americanpresidentjimmycarter / yasd-discord-bot Goto Github PK

The bot edits its own message on completion, which does not notify users with a new ping, because it isn't sending a new message.

For some reason the server just won't start. I don't know why, and this is the third time ive tried it(running linux mint 21)
Im really hoping you understand these logs better than I do.

Logs:
logs.txt

You mentioned that you need 16 GB+ VRAM to run at full resolution.

I have the bot working, but it runs out of Memory instantly at any attempt

RuntimeError: CUDA out of memory. Tried to allocate 512.00 MiB (GPU 0; 7.79 GiB total capacity; 5.08 GiB already allocated; 88.56 MiB free; 5.20 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Is there any way I can either lower the image resolution or somehow use an optimized fork to run the bot on my pitiful 8GB 3060 Ti. Im just trying to set this bot up to use in me and my friends' discord server. Im also not using docker

Currently there is no info on how many buttons, how quick it is going, and when to expect the bot live again. Info like this would be useful for 'quick restarts' to inform users on downtime.

Hi i get an error on running from this line.
intents = discord.Intents(messages=True, message_content=True)
error is: AttributeError: 'Intents' object has no attribute 'message_content'

it looks like a discord.py error.
dicrod.py 1.7.3

The title

Before we could only make squares.

i get an error at start. below the error.
Traceback (most recent call last):
File "bot.py", line 26, in
currently_fetching_ai_image: dict[str, Union[str, bool]] = {}
TypeError: 'type' object is not subscriptable
running on windows. python 3.8

DMing my bot is a feature I'd like to keep, but the prompt completion responses all return errors:

Got unknown error on riff "UeeGqXSH1x18" index 3: 'NoneType' object has no attribute 'id'

There is currently no way to asynchronously set the selected aspect ratio in the client, so if multiple users are tweaking aspect ratios at the same time it will be impossible to tell what the aspect ratio actually is.

https://huggingface.co/docs/transformers/v4.22.1/en/model_doc/clip#transformers.CLIPTokenizer

Solution: use better short ID generator from stdlib eg ''.join(random.choices(string.ascii_lowercase + string.ascii_uppercase + string.digits, k=12))

https://www.youtube.com/watch?v=WsDykBTjo20
https://github.com/hlky/sd-enable-textual-inversion

Create more advanced images from other images with textual inversion

I tried to run
/interpolate and filled out the two prompts

and got
Now beginning work on "interpolate "cat sitting on a sofa" to dog sitting on a sofa" for murmur5786. Please be patient until I finish that.
Got unknown error on interpolate "cat sitting on a sofa" to dog sitting on a sofa: <DocumentArray (length=0) at 140684863006336> is empty

Consider substituting full width commas: ，

Title.
Some good tags to have:

• Which user requested the image
• Which channel the request was made in
• Which guild the request was made in
• ID of the original if it is a riff or an upscale

A quality of live improvement for the bot would be implementing buttons such as MJ has for redoing the prompt or upscaling each of the 4 images.
a good reference for buttons using discord.py seems to be https://gist.github.com/lykn/bac99b06d45ff8eed34c2220d86b6bf4.
another improvement would be that the message "Now beginning work on..." could be edited with the final result.

very nice work on the bot.

It would be good to print the legacy commands after using slash commands so that people might easily copy paste the commands of others.

Not sure if this is a thing, but my tests and the manual don't seem to include this. Negative prompting would be nice as well but based on what I've seen its more difficult to implement.

Both, Riff and upscale-buttons throw an error after a couple of minutes when pressed.

Title. All the parameters should be the same but we should be able to run it again to see more variations.

It would be nice if the same button existed on riffs as well to create more riffs from the same base image.

Image2image is already available via > command, but users in my server were having trouble understanding those commands. I managed to get it hackily working by converting the command into a slash and requiring the user to specify an image url.

Probably shouldn't disable the upscale button on failure too.

Not sure if this is already an option or if it's possible to make it configurable?

For example something like below?

• image foo bar (samples=2)

• image foo bar (images=4)

See:

        if original_request['api'] == 'stablediffuse':
            sampler = original_request['sampler']
            scale = original_request['scale']
            steps = int(original_request['steps'])
            latentless = original_request['latentless']
            strength = original_request['strength']
            await _riff(
                interaction.channel, interaction.user,
                self.short_id_parent, self.idx_parent,
                height=self.pixels_height,
                latentless=latentless,
                sampler=sampler,
                scale=scale,
                steps=steps,
                strength=strength,
                width=self.pixels_width)

Firstly love your work, great project!

Just a few things/issues I came across following the documentation for native install.

1. Got an error that the version of Jax was too old but was able to fix that by not specifying the version, which installed the latest and seems to work fine. pip install jax

2. I had to clone the original SwinIR repo because upscaling was throwing an error about missing main_test_swinir.py script and models/network_swinir.py (this was since you added the buttons, previously did not have this issue.)

3. Not related to your code but updating to python 10.3 on Ubuntu 20.04.5 LTS was a massive pain but I eventually got it working.... More of an FYI for Ubuntu users.

Love the new buttons - just a minor improvement. When a riff-button is pressed it is deactivated. This is fine for upscaling but if I want to do another riff of the same image I have to fall back to the traditional copy-pasting.

Thank you for your work - huge improvement!!

Bug causing a deep fried effect on every 4th image.

Stuff like /image prompt:testing 123, test fooo height:384 sampler:ddim scale:10 seed:123 works too if you hit tab after pasting it in.

Title. It makes it difficult to manage generations as the bot has to be babysat to try out multiple ideas

When using the manual ">upscale" command the bot ignores the index and uses 0. However when you use "/upscale" or the buttons it works fine.

Getting an error after issuing >help command in discord, any ideas?

discord.ext.commands.context.Context object at 0x7f68d61344f0> Command raised an exception: AttributeError: 'NoneType' object has no attribute 'id'

Some resolutions do not work, here is a list
Square resolutions 512x512 and above = working
Square resolutions less than 512x512 = noisy, fuzzy, weirdly noisy
Any vertical or landscape resolutions = stretched into a square
Vertical or landscape resolutions less then 512x512 = noisy, fuzzy, weirdly noisy AND stretched into a square

Assuming all resolutions are multiples of 64
resolution edited in dalle/dalle-flow/executors/stable/config.yml

Add option for users with certain role bypass 72 hour wait time.

git clone https://github.com/CompVis/latent-diffusio324n.git
git clone https://github.com/StableDiffusion/latent-diffusion.git

both do not exist