beginning

In recent years, the state of video production has changed drastically due to the evolution of AI technology. In particular, in K-pop and other MV productions, content utilizing AI is increasing, and higher quality and unique visuals are being pursued. In this article, I will introduce the process of video production using generative AI and its possibilities.
※2024/12 to 2025/01

‍
Use of AI in MV production

1. Reviewing the production flow

The general flow for normal MV production is as follows.
　1. Music Creation & Analysis
　　Analyze lyrics, melody, rhythm, and tempo.
　2. Story settings
　　The concept, story, and character settings for the MV have been decided.
　3. Worldview settings
　　Gather references for buildings, fashion, colors, etc., and create a sense of unity as a team.
　4. Video production
　　Cut, create a storyboard/V frame, and edit after generating the video.

There is almost no change in the AI video production flow this time, but I set it up and executed as follows.
　1. Generate lyrics and music
　　After generating the lyrics, the song is generated with SUNO AI.
　2. Story+Worldview Setting
　　The concept, story, and character settings for the MV have been decided.
　　Collect references on buildings, fashion, colors, etc. on Pinterest etc.
　　If necessary, images are generated with Midjourney.
　3. Video production
　　No cuts or content were created, and editing was performed after the video was generated.

2. Music generation

Using the AI tool “SUNO,” K-pop style songs are created based on lyrics generated from ChatGPT. Multiple variations were compared, and the most suitable song was selected.
For improving sound quality and separating stems”Ultimate Vocal RemoverUtilize”.
The name of the song is “Echo7Light” *named by ChatGPT

Generative music ①
https://suno.com/song/5b8523c7-29e9-4eeb-be98-74c17b15cee2
Generative music ②
https://suno.com/song/65480018-60cf-4c0b-a0cc-981a4060c5b8
Generative music ③
https://suno.com/song/5b8523c7-29e9-4eeb-be98-74c17b15cee2
‍Generative music ④
https://suno.com/song/8db05ac0-fda1-4e1f-a372-01d0eae366bb
Generative music ⑤
https://suno.com/song/7e1eb66d-28c1-436e-a6f5-18fa05213902
Final nominees ①
https://suno.com/song/e51152ff-b13a-42f4-bb68-e491ba63cc44
Final candidate ② ※Adopt this
https://suno.com/song/3f0ad17c-0a9f-4c4c-93e4-1f47ed17062b

‍

Comparison of AI video generation

Comparative video generation AI

• Runway(A video creation and editing tool made in the USA)
• Vidu AI(developed by Tsinghua University, China)
• Hailuo AI(Manufactured by MiniMax, China)
• KLING(Made in Kuaishou, China)

comparison results

• KLINGIt produces the most MV-like video.
• RunwayThe bankruptcy in the background is conspicuous.
• Vidu AIThe camera work is weak.
• Hailuo AIIt is unstable and difficult to control.

How to compare

① Create the original image in Midjourney

② Enter the same prompts and images for each video service.
　*Use the default settings for each tool

Freestyle Dynamic Camera Movement: The Camera Moves Fluidly Around a Silhouetted Woman Standing in Front of Sheer Curtains, Dramatic Lens Flares as the Sunlight Streams Through The setting evokes a retro cinematic aesthetic with muted tones, soft light, and a nostalgic atmosphere. The woman's delicate lace dress and her motionless, contemplative pose add an air of mystery. Cinematic retro footage

③ Compare the output video
　※With the video in between
　*The accuracy of AI tools is improved by increasing the number of trials, but it is also necessary to consider the fact that they are costly.

MV world view/story setting

In AI video production,Surreal (surreal) worldviewis popular. (watch closely)

• Go along with the musicDark and cool vibeIn order to output, a reference image was generated with Midjourney.
• KlingAIGenerate videos and adjust lip sync.
• Trial and error for each sceneProduce videos while doing it.

Issues and countermeasures in MV production

issues

• Highly dependent on AI tools → Work has stalled due to server downtime, etc.
• Specifying camera work is difficult → Unlike still images, dynamic effects are unstable.
• Cost increase due to number of trials → Generation credit is required.
• The detail expression of small details is weak → It is difficult to reproduce costumes and branded products.

solutions

• Combine multiple AI tools → Distribute dependency risk.
• Specify camera work in detail → Review storyboards and reference images as appropriate.
• Accumulate and share trial data → Build an efficient production workflow.

Generative AI service cost for MV production

tool

Cost (monthly)

Midjourney

5000 yen

Suno

1500 yen

KlingAI

4500 yen

totals

11,000 yen

summary

Video production using AI has the potential to produce high-quality images at a lower cost compared to conventional MV production. However, trial and error is required after understanding the risk of being too dependent on specific tools and the limitations of AI.

Kadinch will continue to explore new possibilities in various fields, including video production and technology development, by utilizing AI technology that is further evolving in the future.

‍