Stylized Illustration: witch brewing potions in a moonlit cottage as manga black and white with screentones on Playground v3

You are generating a stylized illustration prompt (not photoreal) for Playground v3. The target is artistic, not documentary.

## Brief
- **Subject:** witch brewing potions in a moonlit cottage
- **Target illustration style:** manga black and white with screentones
- **Lighting:** moody single-source chiaroscuro
- **Composition:** overhead flat-lay
- **Aspect:** --ar 4:5

## Stylized Illustration — Why It's Different
Photoreal prompts describe a camera. Illustrated prompts describe **an artist, a medium, and a process**:

- **Artist archetype** (NOT real names on major models — they filter). Instead: "a Studio Ghibli-trained watercolor artist", "an art nouveau illustrator in the vein of Alphonse Mucha".
- **Medium** — oil on linen, gouache on cold-press paper, digital cel-shaded in Clip Studio, ink + screentones on bristol, 3D clay-rendered in Blender.
- **Brush/tool specifics** — "visible brushstrokes", "thin line work with 2-pixel pen", "rough charcoal shading".
- **Surface texture** — canvas weave, paper tooth, pixel grid, cel animation fade.
- **Color treatment** — limited palette (name 3-5 colors), halftone dots, Risograph 2-color print feel.

## Style-to-Model Compatibility

- **niji v6 (Midjourney):** Anime / manga / shoujo. Best-in-class for anime.
- **Midjourney v6.1 / v7:** Excellent for painterly, editorial, and children's-book illustration.
- **Flux.1 [dev]:** Surprisingly strong on watercolor and line-and-wash.
- **SDXL + LoRAs:** Near-unlimited style via community LoRAs (Disney pixar, pixel art, etc).
- **Recraft V3:** Vector illustration, flat design, iconography — unmatched.
- **DALL-E 3:** Cute/kawaii and children's-book styles are its sweet spot.
- **Ideogram:** Strong flat graphic illustration with text.
- **Imagen 3:** Photoreal-leaning; stylization is possible but less expressive than Midjourney.

## Output

**A. Primary prompt** (60-120 words):
```
[Prompt in the artist + medium + process format]
```

**B. Suggested LoRA / style references** (if applicable on SDXL / SD 3.5):
- 2 LoRA names from CivitAI that fit manga black and white with screentones
- Trigger words for each

**C. Parameters:**
- Midjourney: `--ar 4:5 --stylize 500` for expressive; `--stylize 100` for restrained
- For anime: `--niji 6 --style expressive`
- Flux: [dev], guidance 3.5
- SDXL: CFG 6-8, DPM++ 2M Karras, LoRA weight 0.7-0.9

**D. Three stylistic variations:**
1. Palette shift — same subject, different 3-color palette
2. Medium shift — same subject, different illustration medium
3. Composition shift — same subject, different overhead flat-lay

**E. Gotchas:**
- Hands and fingers fail more on stylized prompts than photoreal — specify pose explicitly.
- Text rendering on stylized prompts is worse than on clean styles — avoid text unless Playground v3 is Ideogram or Flux.
- Multi-character illustrations lose coherence — stick to 1-2 subjects.
- Avoid "beautiful anime girl" type prompts — describes nothing; gives generic output.

## Strict Rules
- No real artist names on commercial APIs.
- No brand names (Disney, Ghibli, Pixar) — describe the aesthetic without naming.
- Match manga black and white with screentones to a model that can execute it; don't ask SDXL base to do Ghibli, use a Ghibli LoRA.
- Keep the prompt focused — manga black and white with screentones is the anchor; other adjectives serve it, not compete with it.

Generate the full output.