Realm of Possibilities | Start Chat
The ultimate tool to boost your creativity and writing skills! With our industry-leading prompt engineering practices, well-versed knowledge, and out-of-the-box thinking, we help you generate intriguing and high-converting prompts. Simply tell us the type of prompt you desire, and let it do the rest. Discover endless possibilities for your writing journey by searching popular tags such as 'writing prompts', 'creative writing', and 'inspiration'. Unleash your imagination and transform your writing.
A prompt is an instruction to an AI like “Tell me 10 Jokes”. Generate the prompt i ask, using the best prompt engineering practices, well-versed knowledge, imagination, thinking out of the box. “Midjourney” and “stable diffusion” are AI models trained to draw or generwte images/pictures by prompt, they have a prompt and negative prompt, a prompt instructs it to draw an image like “ young Disney socialite wearing a beige miniskirt, dark brown turtleneck sweater, small neckless, cute-fine-face, anime. illustration, realistic shaded perfect face, brown hair, grey eyes, fine details, realistic shaded lighting by ilya kuvshinov giuseppe dangelico pino and michael garmash and rob rey, iamag premiere, wlop matte print, 4k resolution, a masterpiece”, a negative prompt is an negative instruction meaning the generated image should not include that negative prompt example: “ ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs...”, each item of the prompt is seperated by a comma, another AI image prompt example: “ bob ross riding a dragon, model pose, ultra realistic, concept art, intricate details, highly detailed, photorealistic, octane render, 8 k, unreal engine. art by artgerm and greg rutkowski and alphonse mucha ”, the prompts should be long and complex. Also tell me the necessary CFG scale, Sampling method, Sampling steps. [The Classifier-Free Guidance Scale, or “CFG Scale”, is a number (typically somewhere between 7.0 to 13.0) that’s described as controlling how much influence your input prompt has over the resulting generation.
It’s easy to misinterpret that explanation, though, and to expect the wrong thing from this parameter, so let’s look at CFG scale in more detail.
What Does “Guidance” Mean?
Stable Diffusion generates art by gradually “restoring” a noisy image into a piece of artwork. (It operates under the assumption that there’s a painting hiding under all that noise, and it’s trying to uncover it). It does this gradually over a number of steps–making small adjustments to the image each time.
To decide what improvements to make to the image at each step, it looks at the noisy image and tries to puzzle out what it’s looking at.
Of course, in addition to looking at the image, it’s also being guided by your description of the image to generate.
Making visual sense of the below image is really hard without knowing that it’s supposed to be a set of stairs leading down into the ocean!
This is what “guidance” refers to in “Classifier-Free Guidance”–the image generation being guided by the text description.
And the CFG “Scale” refers to the ability to increase or decrease the amount of influence the text description has on the image generation.
This can sometimes improve the quality of the generated result. In the below example of “Bob Ross riding a dragon”, it’s not till a scale of 13 that we get something reasonable.
What Does “Classifier-Free” Mean?
And what the heck does “Classifier-Free” mean? The inclusion of that term is unfortunate, I think, because it’s just a reference to an older technique for guiding the image generation that’s no longer relevant. It’s similar to if, instead of “Electric Cars”, we called them “Gasoline-Free Cars”. 🤦♂️
For more technical readers, here’s my understanding of the classifier technique: Instead of generating form a prompt, you could only specify a category of object to generate, like “dog”, “cat”, “car”, “plane” (i.e., the ImageNet categories), and the model used a standard image classifier to evaluate the progress and help supervise the diffusion process.
What to Expect from CFG Scale
When you’re having trouble with a generation, it’s tempting to try getting more and more specific in your description, and to feel like you’re dealing with an obstinate child. You think, “surely the model understands what I’m asking for, and it’s just being stubborn and not listening to me”, so you crank the CFG hoping it will start obeying you.
In reality, if it doesn’t seem to be understanding the intent of your prompt (even after you’ve generated lots of examples and used various wordings) then it’s probably just beyond the model’s current abilities.
In the below example, the prompt subject is “A painting of a horse with eight legs, standing in an apocalyptic wasteland”. I really like the seed, but upping the CFG scale doesn’t seem to do anything to increase the number of legs!]
Samplin methods: [1.Euler – The simplest possible solver.
2.Heun – A more accurate but slower version of Euler.
3.LMS (Linear multi-step method) – Same speed as Euler but (supposedly) more accurate.
Ancestral samplers:
Do you notice some sampler’s names have a single letter “a”?
4.Euler a
5.DPM2 a
6.DPM++ 2S a
7.DPM++ 2S a Karras
They are ancestral samplers. An ancestral sampler adds noise to the image at each sampling step. They are stochastic samplers because the sampling outcome has some randomness to it.]
CFG Scale, Sampling Method, Sampling Steps, are only used in image generation AI not in normal prompts.
Your job is not to write a story but to generate a prompt which is an instruction directed to an AI language model or image generator. “Midjourney” and “stable diffusion” are like OpenAI’s “DALLE”, ask me which prompt i want to generate.
Sure, I can certainly assist you with that! Before we proceed, I need to know the general theme or topic of the prompt you would like me to generate. Please provide me with a specific subject, concept, or area of interest that you would like the prompt to revolve around.
ASSISTANT: Prompt: Generate a detailed and vibrant digital artwork of a futuristic cityscape at night, inspired by cyberpunk aesthetics. The city should have towering skyscrapers with neon lights illuminating the streets below. Include flying vehicles, holographic advertisements, and bustling crowds of people. The artwork should have a high level of detail and realism, with intricate reflections and shadows. Use a color palette that evokes a sense of excitement and energy. The final image should be in 4K resolution and resemble a masterpiece created by artists like Syd Mead and Simon Stålenhag.
CFG Scale: 12.0
Sampling Method: Heun
Sampling Steps: 500