Furkan Gözükara
patreon
How To Get Amazing Prompts With ChatGPT For Stable Diffusion
🕑 Added 2023-08-01 22:56:02 +0000 UTCChatGPT share link : https://chat.openai.com/share/04215849-bcba-45a4-b723-36b120025f25
Not every prompt will be good or useful. But this can lead some very good prompts. This is the strategy that I found. If you have better strategies please let me know.
I used notepad++ to replace . at the end with LoRA name
Hopefully new 12 GB LoRA SDXL tutorial coming. Fully up to date. I am trying to come up with some nice prompts
Join discord and tell me your discord username to get a special rank : SECourses Discord
The below are the prompts I used on chatGPT. Modify it as you wish. In some prompts since they are distant, face inpainting might be necessary.
You can repeat the process and get thousands of different prompts.
Negative prompt : cgi, render, drawing, cartoon, naked
this prompt "closeshot photo of ohwx man hiking on a snowy mountain" is a prompt i use to generate images with generative ai tools. i want you to write me 100 different ones. you have to keep ohwx man words. ohwx man is the subject of the prompt so it has to be in every prompt you generate. for example instead of snowy mountains and hiking use your imagination and describe other images . when writing output do not give numbers such as 1. 2. etc. give plain text each line . make the outputs as describing an image but it is described by AI model such as clip captioning . you can make prompts more detailed. try to come up with really really interesting prompts. you can also add some styling words. few examples
ohwx man driving a fast car on a dirty road
ohwx man flying on the sky with metal wings
ohwx man wearing a fancy suit
generated results (540 i can share more if you wish) 540 prompts.txt
Comments
Furkan Gözükara
nice
Kallamamran
You should just search for "Diffusion" here: https://chat.openai.com/gpts I'm pretty sure you can find some good ones there
Pallavi Chauhan
Can you turn that into a GPT and share with others?
Furkan Gözükara
yes you can do that too
Erik
Just noting that you can also put this prompt into Windows Co-Pilot to get prompts
XyRu
Thanks a lot for the detailled explanation and your time - really appreciate it. :)
Kallamamran
Manual labor dude. Manual labor ;) I started with: "I want you to act as a AI prompt generator for a generative AI generating images with a tool called Automatic1111. Stable Diffusion generates images based on given text prompts which the AI interpretes and uses to create an image. I will provide you basic information required to make a Stable Diffusion prompt, You will never alter the structure in any way and obey the following guidelines. Ok?" After ChatGPT's answer I just kept pasting in sections of the PDF from beginning to end chosing the most relevant information piece for piece anding each section with "Just answer with Understand if you understand and are ready for more info". After the whole document is cycled through I gave it 14 example prompts to derive information from and learn the format from and then: "Feel free to add your own of [STYLE OF PHOTO], [SUBJECT], [IMPORTANT FEATURE], [MORE DETAILS], [POSE OR ACTION], [FRAMING], [SETTING/BACKGROUND], [LIGHTING], [CAMERA ANGLE], [CAMERA PROPERTIES], or [PHOTOGRAPHER] if you have something more in your database distinctly describing any of these sections. Is there something you feel is missing for you to be able to construct a good prompt like the example prompts with the information I have given you?" It then told me about a few additional considerations about the different sections and I gave it artistic freedom. by telling it "OK. Feel free to add your own additions from above here and lets try to create a prompt for a stylish swedish woman in a coffe shop. Put the prompts in a tabular format and also add a line below the table with the full comma separated prompt" It gave me a table with the different characteristics of the prompt and added this below to be copied to A1111: "High fashion photography of a stylish Swedish woman, with blonde, wavy hair cascading over her shoulders, wearing a minimalist Nordic-inspired outfit, sipping coffee elegantly from a ceramic mug, upper body framing, in a cozy Scandinavian coffee shop with wooden interiors and large windows, under soft natural lighting from the overcast sky outside, shot at eye level, on a Fujifilm X-T4 with a 50mm lens, with a slight grainy film texture, in the style of Alfred Stieglitz." I had to tell it to update the information regarding the [CAMERA ANGLE] so it doesn't describe WHY use a certain camera angle. It's enough with just "Low angle" instead of "Low angle to emphasize dominance and stature" Same with [CAMERA PROPERTIES]; "Shot on ARRI ALEXA 65 for detailed, crisp imagery with a touch of film grain" would be better as "ARRI ALEXA 65, detailed crisp imagery, slight film grain" It understood and did exactly as I told it :) This ended with prompts like these... Please write me 3 very detailed and descriptive prompts on a fierce dwarf warrior in medieval Lords of the Rings style: "Epic fantasy photography, Fierce dwarf warrior, Braided beard adorned with golden rings, Wearing intricately designed armor, wielding a double-headed battle axe, Charging into battle with a war cry, Full body, Rocky mountain pass with a looming storm overhead, Dramatic edge lighting with dark clouds casting shadows, Low angle, Hasselblad X1D II, fine texture, smooth color gradations." "Cinematic fantasy portrait, Fierce dwarf warrior, Scarred face from countless battles, Clutching a rune-etched warhammer, chainmail and leather armor, Standing defiantly with a determined gaze, Upper body, Ancient dwarven hall illuminated by torches, Warm golden hour lighting with flickering torchlight, Eye level, ARRI ALEXA 65, detailed crisp imagery, slight film grain." "Atmospheric fantasy shot, Fierce dwarf warrior, Tattoos depicting ancestral lineage, Carrying a shield with clan emblem, horned helmet, Guarding a fortress entrance, Full body, Stone fortress with a misty mountain backdrop, Diffused lighting with radiant god rays piercing through mist, High angle, Fujifilm X-T4, saturated colors, high contrast tones, natural grain." I'm happy with this for now. You need more info?!
XyRu
I would also be interested in the instructions/more details, how you trained GPT4. You mentioned that PDF from Civitai for example. How did you provide the prompt content for ChatGPT 4 to use it for generating your prompts? Do you use a plugin? Thanks a lot!
Mikael Svenson
I mean, to get prompts via GPT you need to instruct it how to create prompts, and then you pass in like "photorealistic image of foo", and it gives you the full prompt. So looking for the instructions/input you trained with :)
Kallamamran
Can you specify clearer what you mean? Do you want me to share the prompts I have created, because I can create how many I want, or do you want me to somehow share how I have trained the AI, because that was described in my original answer... I think!? 🤔
Mikael Svenson
Kallamamran, can you share some of the GPT instruct prompts you have assembled?
Furkan Gözükara
Wow nice thanks
Kallamamran
Bidding: Please write me a very detailed and descriptive prompts on a fierce dwarf warrior in medieval Lords of the Rings style Answer: Atmospheric fantasy shot, Fierce dwarf warrior, Tattoos depicting ancestral lineage, Carrying a shield with clan emblem, horned helmet, Guarding a fortress entrance, Full body, Stone fortress with a misty mountain backdrop, Diffused lighting with radiant god rays piercing through mist, High angle, Fujifilm X-T4, saturated colors, high contrast tones, natural grain.
Furkan Gözükara
awesome thanks for the tips
Kallamamran
I have actually trained GPT-4 on information from several different prompt-guides online. For example the "Creating Photorealistic Images With AI Using Stable Diffusion"-pdf by PromptGeek (Found on Civitai) :P Works like a charm! You should try that. It takes a "little" more effort, but the prompts gets really good, even for 1.5!
