[3.2][3.3][RC] AI Labs with Midjourney, FaceSwap, PixVerse, Pika, Claude, ChatGPT, Gemini, DALL-E, Stable Diffusion

[nou nou]

I'm using a free account. So far, the interest in video generations on my board has been very mild. I'm not sure if I'm going to upgrade to a paid account at this point. For me, it's not an issue of length, but rather the quality of the generation. I feel like Pika is quite hit or miss at this point, mostly a miss actually. I'm sure it will improve over time.

OK thanks - I may sign up for a paid one for a month or so and see what happens then, I'll report back.

Yeah a lot of miss, agreed.

[privet.fun]

Stress test today - I launched a prompt to @Midjourney @Stable Diffusion @DALL-E @Pika at the same time.

I got this in my own post so all good there:

Screenshot 2024-05-12 150232.png

DALL-E and Midjourney came back with an image, Stable Diffusion and Pika never arrived. I did the same prompt again with Stable Diffusion and Pika and they came back fine.

Possibly/probably a concurrent connections issue?

There were no error messages anywhere though - not sure if something could be done about that?

Midjourney and Pika use callbacks/webhooks to receive results and updates from the useapi.net API. They do not keep the server connection open once the initial POST job submission is completed. All other bots will execute the POST and receive results right away. However, POSTs to most AI APIs (especially OpenAPI) can take up to 3 minutes. Gemini is the fastest among them, and Claude is somewhere in the middle. Therefore, for these, the web server connections will be kept open.
I'd imagine having extra 5-6 web server connections open should not make any difference but I guess it depends on you particular hosting/web server config.

[privet.fun]

OK thanks - I may sign up for a paid one for a month or so and see what happens then, I'll report back.
Yeah a lot of miss, agreed.

I'm very curious, please post a link to your Pika generations thread (if you don't mind )?
Your board probably has a much better audience for such bots.

[privet.fun]

Honestly adding Pika was a no-brainer since it's free and included in the useapi.net API monthly subscription.
They also offer a face swap feature which seems to be included in the API subscription too.
I've played around with that a bit but never actually added it. Maybe I will.

[nou nou]

Maaaaaan

Does my subscription on Pika.art work for Pika beta on Discord?
No, They are two separate systems
Your Pika.art subscription is only for the website for using pika 1.0
This subscription does not affect your access to Pika beta on discord.
The credits from Pika 1.0 are for use on the Pika.art website only.

Boo. Ah well I'll play on their website a bit then. Sheesh.

[nou nou]

I've played around with that a bit but never actually added it. Maybe I will.

Go for it yes please

[nou nou]

Quick (perhaps) feature request.

Is it possible, in addition to the {request} variable in the response template, to have a {short_request} (or something) that is a quote with a maximum length of X characters?

It's handy to have some context especially when different bots are replying at the same time, but full quotes are often too long IMO.

See https://github.com/privet-fun/phpbb_ail ... mini-setup:
max_quote_length, if provided, the quoted response text will be truncated to the number of words defined by the max_quote_length value. Set it to 0 to remove all quoted text entirely.

Is that what you want?

This doesn't seem to work.

So I have this in the ChatGPT setup:

Code: Select all

"max_quote_length": 10

And here is the result of a prompt:

Screenshot 2024-05-14 202037.png

[nou nou]

I'm having a weird issue - all of the DiscordCDN links are dead up until yesterday. Did something change?

I'm getting this error:

"discord_cdn configuration missing tokens of channels array values."

But I didn't change the configuration and the values seem correct.

[privet.fun]

This doesn't seem to work.

You're right, I need to update code a bit to trim original reply quote, right now it is only applied to follow-up quoted text.
You can try to quote reply to see what I mean.

[privet.fun]

I'm having a weird issue - all of the DiscordCDN links are dead up until yesterday. Did something change?

I'm getting this error:

"discord_cdn configuration missing tokens of channels array values."

But I didn't change the configuration and the values seem correct.

You most likely missing configuration for DiscordCDN https://github.com/privet-fun/phpbb_ail ... roxy-setup

[nou nou]

No, the configuration is there and it worked before - let me PM you the setup.

[privet.fun]

No, the configuration is there and it worked before - let me PM you the setup.

Thank you for reporting, I can see that issue was resolved.
I'll add JSON config validation step to avoid this in the future, should be an easy fix.

[nou nou]

User error nonetheless, thank you for your help!

Can I adjust the Stable Diffusion config to use SD3? Looks like I should be able to by pointing to the SD3 path I find in the API docs?

ChatGPT works a treat with GPT-o4, very fast!

[privet.fun]

User error nonetheless, thank you for your help!

Can I adjust the Stable Diffusion config to use SD3? Looks like I should be able to by pointing to the SD3 path I find in the API docs?

ChatGPT works a treat with GPT-o4, very fast!

I'll have to make minor updates to support Stability.AI most recent SD3 API https://platform.stability.ai/docs/api- ... ~1sd3/post
Seems like they also added support for image + prompt (image-to-image) in addition to prompt (text-to-image).

I'm honestly a bit hesitant to spend time on this since Midjourney is so massively better and also cheaper vs paid version on Stability.AI
With current https://platform.stability.ai/pricing $10 is roughly 150 single generations (images) with is a lot more expensive comparing to Midjourney $10 basic plan which if memory serves me right gives you at least 200+ something generations (4 images each) plus tons of extra features.

[nou nou]

Myeah, I was just looking at their prices - ooof! Hefty for SD3... nice results, though. Hm.