00Method, not mystery
A clearer way to learn visual style
AI Style Atlas is a controlled comparison. Every image begins with the same fictional person and the same reference photograph, so the differences you see come from the visual direction rather than a changing subject.
One identity, one repeatable process
We begin with a vertically framed base portrait, pair it with a complete image-to-image prompt, and generate one result for every entry. Keeping identity and framing consistent turns the catalogue into a practical visual reference instead of a mood board.
01 · Hold the subject
The face, hair, expression and core pose stay recognizable unless a style specifically requires a costume change.
02 · Change the language
Medium, line, color, surface, light and composition are described explicitly for each visual tradition.
03 · Compare and reuse
Move the comparison slider, inspect the result and copy either native-language prompt into your image tool.
Style-first principle
Each showcase image prioritizes the visual style over photographic fidelity. We retain recognizable cues such as face shape, long black hair and outfit motifs, while redrawing the entire person—including face, eyes and skin—in the target medium so a realistic face is never pasted onto a stylized body.
Image-to-image or text-to-image?
The prompts are written for an attached portrait. They can still guide text-only generation, but identity consistency will be lower without a reference image.
With a reference image
Attach a clear portrait, paste the prompt and keep the requested 2:3 framing. This is the closest match to the atlas workflow.
Without a reference image
Replace identity-preservation language with a precise subject description, then keep the style, lighting and output instructions.
Tool differences at a glance
| Tool | Best used for | Working note |
|---|---|---|
| ChatGPT | Conversational refinement | Ask for identity preservation again if the face drifts. |
| Gemini | Natural-language iteration | Give layout constraints in a short, explicit final sentence. |
| Midjourney | Strong aesthetic exploration | Use the listed parameters as a starting point, then tune stylization. |
| Stable Diffusion | Fine control and local workflows | Use the negative prompt and adjust image-to-image strength carefully. |
| Flux | Prompt fidelity and texture | Keep material and lighting terms concrete. |
Open prompts, fictional person
Every person shown is AI-generated and fictional. No real individual is represented or endorsed.
All AI Style Atlas prompts are released under CC0. Copy them, adapt them and use them in personal or commercial work.