[AI Generated] Random 4:30 AM experiment with ChatGPT

LuoirM

Voidiris' enthusiast feet enjoyer.
Joined
Mar 5, 2021
Messages
1,435
Points
153
I asked ChatGPT to analyze one of my old poem about a woman in my dreams, then I told it to generate me some prompts based on what it think about the poem so I could run those prompts in a different model engine, this is my document.

Part 0. Model and website
I was shown by a senpai, and I don't know if I'm allowed to share his settings, but the site is Tensor

Part 1. Rough translation of the poem from Vietnamese to English, once again using GPT
She stood on the windy hill, a radiant light,
Pausing to gaze at me I froze in awe.
Snow fell on golden streets beneath the darksky,
She wrapped her scarf beside the person who slept peacefully.

I shall lie down and die beneath the shade,
While golden sunlight glimmers on the dreamy lake.
I vow to kneel before the light in her eyes,
When winter melts, revealing the corpse of a young man.

Part 2. ChatGPT's prompt after asking it to break it down to keywords and told it I like anime style
anime style, beautiful girl, (windy hill), (twilight), glowing aura, flowing scarf, snow falling, golden city lights, dark sky, (serene expression), distant gaze, melancholic, ethereal, soft lighting, frozen lake, surreal shadows, long hair, white coat, delicate features, poetic atmosphere, dreamlike, winter theme, cinematic composition
I added in:
High resolution, 8K, High visual, best quality, beautiful illustration, amazing quality, high resolution, absurdres, ultra-detailed, depth of field, newest
1745097916353.png

Part 3. Asking ChatGPT to switch the mood
I just ran it, it seems the model focuses more on the ice lake and the evening, can you try writing the prompt so that it focuses more on sentence 1 "She stood on the windy hill, a radiant light," and sentence 6 "While golden sunlight glimmers on the dreamy lake.."? And imagine the female character's appearance more according to your ideas.

Part 4: Its prompt
anime style, (beautiful glowing girl), (windy hill), long white hair, golden eyes, soft smile, radiant aura, (sunlight halo), flowing dress, white and gold outfit, ethereal, elegant, standing gracefully, glowing in the wind, golden sunlight, soft light rays, shimmering atmosphere, dreamy background, (sunlit lake), light breeze, poetic feeling, serene, majestic, fantasy, cinematic lighting, warm tones
I added in:
16k, masterpiece, best quality, highres, absurdres, very detailed, cinematic lighting, high detail
1745098371827.png

Part 5. Disastifaction
The result too absurd, not real or human-like which I expected, I ask ChatGPT to do better
New prompt:
anime style, (beautiful girl on a windy hill), soft sunlight, golden hour, white scarf flowing, shoulder-length black hair, (warm smile), soft expression, brown eyes, standing gently, (subtle glowing light), natural lighting, autumn coat, calm breeze, dreamy mood, sunlit lake in background, poetic, grounded atmosphere, cinematic anime frame, realism, gentle colors, peaceful
I added
16k, masterpiece, best quality, highres, absurdres, very detailed, cinematic lighting, high detail
Negative prompt:
halo, angel wings, divine, glowing skin, holy, fantasy armor, god-like, magical girl, elf, ethereal creature, surreal anatomy, overexposure, sparkles, glowing eyes, unnatural colors
I didn't add anything

Part 6. Heartbreaking final results
1745098637231.png


Part 7. Conclusion
Write your own damn prompts.

Signed,
LuoirM | SAVI | DesperateWorshipper
 

Tyranomaster

Guy who writes stuff
Joined
Oct 5, 2022
Messages
746
Points
133
I'm working on a project that does this. The trick isn't to have it write the whole prompt. It's to know what goes into a good prompt and ask it the right questions to get the answers. That project is on the backburner because I have 2 other projects that have deadlines coming up though.

Btw, this is essentially what "thinking" mode is. The AI is given fixed prompts that cause it to give analytical responses that eventually fills in a new prompt with a deeper informed question. The issue that arises is that "thinking" can end up down an unrelated rabbit hole if the AI missunderstands something.
 
Top