解説Algorithmically-generated landscape artwork of forest with Shinto shrine.png |
Demonstration of the usage of negative prompting on algorithmically-generated artworks created using the Stable Diffusion V1-4 AI diffusion model. The purpose of a negative prompt is to instruct the AI to omit certain objects, motifs or visual elements when generating an image, as opposed to a positive prompt which instructs the AI to include such things.
This image aims to illustrate the process in which negative prompting within Stable Diffusion can be used to fine-tune the output of an AI generated image based on the desires of the user, as one part out of three images showing each step of the procedure.
- Procedure/Methodology
All artworks created using a single NVIDIA RTX 3090. Front-end used for the entire generation process is Stable Diffusion web UI created by AUTOMATIC1111.
A single 768x512 image was generated with txt2img using the following prompts:
Prompt: Hakurei Shrine in distance, Gensokyo, nature landscape, landscape art, far view from distance, traditional Japanese architecture in distance, Shinto shrine in distance, forests, mountains, rivers, art style of Craig Mullins and jordan grimmer and tyler edlin and darek zabrocki and raphael lacoste
Steps: 50, Sampler: Euler a, CFG scale: 7, Seed: 1411213889, Size: 768x512
From there, two additional images were generated using the same seed and positive prompt, however this time using negative prompts:
- Second image
Prompt: Hakurei Shrine in distance, Gensokyo, nature landscape, landscape art, far view from distance, traditional Japanese architecture in distance, Shinto shrine in distance, forests, mountains, rivers, art style of Craig Mullins and jordan grimmer and tyler edlin and darek zabrocki and raphael lacoste
Negative prompt: green trees
Steps: 50, Sampler: Euler a, CFG scale: 7, Seed: 1411213889, Size: 768x512
- Third image
Prompt: Hakurei Shrine in distance, Gensokyo, nature landscape, landscape art, far view from distance, traditional Japanese architecture in distance, Shinto shrine in distance, forests, mountains, rivers, art style of Craig Mullins and jordan grimmer and tyler edlin and darek zabrocki and raphael lacoste
Negative prompt: round stones, round rocks
Steps: 50, Sampler: Euler a, CFG scale: 7, Seed: 1411213889, Size: 768x512
Afterwards, for all three images, the image was extended by 128 pixels on both the left and right sides using a single pass of the "Outpainting mk2" script within img2img. This was done using the same seed value of 1411213889 earlier, along with a setting of 100 sampling steps with Euler a, denoising strength of 0.8, CFG scale of 7, mask blur of 25, fall-off exponent value of 1.8, colour variation set to 0.03. The prompts used were identical to those utilised during the first step. This subsequently increases the image's dimensions to 1024x512, while also revealing additional foilage and architectural elements which were previously absent from the original AI-generated image.
Then, two passes of the SD upscale script using "Real-ESRGAN 4x plus anime 6B" were run within img2img. The first pass used a tile overlap of 64, denoising strength of 0.3, 50 sampling steps with Euler a, and a CFG scale of 7, using an identical seed of 482112941 for all three images. The second pass used a tile overlap of 128, denoising strength of 0.1, 30 sampling steps with Euler a, and a CFG scale of 7, using an identical seed of 3320472043 for all three images. |