Stable diffusion thread

[doogyhatts]

so hard. you made YT or TK presence liao?

YT presence.

But YT changed their algo liao.
Now, I have to figure out how to make 20 min video.

 

[AikiBoy]

YT presence.

But YT changed their algo liao.
Now, I have to figure out how to make 20 min video.

what you mean YT change algo? what your channel?

I got a simple gaming one but quite sux lol

 

[doogyhatts]

what you mean YT change algo?

Google did a core update to YouTube, such that a lot of channels doing evergreen content got shafted one side.
So these channels now get very little impressions, hence little views compared to what they got before.
A lot of these channels are likely to be gaming content related.

 

[AikiBoy]

Google did a core update to YouTube, such that a lot of channels doing evergreen content got shafted one side.
So these channels now get very little impressions, hence little views compared to what they got before.
A lot of these channels are likely to be gaming content related.

Then what niche is good.

I think many ppl now go for shorts.

I made an AI music video...short got more view than normal video

 

[AikiBoy]

I like kling ai face recognition feature... but consistency still need improvement.

Created this girl from comfyUI

 

[doogyhatts]

Then what niche is good.

I think many ppl now go for shorts

I tried Shorts before, but it tanked all my long-form views, so I stopped doing it.
Now, I have to rebuild the channel using long-form 20min video.

There are some trend changes to AI content.
AI girl lookbooks for long-form content are now mostly dominated by India.
Although, those from Taiwan are also doing quite ok.

Long-form storytelling content is mostly dominated by Japanese creators. (search for Chrono Trigger AI)
Yesterday, I discovered a Japanese channel doing SCP content. (search for @CreepyTheater_K)

For Shorts, search for @StunningBeauties67 channel, which is doing sports girls.

For music video content, you have to target the 1 hour mark.
This channel shows you how to do it.
https://www.youtube.com/@aifascinated

 

[doogyhatts]

@AikiBoy
There is another part on Japanese content, that is more closely related to their anime trailers.
They tend to include music with singing vocals, with characters speaking, and action-based animations ongoing.

 

[project_00]

How are u guys dealing with the missing models from CivitAI?

 

[doogyhatts]

How are u guys dealing with the missing models from CivitAI?

There are two archives. Diffusion Arc and Civitai Archive.
But I had downloaded some of those that I would need eventually.

 

[AikiBoy]

@AikiBoy
There is another part on Japanese content, that is more closely related to their anime trailers.
They tend to include music with singing vocals, with characters speaking, and action-based animations ongoing.

i wonder what AI they use. i also want to do story. but i sux

 

[doogyhatts]

i wonder what AI they use. i also want to do story. but i sux

It is stated right there at the end of the video.
Vidu, Midjourney and Suno.

 

[AikiBoy]

It is stated right there at the end of the video.
Vidu, Midjourney and Suno.

need skills to combine them.

 

[doogyhatts]

need skills to combine them.

It will take some time to work on them as the workflows are getting more complex.
And premium solutions are also getting more expensive.

FYI, the best lip-sync solution is on Dreamina, ByteDance's platform, extremely expensive.
It is based on their OmniHuman-1 model.
At the moment, there are no open-source solutions that can compete with it.

 

[AikiBoy]

It will take some time to work on them as the workflows are getting more complex.
And premium solutions are also getting more expensive.

FYI, the best lip-sync solution is on Dreamina, ByteDance's platform, extremely expensive.
It is based on their OmniHuman-1 model.
At the moment, there are no open-source solutions that can compete with it.

so AI channel/music can get monetise?
i working on Music... see those 1hr relax music..
but i just dont want anyhow upload random music.

i want something realistic. if you saw my link earlier. i got full length song. lol but can only make 1 minute MV from Kling. prompt and fine tune kills me.

 

[doogyhatts]

so AI channel/music can get monetise?
i working on Music... see those 1hr relax music..
but i just dont want anyhow upload random music.

i want something realistic. if you saw my link earlier. i got full length song. lol but can only make 1 minute MV from Kling. prompt and fine tune kills me.

AI content for music & animation can be monetised.
I also have such content but for monsters and horror music.

For 1 hr music video, you need to do your video generations on a local machine using a suitable GPU with very good performance and sufficient vram.
Aim for 10 minutes or less for each run.
You won't immediately finish all of them in a day, but you need to keep the generated clips and not delete them.
Eventually, you will have a huge stash of video clips, just randomly reuse and recombine them into a new 1 hr video, together with new animations.

 

[AikiBoy]

AI content for music & animation can be monetised.
I also have such content but for monsters and horror music.

For 1 hr music video, you need to do your video generations on a local machine using a suitable GPU with very good performance and sufficient vram.
Aim for 10 minutes or less for each run.
You won't immediately finish all of them in a day, but you need to keep the generated clips and not delete them.
Eventually, you will have a huge stash of video clips, just randomly reuse and recombine them into a new 1 hr video, together with new animations.

i want consistancy on the model.
Kling face reference when go 100%, face will deform. only works well in close up.
Credit used up... must wait next month

 

[doogyhatts]

i want consistancy on the model.
Kling face reference when go 100%, face will deform. only works well in close up.
Credit used up... must wait next month

You can fix the face consistency using a Flux+PuLID workflow.


Kling is also getting very expensive.
Try to shift everything you can to Comfy.

 

[AikiBoy]

You can fix the face consistency using a Flux+PuLID workflow.


Kling is also getting very expensive.
Try to shift everything you can to Comfy.

my pc not that good. lol plus comfy so many things to learn.

 

[doogyhatts]

my pc not that good. lol plus comfy so many things to learn.

I can run it using a GPU with only 8gb vram.
Slowly learn Comfy, not possible to learn everything at one go.

 

[doogyhatts]

@AikiBoy
Another channel to look at is @NeonNightsJapan.
This one doesn't need to do 1 hr music video.
And back-view shots are popular.