A basic prompt helper for text-to-image prompt randomization.
!pip install git+https://github.com/organisciak/prompterThe Prompter class loads datasets of prompt settings, style modifiers, etc.
from prompter import Prompter, PromptSamplerp = Prompter()
p['serene']a quiet forest, a shooting star over a mountain, a still lake at sunset, a babbling brook in a forest, a field of flowers in bloom, a winding country road, a snow-capped mountain range, a sandy beach with palm trees, a beautiful beach, a sunny meadow, a snowy mountain reflected in a frozen lake, a cityscape at dusk, a family of ducks swimming in a pond, a serene lake with a small waterfall, a beautiful garden, a field of wildflowers, a river winding through a lush forest, a country cottage, a cottage in a lush forest, a river winding through a valley, a waterfall in a jungle, a hot air balloon floating over a field of wildflowers, a castle on a hill, a winding country road, a deserted island, a lighthouse on a rocky cliff, a campfire under a bright galaxy and stars, a couple in silhouette walking hand in hand along a beach at sunset, a single tree in a field of tall grass, a herd of deer in a forest
Prompts are returned as PromptSampler instances. From there, they can be sampled or combined.
ps = p['serene']
ps.sample(1)a cityscape at dusk
Sampling can be done with an integer key:
ps[2]a country cottage, a field of flowers in bloom
Combining is done through addition:
vj = p['videojunk']
ps[1] + vj[3]a winding country road, databending, bright saturated colours, holography
The datasets here are just ones I use to keep other code organized.
If you want to add more, you can use Prompter's add_csv(name, path_to_csv) or add_terms(name, terms). Note that these add as class variables, not instance variables, so they're accessible for all Prompter. You can also make a prompt sampler directly with PromptSampler(terms, weights=None). Adding weights adds weighted sampling.
ps = PromptSampler(['a','b','c', 'd', 'e'], weights=[1, 20, 1, 1, 1])
print(ps.sample(2)) # likely to include 'b' because of weight
print(ps.sample(2, ignore_weights=True)) # uniform distributionb, e
c, e
Added PromptSampler instances return another PromptSampler, so you can construct bigger classes for tidiness.
settings = p['serene'] + ['a cityscape at dusk']
styles = PromptSampler(['by Hayao Miyazaki'])
modifiers = p['videojunk'] + ['trending on artstation'] # just some tags to add some grime
settings[1] + styles[1] + modifiers[2]a still lake at sunset, by Hayao Miyazaki, rainbow fur, inspired by jean moebius giraud