Yesterday, a group member asked: “Group admin, you used to be a writer. How did you become so skilled at video creation? Could you briefly share your learning process and experience?” So I’m writing this blog post to discuss my learning journey and insights. When it comes to technology, I’ve always held a relatively radical view – I don’t believe that focusing on the technical aspects of AI is useless. AI is a combination of technology, art, and literature. Only when these three elements are balanced can great works emerge.
The learning process is well-known to everyone: for image generation, start with Midjourney, then move to Flux, and then select different video generation platforms based on their respective strengths. Midjourney is very simple to use – you can master it in an hour – but its aesthetic quality and style consistency are excellent. While Jimeng and Doubao can also achieve stylization, there’s still a significant gap compared to Midjourney.
After that, I learned how to use the Flux series and LoRA training. This is also very simple. Now you can directly run the entire Flux suite on Liblib and RunningHub, including the newly released Flux Kontext and Flux Krea. The Flux series hasn’t actually been eliminated by GPT-4o as some anti-tech people claim. Moreover, the remote use of ComfyUI eliminates a significant amount of configuration hassle, making it very simple. Anyone who still says it’s difficult simply hasn’t truly participated in the AI film and video production process.
LoRA training is currently also very simple – you can generate multiple images from a single image directly, and remote training is available, making it convenient and quick without tying up your computer. Therefore, the design of the base image in LoRA training is the most crucial aspect. You need to spend time designing the character’s appearance and clothing, then take it for training. Those who think LoRA is difficult also haven’t truly participated in the AI film and video production process.
Don’t fear technology, because there’s nothing difficult about AI technology currently. Except for some special fight choreography, there’s no technology that exceeds the average person’s comprehension level. Next comes the selection of video generation platforms, which evolve with the times. As a result, every time a platform releases a new model, you need to familiarize yourself with its characteristics.
As of today, August 6, 2025, video models have characteristics such as: Veo3 excels at text-to-video and smooth character movements; Hailuo 02 is great for fight scenes and all kinds of daily work; Kling 2.1 Master version excels at preserving character details; Vidu’s multi-reference feature is magic; Higgsfield is a bit expensive but has pre-set special effects. Video models need to be used in combination to achieve the best results.
Previously, a friend asked me how she could get started with AI video production without using a VPN. After thinking about it, I recommended the complete Jimeng suite to her. So if you’re an absolute beginner and don’t want to use a VPN, you can try the Jimeng series. However, if you require the American blockbuster quality, a VPN is still necessary.
Crossing over from author to AI video requires not only an AI technical foundation but also a solid understanding of the basics of audiovisual language. Audiovisual language can be learned through books or by watching films. I think directing techniques are technical skills, but to develop a feel for it, you must practice. Your initial videos might be completely inadequate in terms of audiovisual language. However, with more practice and study, you’ll naturally improve and become better.
On the contrary, what’s truly difficult might be dedicating yourself to creation. This requires a certain foundation and determination to focus single-mindedly. It wasn’t until after May this year that I pushed aside other matters to focus entirely on creation. I finally realized that I couldn’t stop creating the Computing Universe series for even a moment. As I age, my time becomes increasingly limited. If other things continue to encroach on my creative time, it would be a disaster.
If you want to join the ranks of AI video creators, you don’t need to take a year like I did to achieve complete proficiency. Current technology can indeed accomplish more and more – you can get started in just two weeks. Then it’s time for your creativity to shine. Bilibili University can solve everyone’s beginner problems. But to become proficient, you must create works. The more works you create, the better your feel for it becomes.
The reason I make AI videos is that they’ve shown me a completely new world. I would like to visualize my novels, enabling more people to directly experience the grandeur and beauty of the Computing Universe. I won’t forget why I’m doing this, because I want to continue in this industry for the long term. What’s your reason for making AI videos?
Follow me as we explore AI filmmaking together.
