I’m aiming to get better at telling the story to generate the images I really envisioned. Taking @krank Star@lemmy.world 's post as a aid to describe the image
Reference
Early attempts:
“Streetpunk Sexy” - Visualize a portrait of a young 22 year old woman on an medieval village street where the second stories jut over the street. rainy night,The shot, taken from a distance, captures her walking away at night under dim street lamps. She is wearing a long and flowing red dress and riding cape, the humid atmosphere ads mist to the street. messy hair escapes the red hood of the cape, The photo,taken with a Canon EOS 5D Mark III, showcases the natural smoon light and oil street lamps, in hyperrealistic detail under soft lighting. The depth of field focuses on her against a backdrop of soothing,muted tones and high contrast,creating a dark,yet gorgeous (1.2) and immersive night scene.
Negative prompt: text, watermark, low-quality, signature, moiré pattern, downsampling, aliasing, distorted, blurry, glossy, blur, jpeg artifacts, compression artifacts, poorly drawn, low-resolution, bad, distortion, twisted, excessive, exaggerated pose, exaggerated limbs, grainy, symmetrical, duplicate, error, pattern, beginner, pixelated, fake, hyper, glitch, overexposed, high-contrast, bad-contrast
Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 510970194, Face restoration: CodeFormer, Size: 512x512, Model hash: a19b862e79, Model: fullyREALXL_v30ForREAL, Denoising strength: 0.7, Hires upscale: 2, Hires upscaler: Latent, Pad conds: True, Version: v1.7.0
One trick I’ve noticed with LLMs that might translate to diffusion is that they struggle with generic pronouns. Try naming your entity and using the name to clarify and define an individual. The smaller the model, the more they seem to struggle with character identity across sentences. Just a thing to maybe try.
The prompt was built using the prompt in https://civitai.com/images/4771726 as I had aslo been trying to recreate that image.