SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.
見出し画像

Discarding past video generation methods and knowledge to move to the next phase / Archive video (4th session) released: "Would you like to produce an AI drama?"

CreativeEdge CL+

The recorded video of the "4th Session: Would you like to produce an AI drama?" streamed yesterday, Friday the 20th, from 8:00 to 9:00 PM, has been released on YouTube.

Archive video (4th session)

  • Duration: 55 minutes 42 seconds

Video generation AI models have reached the 5th generation (such as Kling 3.0 and Seedance 2.0), and I am resetting my past know-how and knowledge of video generation to start exploring completely new ways of utilizing them from scratch.

This series (How to make an AI drama) is like a ritual to "discard" past video generation methods and knowledge.

The importance of fixed character management shown by Element 3.0 in Kling 3.0, and intermediate layers that control AI video production—rather than just UI wrappers like Cinema Studio 2.0 provided by Higgsfield AI—will become of great value in the future.

However, this 5th generation will only last until the next-generation model (world model) from Google Veo arrives.
It will likely move to the next phase in less than half a year.

While practicing video content production where multi-model orchestration is the norm, it is necessary to build a flexible workflow that is always conscious of "switching to new technologies" (without depending on a specific generative AI).
I would like this series (How to make an AI drama) to be the gateway to that.

The final session is scheduled for Wednesday, February 25th, from 8:00 to 9:00 PM.
*This was supposed to be the final session, but as the verification of Kling 3.0 progressed further, it could not be contained within one hour and was extended.
The registration page will be published on Peatix.

Live stream content (partial):

This time, the focus is on "vertical short videos"

Two projects: "Feature-length drama" and "Vertical short video"

Prop Elements also include size information

Prop registration (including references for understanding size)

Reference for destruction scenes

For parts scattering in destruction scenes, a "disassembly diagram" reference is effective

Screenshot reference for connecting to the next scene

For continuous scenes, take a screenshot from the generated video and use it as a reference

Write prompts while inserting references (@)

How to write prompts for Kling VIDEO 3.0 Omni (including references)

Techniques to trick AI (when you want to generate only footsteps from video)

When generating only "footsteps" from a video, you need to interfere with face recognition

Be conscious of reference strength

If the camera angle of the background and the person differs, generation accuracy decreases

Add Elements if the person reference is weak

If you don't add Elements to increase the strength of the person, the face will become someone else

Method to completely control movement: Start & End Frames-to-Video

Start & End Frames-to-Video

Effective use of 3DCG tools (using DAZ Studio here)

Create poses in DAZ Studio and use those screenshots as references

How to use Nano Banana Pro 2K and 4K

Nano Banana Pro 2K and 4K (4K consumes twice the credits)
Nano Banana Pro 4K is higher quality than 2K (4K is essential for infographics, etc.)

The era of remaking characters due to technical updates is over

We have entered a stable period where fixed characters can be maintained

Content that relies on the surprise or excitement of generative AI technology has a short lifespan

Short videos consisting only of world-building have a short lifespan

Nano Banana Pro's remix technique should be mastered as a top priority

Use images generated by Midjourney as source material and finish them with Nano Banana Pro
You should master Nano Banana Pro's remixing techniques

Starting from a sketch

Start with "pictures" that offer a wide range of expression
Sketch to Image technique

Battle scenes are more difficult than you might imagine (the same applies to Seedance 2.0)

You need to be prepared to consume a large amount of credits

Combining references and elements is the foundation

If there is a contradiction between the reference and the element, the generation accuracy will decrease

Ethical guardrails and moderation

Images that are easily misjudged by AI (it won't let you generate them)

Creating character boards

Using Firefly Boards to create character boards

Planning work for vertical short videos

Using 4-panel comics as "material" to create storyboards

The final session is on Wednesday the 25th at 8 PM

This time it really is the final session!


Previous archive videos:

3rd session

2nd session

1st session


The final session is scheduled for Friday, February 20th, from 8:00 PM to 9:00 PM.
The registration page will be published on Peatix.


Updated: Saturday, February 21, 2026 / Published: Saturday, February 21, 2026

いいなと思ったら応援しよう！