Posts

@mp40
Oh, also, one thing worth mentioning is that I think the longer the system prompt is, the more likely it is for the system prompt to start going out of the context window. I’ve noticed that since the instructions to uncensor it are at the beginning, it tends to start becoming censored again if you put too much in the system prompt.

@mp40
No problem. It’s one of these spots where I really need to play more with it, and there might be better ways to do some of it, but that’s what was getting me results.

I remember one oddity was that autocomplete on what I was typing kept giving I can’t talk about this topic type lines, but the actual response was uncensored.

There could easily be better models, too. I just remember trying two or three and this was the one that was giving decent results.

@Lord Waite
Thanks!

@Adusak90
I would hope so!

Added links to the static pages to the OP,
https://tantabus.ai/pages/lorachar
https://tantabus.ai/pages/lorachar-minor
https://tantabus.ai/pages/lorachar-bg

There are a lot of characters, still have to go through several dozens that I’m already aware. And I need to generate some 60 preview images, the default for many tend to be anthro / borderline NSFW…

But hopefully in a week or two this’ll make a nice useful list of character LoRAs.

If this is going to eventually bloom into more background pony art, as an enjoyer of these mares you’re doing a god’s work right here, I applaud and cheer for your noble mission.

@mp40
I haven’t done more with it, but with ollama, the key was making a custom “Modelfile” file, and creating a model from that Modelfile.

What you can do is copy the modelfile of an existing model and modify it.

So, first:
ollama pull rolandroland/llama3.1-uncensored

to install the modelfile you are going to base it off of. Then, if you run:
ollama show rolandroland/llama3.1-uncensored --modelfile

It’ll print out on the console that models Modelfile, so just copy that to a file named Modelfile.

Then change the FROM section to say:
FROM rolandroland/llama3.1-uncensored:latest

and add a section at the bottom that says:
SYSTEM"""<your prompt here>"""

And just write a prompt for how the AI is going to act there. You basically want to describe to it what its purpose is, and let it know that it’s uncensored and can describe sexual acts and such, tell it not to add in disclaimers, tell it the exact format that a prompt should be in and the type of words it should use, and give it a few examples of real prompts.

(I’d give one here, but looking at it, I really want to clean it up and improve it. I was explicitly telling it to add the line of score tags, then a source and rating tag, then a description, then several paragraphs of danbooru tags.)

Then run:
ollama create <new mode name> --file Modelfile

Keep the modelfile, try using the model you generated, and if you want to tweak it, do:
ollama rm <model>
change the modelfile, and rerun the create command.

That’s basically how to do it, in any case, the key is going to be playing with creating a prompt until something sticks, and basing it off the right model, as I remember trying it with a different model or two and not having as much luck…

Set up a page to list these with an index and tighter format.

Will be adding G5 and OCs to that page… tomorrow maybe.

Please do list more if you know of them!

@tyto4tme4l
I find this one gives weird aspect ratio screwups on occasion, which is mildly annoying.

@Background Pony #D6A2
“Low Rank Adaptation”

Very short version: it’s a model mod that adds/modifies information in the main model, so you can prompt stuff the main model didn’t have.

Velvet Remedy (Fallout: Equestria)

Base Model: PonyV6
Trigger words: velvet remedy, pony, fallout equestria

what is a lora?

(reserved… there’s a lot)

(reserved)

(reserved)

Not all characters are well known (or at all) in PonyV6 or other SDXL models, so LoRAs are often required for quick gens. Multiple LoRAs for lesser known characters already exist, but are not always easy to find. Herein I’d like to list known ones.

Of course, please comment in the thread with the ones you’re aware of, and any additional info you might have for it, like base model (PonyV6 for 99% of them, but still), trigger words, strength, etc. and I’ll add them to the first post(s) the page above as soon as I can.

@Lord Waite
Have you done anything else with this? I’m looking for resources on how to build or curate a llm of my own but the “uncensored” model is still denying some of my prompt requests, do I just need to try other jailbreak prompts till somthning works or?

I believe for ADetailer, you’d be installing the Comfy-Impact Pack. (Which is easiest if you have the ComfyUI Manager installed, of course.)

And yeah, definitely will take a bit of getting used to. If you go under Workflow->Browse Templates, the Image Generation template there is pretty much the default one. Change the model to pony v6, change the width and height, put a prompt in, and hit Queue…

Watching some videos will probably help, just some things won’t match because the ui has changed.

Finally got ComfyUI to work on my end, but the UI looks a bit overwhelming, so need to set aside many consecutive hours one day to really dig into it.
And I must be blind, because I couldn’t find ADetailer in it. Will look more into it later though.

I always feel like it helps to have a bit of a base understanding on how the models work on these things.

Initially, someone created a large dataset of images and descriptions. The descriptions were tokenized, and the images cut up into squares. Then, random noise was generated based on a seed. It took one square, generated random noise based on a seed, and attempted to denoise the noise into the image on the square. Once it got something close, it discarded the square and grabbed another one. At the end, all of this was saved in a model.

Now, what happens when you are generating an image is that your prompt is reduced to tokens by a text encoder (XL based models use CLIP-L and CLIP-G), random noise is generated by the specified seed, and then the sampler and noise schedule is how it denoises, with as many steps as you specify.

Some schedulers introduce a bit of noise at every steps, namely the ancestral ones (with an a at the end), and sde, but there may be others. With those ones, the image is going to change more between steps and they’ll be more chaotic. Also, some will take less steps then others to get to a good image, and how long each step takes will vary a bit. I believe some are just better at dealing with certain things in the image, too, so it’ll take some playing around.

Now, the clip text encoder actually can’t cope with anything more than 77 tokens at once, and that includes a start and end token, so effectively 75. So if your prompt is more than 75 tokens, it gets broken up into chunks of 75.

The idea behind “BREAK” is that you are telling it to end the current chunk right there and just pad it out with null tokens at the end. The point is just that you’re making sure that particular part of the prompt is all in the same chunk. I’ve had mixed results on it, so I try doing it that way occasionally, but also don’t a lot of the time. It is going to have trouble with getting confused anyways. This is just an attempt to minimize it a bit.

(Text encoding is one of the differences between model architectures, too. 1.* & 2.* had one clip, XL has two, then when you start getting into things like flux and 3, you start dealing with things like two clips and a t5 encoder, and the t5 encoder accepts more like 154 tokens. I also didn’t get into the vae, which is actually what turns the results into an image…)

I’ve seen some guides mention to use BREAK in prompts to help guide the model. E.g.
Description of scenery
BREAK
Character 1 wearing denim jeans and red sweater sitting on a bench
BREAK
Character 2 wearing black suit with bowtie walking in the background

But I’m not having much success with it, it still gets confused as to who wears/does what.
Any of you using it successfully?

The quantity of steps depends on the sampler, for Euler it’s 25+ sampling steps, but sometimes it can be lower. I guess it depends on composition and it’s never constant. I recommend just trying different settings and checking if increasing the steps substantially improves the image

@Scarlet Ribbon
Okay, interesting.

In your opinion, what sampler is best to use when creating pony images in Pony Diffusion? On CivitAI, the default appears to be DPM++ 2M Karras.

@MareStare
Okay.

Another question - generally, how many steps for generation produce the best results?

@Zerowinger
You can use full sentences to describe the prompt with Pony Diffusion as well. Citing from their page on civitai the recommended prompt format:

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, just describe what you want, tag1, tag2

where tag1, tag2 are simple words/word combinations similar to derpibooru tags like “unicorn, blushing, trio, duo”, etc

@Zerowinger
Different models are trained in different ways, leading to some models being better for natural language, and others better for tag-based prompting. Pony doesn’t completely fail with natural language prompting, but in my experience it performs much better with tag-based. If you add source_pony to your prompt, you can damn near just use Derpi/Tanta tags to get most of the results you’re looking for.

@MareStare
So basically, including that string is necessary for higher quality images then? What about the rest of the prompting? On Imagen, I’m used to using full sentences and phrases to describe exactly what I want the output to be, with Pony Diffusion it seems the go to format is to list each individual aspect as a prompt separated by a comma.