Yesterday, Dreamina 4.0 started its beta testing, and Nano Banana was released over a week ago. I plan to use both of these in my new films. These two mean that character consistency has been completely solved. I used to be an AI creator who was very competitive about character, scene, and style consistency. Back in the SD1.5 era, I was already training LoRA to maintain character consistency.
When multi-reference was introduced previously, I also wrote an article stating that there was no need to compete on character consistency anymore. After using multi-reference, significant gaps remained in image quality and character texture. But Dreamina 4.0 and Nano Banana should take texture quality to the next level. Using prompts to modify images has become mainstream, and multi-angle storyboarding is also easily achievable.
Does this mean I’ll abandon Flux’s LoRA? I’ll still train LoRA for main characters because LoRA training is very simple – it can be done with just one image. When fine-tuning characters, you never know when LoRA might come in handy again. So my superhero assembly will continue. Additionally, Flux’s LoRA has remarkable effects when performing face swapping in Krea, resulting in good texture quality.
What changes will there be in overall film production after character consistency is completely solved? First, what I really need is increased productivity. I’m a very dedicated person who spends an average of 6 hours daily on AI, but I currently need a month to make a short film. Due to nitpicking about character consistency, two-person positioning, and other issues, Photoshop work is still required.
Currently, if you ask about any good film you’ve seen, they all involve extensive Post-Production work, often using Photoshop.
But when multi-angle generation and character consistency are solved, more time can be spent on creation rather than just on Photoshop. Of course, Photoshop remains a fundamental skill that no one can overlook. If you think one-click film creation can solve all problems, it means your work isn’t professional enough yet.
If I can update once a week, then I’ll start making long-form series. I hope this update cycle can handle 7-10 minutes of content. If this productivity improvement can achieve this level, I’ll feel excited rather than feeling that my previous investment in character detail work was wasted.
When you join AI creation, you’ll also become at peace with the rapid technological advances. The technology used today may have a shelf life of only one month. The plateau period won’t arrive for at least another year or two years. The emergence of new technology always brings amazement, and this sense of wonder is the same feeling I had when I first encountered AI video.
No longer competing on consistency, so how do we compete on stories? Some people argue that many individuals now utilize AI to write stories, and major studios have a wealth of high-quality scripts, so stories can’t be completed. I have a large collection of long and short stories from the Computational Universe worldview. My long-form creation schedule extends to 2032, and as for making them into AI films, there’s a queue for those, too.
I’m not sure if I can compete with people who use AI to write stories, but I have many stories to tell. My feature films have been in a wait-and-see state because my solo productivity is too low, and even with a team, it remains too low. This easily results in spending 3-6 months to complete a film, only to find it unwatchable due to outdated technology.
This would be a fatal blow to high-investment feature films if productivity is insufficient. For long-form series, if weekly updates aren’t achieved, it’s a big problem. Non-serialized works face the aforementioned issue of technology becoming outdated by the time they’re completed.
I have endless stories to tell, and I don’t know when I’ll be able to finish them all. The short films I’m currently making have 25 creative ideas waiting in the queue. My long-form stories have involved numerous massive projects, and being able to make feature films has been my dream since I started creating AI videos. My research direction has always been a realistic, long-form narrative.
I’ve been waiting for technology to mature, but it’s currently in a rapid growth phase. Around this time last year, I thought I needed to wait another year before starting the AI visualization of the “Computational Universe” novel series. Now it seems “Computational Universe” still needs another year, but stories like “Computational Journey” can be made now.
Why don’t I use AI to write stories yet? Maybe because this is my last strength in the face of AI. I wrote my first blog post in 2005 and have been writing various stories for nearly 20 years since then. People who can get into my stories find them irresistible, while those who can’t often encounter certain reading barriers. Such stories with barriers are best visualized to make them more understandable.
I really love AI – it’s modern magic.
Follow me and explore AI filmmaking together.
