term consistency
In many cases, 10–15 seconds simply isn't enough, and users often have to look for ways to extend generated videos.
Using the last frames of one clip as the starting point for the next often breaks temporal consistency, resulting in noticeable seams between clips.
It would be amazing to have the ability to generate long, coherent videos. Ideally, the model could generate them in 10–15 second segments, with each new segment conditioned not only on the final frame of the previous one, but also on the overall storyboard, motion, dialogue, audio, and the broader context. That way, users could launch a long generation job and end up with a consistent, extended video rather than a collection of disconnected clips. While also finding a way to prevent quality and consistency from degrading over time.
Thank you!