[For AI Beginners] No GPU Required! I Made a Stylish 30-Second Music Video Using Only Free AI Tools | ChatGPT | Google Flow | Gemini | Shotcut
Recently, the evolution of image, video, and music generation AI has been remarkable, and many people are likely thinking, "I want to try creating works with AI even on a computer without a GPU."
So, this time, I created a 30-second music video using only computers without a GPU and free online AI services.
I combined services that are all free to use, covering image, video, music, and editing.
Furthermore, I created the same content using two different video generation AIs and compared the differences. The story is set on a summer island. It is a
love story about a man and a woman who end up taking shelter from a sudden squall and find their hearts connecting through a chance encounter.
Note that the input images, prompts, and music used were all identical, and only the differences caused by the video generation AI were compared.
Free Local AI vs. Free Online AI
Free Local AI (LTX-2.3)
Free Online AI (Omni Flash)
Continue the drama here
Free AI services used this time
The three services used this time are as follows.
Image Generation: ChatGPT (Image 2)
Video Generation: Google Flow (OmniFlash)
Music Generation: Google Gemini (Lyria 3)
Also, for comparison, I generated a video using LTX-2.3 with the same prompt and the same image.
How to use free online AI is here
Image Generation: ChatGPT (Image 2)
Video Generation: Google Flow (OmniFlash)
Music Generation: Google Gemini (Lyria 3)
Video Editing: Shotcut (Free Video Editing Software)
Click here for how to use the free local AI LTX-2.3
Video Generation: ComfyUI (LTX-2.3)
Good points I felt after using OmniFlash
When I actually used it, OmniFlash generated videos with a higher level of perfection than I imagined for me.
What was particularly impressive was:
The movements of the people are natural
The facial expressions are rich
The camera work is dynamic
It automatically adds effects like a scene from a movie
these are the points.
Compared to LTX-2.3 using the same prompt, I felt that OmniFlash had a stronger sense of impact and realism in the footage.
There were also a few points that concerned me
Of course, it wasn't all good points.
I will introduce what I noticed while actually using it.
1. There are cases where reference images cannot be used
When uploading an image as a reference image,
"There is a possibility that it resembles a famous person"
for that reason, video generation was rejected at times.
Therefore, there are cases where even images you have carefully prepared cannot be used.
2. Watermarks are included
In the free version, a star-shaped watermark is displayed in the bottom right corner of the generated video.
At this time, it is not possible to remove this watermark.
3. Lip-sync is convenient, but you cannot use your own voice
Although I didn't use it this time, OmniFlash has a lip-sync feature.
If you enter dialogue, it will speak naturally, even in Japanese.
However, you cannot upload your own recorded audio; it is designed to select and use TTS (text-to-speech) voices provided by Google.
If you select the same voice, it will speak with the same voice every time, making it easy to use for series projects.
4. Character faces are prone to changing
This was the point that concerned me the most this time.
With LTX-2.3, the face of the person specified at the beginning was relatively maintained.
On the other hand, with OmniFlash, there were some cuts where the face looked like a different person halfway through the video.
While the character movements and direction are attractive, there is still room for improvement in terms of character consistency.
Music production with Lyria 3
For the music, I used Lyria 3, which is available through Google Gemini.
This time, I wrote the lyrics and composed the music based on the story:
'A sudden squall hits a summer island, and two people meet while taking shelter from the rain'
I composed the lyrics and music with this story in mind.
The generated lyrics were in English, but the song had an atmosphere that matched the content specified in the prompt. The English title is
'Stay Through The Grey' and the Japanese title is
'Even When You're Melancholy, Stay By My Side'.
It is quite surprising that you can create such natural music for free.
However, at the moment, the music that can be generated with the free version is limited to up to 30 seconds.
Even with 30-second music videos per track, by stitching multiple scenes together, you might be able to create a long-form music video that feels like a full drama.
For example, starting with a scene where two people meet, enjoying beach volleyball together, driving along the coast in a convertible, and talking at a bar at night... We are entering an era where you can easily create music videos with such narrative quality by utilizing AI.
My thoughts after comparing OmniFlash and LTX-2.3
From here on, these are just my personal impressions.
OmniFlash feels like it is a step ahead of LTX-2.3 in terms of:
natural movement
richness of expression
camera work
visual direction
.
However, because of that, there is a strong tendency for the AI to add its own direction automatically, which can lead to:
camera work that wasn't requested
suddenly starting to speak in English
unexpected visual effects being added
.
If that direction fits the work, it results in a very good outcome, but sometimes the video ends up different from what you imagined.
For better or worse, I felt that this AI has a 'gacha' element where you look forward to seeing what kind of direction it will produce.
Furthermore, regarding the point that human faces tend to change midway through, I look forward to future updates.
Summary
As a result of this experiment, I was able to produce a 30-second music video using only free online AI services, even on a computer without a GPU.
Omni Flash, in particular, offers high quality in terms of visual expression, camera work, and character movement, resulting in a finish that is hard to believe is free.
On the other hand,
there are cases where reference images cannot be used
watermarks are included
facial consistency is somewhat weak
are some points to keep in mind.
Even so, considering that you can produce such high-quality work for free, I felt that this is a service I can highly recommend to those who want to start AI video production.
If you have given up because you don't have a GPU, why not give it a try?