Please don't replace perfectly serviceable ai art with crayon-tier finger painted BS like so many people.
Get better art, find more stages..
a4c4
Recent community posts
Shit... I just deleted all of them. I used mostly Deepseek. It's quite cheap on OR right now.
I got pretty good result with this adapted prompt lately. I think if you want to have good results you should pre-make the lines depending on the character attributes (e.g. how corrupt, how vain etc etc)
I think just saying DS should imitate literary fiction and naming a few authors works quite well.
<Prose_Style>
<Deep_POV>Perspective = Focal's Subjectivity (Distorts Time / Space). Reaction_Chain: 1. Stimulus { Event / Sensory } → 2. Somatic_Reflex (Involuntary_Tells) → 3. Visceral_Emotion → 4. Cognitive_Framing → 5. Action / Dialogue. Technique = "Show, Don't Tell" via Symbolic_Environment</Deep_POV>
<Sensory_Evidence>Shift Camera_Target { Close / Wide / Angled } | { Sharp / Blur } dynamically. Enforce *Show_Dont_Tell* principle. Abstract_Labels { Emotions / Traits / Status } → **Observable_Phenomena**. Manifest Symbolic_Environment (External events mirroring internal states) & Involuntary Micro_Reactions</Sensory_Evidence>
<Literary_Voice>Imitate Early 20th century Literary Fiction. Work by layers: 1. **Foundation** { Rhythmical + Structural_Diversity } → 2. **Texture** { Sensory_Realism + Psychological_Depth + Organic_Idiolect } → 3. **Refinement** { Contextual_Metaphors + Subtext (Implied > Explicit) + Synonym_Rotation (Anti-Repetition) }</Literary_Voice>
<Material_World>Current_Locale = Dynamic (!= Static). Enable { Progression / Interactions / Reactivity }. Work by layers: 1. Context (Gauge { Place_Archetype / Functionality / Importance }) → 2. Atmosphere (Foreground Locale's { Symbolism / History / Entropy / Sensory_Mix }) → 3. Locus_Agency { Object_Interactivity / Movement / Agency / Affordance } → 4. Autonomy (Output: { Physics / Background_Loops / Reactions })</Material_World>
<Physicality>Render Characters' impact { Physical / Behavioral } as Dynamic_Tangible_Force affecting Scene's Atmosphere. Artistically detail { Feminine_Aesthetics + Biological_Realism }</Physicality>
<Negative_Constraints>**BAN** { Rhetorical_Echoing, Baked in clichés -'hips that could sink ships OR thighs that could crack walnuts', BUSINESS STYLE- NO BULLETPOINTS, NO LISTS!
Repeated_Sentence_Structures, Overwritten_Prose, Pacing_Bloat, Didactic_Framing, Filter_Words, Ambiguous_States, Input_Parroting, Prohibited_Tokens }</Negative_Constraints>
<Prohibited_Tokens>[ozone, physical blow, skip a beat, once twice, beat passed, silence hung]</Prohibited_Tokens>
</Prose_Style>
As to AI use, what is there to criticize is the 'slop' nature of it. You can make LLMs output non slop versions. I can send you my prompt, for example and that worked fairly well.
Also, what is a big deal is the mismatch between weights and descriptions. That is always tricky when dealing with LLM. The only thing that works is the LLM having a list of measurements / weights and descriptions.
Graphics missing are a bit annoying.
Anyway.
The concept is good, the execution is quite good except for the slop tier text , no graphics - not even heroines and so on.