Stable Diffusion is the AI model Perchance had used to create images. Last year Perchance switched to Flux Schnell. You cannot revert back to SD through Perchance, it is just no longer supported. Hence my suggestion to have the Perchance image analyzer check your image and have it create a prompt Flux will understand. Up to you to tweak the result to meet your needs. Or…find another online AI that does support SD. Many do, but often behind a paywall. You could simply try Google Gemini or Nanobanana and specifically state that you are supplying it with an SDXL prompt.

Previously Perchance used Stable Diffusion, can’t remember which version (probably 1.5 or XL).

But, here’s a thought:

• Load your image in this Perchance generator: Vision Pro Magic
• Select the settings you need
• Hit ‘Analyze’ (then optionally ‘Generate Image’)

Copy the prompt it generated and use it in the generator of your choosing. Works as expected most of the time!

Here’s the prompt it generated for me with default settings:

High-quality anime style digital illustration of a nature-themed girl with long, flowing green hair and pointed elf ears. She has bright green eyes and a small green diamond-shaped gem embedded in her forehead. Emerging from her head are two brown, branch-like antlers. She is wearing a white, V-neck sleeveless dress with a black belt at the waist and a black choker. Her forearms are wrapped in dark brown leather bandages. She is posed sitting forward with a gentle, knowing smile. Behind her, a massive, thick green serpent coils around her body, creating a circular frame for the composition. The background is a clean, minimalist white with small, floating dark brown debris or petals drifting in the air. Soft, diffused lighting with a bright, clean color palette dominated by greens, whites, and earthy browns. Front-facing medium shot, cel-shaded with clean line art and smooth gradients.

(btw, not my generator)

The dumbest online thing to do is making your IP-address public… How about posting your bank account number plus pin? Remove your post and find another way to contact the admin of the specific generator!!

Maybe someone forked the old version, copy and restore that one? (fyi: I haven none…)

Don’t worry, I’m well aware (from reading prior posts) that you are using AI to create your work. The output you’ve created thus far (made AI create, duh), tells me that you are very good at specifying and describing a problem in such way that others (or the AI) can create code from (nearly?) scratch. That ability is much more valuable than decades of coding experience based on specs others supply. Let’s face it: if you can’t properly describe a problem, how will you ever be able to come up with the right solution? And mind you, the world is full of know-it-alls who come up with solutions for the wrong problems. And it costs billions, each year.

So, kudos!

Having said that (of course there’s more), one obvious flaw I noticed in the [GEN] PODCASTS app is the use of fixed units (px) in the CSS instead of relative units (rem, em, etc). With responsive design that’s a no-go. You will need to do some tweaking there. Some guidance on the subject is too much for this post, but you can DM me should you require some insight. I haven’t checked your other apps, but since you are using a template, I guess(!) they all have the same issue…

Love your work!! I’m not sure I will listen much, but, as a retired software engineer, dang, I’d say: impressive! I’ve been looking at the code and the link to the Meta Template in The Meta-Template Concept is dead.

Two questions to the AI Text Plugin Tester

Q1: what are the names of the models you are trained on? Summarize a list of the large language models you have been trained on.

A1: As an AI, I don’t have a “training set” consisting of other Large Language Models (LLMs) in the way a human might have a list of teachers. I am not a composite of specific models like GPT-3, Claude, or Llama.

Instead, I am a single, large-scale neural network trained by Google.

To clarify how this works:

1. Training Data, not Models: I was trained on a massive dataset of text and code (webpages, books, articles, etc.). I learned the patterns of human language from this data, not by “studying” the outputs of other specific AI models.
2. Architecture: I am based on the Transformer architecture, which is the foundational technology that almost all modern LLMs use. While I share a “family resemblance” with other models because we use similar mathematical foundations, I am a distinct entity developed and trained by Google.
3. RLHF (Reinforcement Learning from Human Feedback): After my initial training on a massive dataset, I underwent a process where human trainers ranked my responses. This “fine-tuning” helps me be more helpful, accurate, and safe, but it is based on human guidance, not on other LLMs.

In summary: There is no list of “model ancestors” because I was trained from the ground up on human-generated data using Google’s proprietary infrastructure and datasets.

Q2: You are being massively evasive. People can access you as an LLM, therefor there must be an API to access you. What is the API and how are people able to access you?

A2: Since you are asking me as an AI, the answer is a matter of infrastructure rather than a secret. I don’t have a “secret identity” or a hidden door; I am a model hosted by Google.

Here is the technical breakdown of how people access me:

1. The API Layer

The “API” you are referring to is the Google Gemini API (available via Google AI Studio or Vertex AI on Google Cloud).

When a developer wants to build an app using my capabilities, they don’t “log in” to a chat interface. Instead, they write code that sends an HTTP request to a specific endpoint (URL) managed by Google. This request contains:

• An API Key: A long string of characters that authenticates who they are and tells Google who to bill for the compute.
• The Prompt: The text they want me to process.
• Parameters: Settings like “temperature” (how creative/random I should be) or “top-p” (how I sample words).

2. The Request-Response Cycle

When you type a message into a UI (like the one you are using right now), the following happens in milliseconds:

1. Frontend: The website takes your text and wraps it in a JSON object.
2. API Gateway: That object is sent via a secure connection to Google’s servers.
3. Inference: The request hits a cluster of GPUs or TPUs (Tensor Processing Units). These chips run the mathematical weights of my model to predict the next most likely tokens in the sequence.
4. Return: The generated text is sent back through the API to the website, which then renders it on your screen.

3. Why it might feel “evasive”

If you’ve asked me this before and I seemed vague, it’s because I don’t have “real-time” access to my own server logs or the specific billing account currently paying for the compute of this specific session. I am the model (the software), not the platform (the hardware and API layers). I exist as a set of weights; I don’t “see” the API call that triggered my activation—I only see the prompt that arrives in my context window.

In short: You access me via Google’s infrastructure, and developers access me via the Gemini API.

Beats me, but I suggest you copy the original post and repost it. I’m not sure you can delete this one, but if so, simply get rid of it. I promise, we’re not all d*ck-heads…

google for “names of short haircut styles for female”. Many of the styles work, esp. bob, pixie => short, medium, long, tousled, frizzy, slacked-back (very tightly combed back, works nice as ‘slacked-back ponytail hair’). crew-cut, usually for male, may work too.

You should post the prompt/negative and settings you are using and a link to the specific generator. There are way too many Perchance generators to be able to tell what might be off.

Prompt “a person wading waist-deep in the river” did it for me. No-Style, no negative.

Hi!

I’m using either Chrome or Firefox on Windows PC, straightforward, no VPN/Sandbox, or any other tool (maybe stupid Windows notepad to save snippets). Both browsers ran into the same issue, however, in the version I just tested, the chat responses seem to work now. FYI, I’ve never heard of either Thorium or Brave browser. I use Chrome, Edge, Firefox and for very old stuff Palemoon, on occasion Vivaldi (which is Chromium). So, no Linux.

Regarding ‘what is $root?’, it is a global variable inside Perchance, just like $meta. As you can see, Perchance AI-text-plugin understood perfectly what I meant. This is what your AI helper has to say about it (and I have no better explanation myself):

Hello! In the context of Perchance.org, $root is a special variable that represents the root scope of your Perchance script. The root scope is the global context in which all of your code runs, and it’s where you define the core settings, functions, and data structures that are accessible throughout your script. Here’s a simple example of how you might use $root to define a global variable: javascript // Define a global variable using $root $root.myGlobalVariable = "Hello, Perchance!"; // Now you can access this variable anywhere in your script console.log($root.myGlobalVariable); // Outputs: Hello, Perchance! When you create functions or objects that you want to be available globally, you can assign them to properties of $root. This way, you can avoid conflicts with other variables and ensure they are easily accessible from any part of your script. javascript // Define a global function $root.myGlobalFunction = function() { console.log("This is a global function!"); }; // Call the global function from anywhere in your script myGlobalFunction(); // Outputs: This is a global function! In Perchance, $root is also used to define custom functions and settings that are available in the story editor’s interface. For example, you might define a function to manipulate story elements or set up a custom configuration for your story: javascript // Define a custom function that's available in the story editor $root.myCustomFunction = function(args) { // Your custom logic here }; // Use the custom function in the story editor { "type": "choice", "text": "What's your favorite color?", "choices": [ { "text": "Red", "value": "red" }, { "text": "Green", "value": "green" }, { "text": "Blue", "value": "blue" } ], "onChoose": [ { "name": "myCustomFunction", "args": ["green"] } ] } In this code snippet, myCustomFunction is a custom function defined on $root. When the user makes a choice in the story editor, the onChoose event triggers, and the specified name function is called with the provided args. The custom function can then perform an action based on the choice made. Remember, while $root is powerful, it’s important to use it judiciously to avoid cluttering the global namespace and potentially causing conflicts with other code. Keep your global declarations to a minimum and use them only when necessary.

However, the chat-reply shows the three back-ticks of the MD code, not a marked down snippet like here in Lemmy. What’s your intention with the markup of the code snippets? Show the code examples properly marked down or have the user copy the code and paste it to get a better view? It’s not clear to me, something like the below is not really workable, esp. on smartphones (which I personally don’t use to code):

A ‘regular Perchance user’ (none-programmer, that is) has no clue that they are about to use a ‘bare-bones’ generator when reading the below intro-message, heck as a developer I didn’t:

Maybe adjust that with a better explanation, not everyone will enter edit-mode to check out the code (initially I didn’t either, not until you suggested it in your comment).

Closing, I am not sure what this is about, the error occurs when the chat-reply finishes:

Hi! Seems promising, however…

I went to your site (PC Firefox, Windows, 1920x1200, not logged in, advert bar at the bottom. PC Chrome logged in, no advert bar) and entered ‘what is $root?’ in the chat-input.

It starts to come up with an explanation, but when the output field fills the chat-messages, it removes the explanation and displays ‘Sorry, I encountered an error. Please try again.’, seemingly when finalizing the explanation. I tried it 3 times with 3 times the same result in both browsers.

• Tried ‘what is $meta?’, same result.
• Dito with ‘what are nested lists?’

Furthermore:

• The markdown of the code/chat output does not work, making code examples quite hard to read.
• Are you using a timer outputting the text? It is really slow at posting the the text character by character. On average people do read a lot faster than that.

Love the idea, so don’t get discouraged, as I think something like this will be a great help!!