Fixed dynamic prompt word seed, fixed sampler seed, still sampling.

[wencxxxxxx]

Fixed dynamic prompt word seed, fixed sampler seed, still sampling.

[cbaoth]

I just had the same issue and had a quick look at DPAbstractSamplerNode to fix this issue in the Random Prompts node:

    @classmethod
    def IS_CHANGED(cls, text: str, seed: int, autorefresh: str):
        if autorefresh == "Yes":
            # Force re-evaluation of the node
            return float("NaN")
        else:
            # Re-evaluate only in case the seed and/or text has changed
            m = sha256()
            m.update(str(seed).encode())
            m.update("".encode() if text is None else text.encode())
            # print(f"Hash: {m.hexdigest()}, Seed: {str(seed)}, Text: {text}") # debug
            return m.hexdigest()

The original code always returned float("NaN") meaning the node will always be treated as changed, according to this.

[wencxxxxxx]

Is the author still maintaining this node?

[wencxxxxxx]

Please optimize this bug, can you

[PrometheusDante]

@cbaoth I tried your fix by editing the sampler.py to this :

@classmethod
def IS_CHANGED(cls, text: str, seed: int, autorefresh: str):
    if autorefresh == "Yes":
        # Force re-evaluation of the node
        return float("NaN")
    else:
        # Re-evaluate only in case the seed and/or text has changed
        m = sha256()
        m.update(str(seed).encode())
        m.update("".encode() if text is None else text.encode())
        # print(f"Hash: {m.hexdigest()}, Seed: {str(seed)}, Text: {text}") # debug
        return m.hexdigest()

RETURN_TYPES = ("STRING",)
FUNCTION = "get_prompt"
CATEGORY = "Dynamic Prompts"

But it still insists on sampling over and over unfortunately

[cbaoth]

@PrometheusDante Sorry, I forgot to mention that you also need to add the following line ...

from hashlib import sha256

... somewhere on top of the file, e.g. like this:

from __future__ import annotations

import logging
from abc import ABC, abstractproperty
from collections.abc import Iterable
from pathlib import Path
from hashlib import sha256

Otherwise you should see something like this in the console output:

WARNING: name 'sha256' is not defined

[PrometheusDante]

@cbaoth Thanks for getting back to me. I switched to the Impact Pack's wildcard processing with a .yaml file in the meantime and like that quite a lot, with the only downside for me being the absence of immediately evaluated variables to use in multiple points of a prompt.

[4lt3r3go]

there's still no workaround for this or any other nodes to solve this issue?

[cbaoth]

There are other wildcard nodes, but all of them had some annoying issues or limitations that were mostly solved when I found this one here, so I'll intend to stick with it for now.

Since it doesn't seem to be maintained any more / at the moment, and it still seems to works apart from this particular issues that I was able to solve, I created a fork for myself where this issue is fixed. In addition I made a few quick changes / extensions that are more to my liking. There is now an advanced node version that has all the settings (plus some new ones) and the regular one which has no options at all (simply uses my preferred default). Since I tend to "stack" various prompts (fragments) I like the node to be as compact as possible. The seed I always convert to an input and use a global one for things to be (more or less) reproducible, so for me there are no widgets left apart from the text input.

My fork is here), but I don't intend to really maintain this version either since the whole comfyui thing is just one of many side-projects for me, and I sadly don't have the time and energy to work on so many fronts at once. So no guarantees but feel free to try if it works for you. I also can't guarantee that I'll not make any (more) incompatible changes in the future, but if the current version works for you, it is always possible to make your own fork with a single click.

[4lt3r3go]

My fork is here)

thank you sir

[LunaticSongXIV]

@cbaoth THANK YOU SO MUCH, this fixed every problem I've ever had with this node, and no other node I've found has proper feature-parity with A1111's module.