Anime styles are driven by five controllable cues: linework, shading, color, proportions, and texture. In AI image generation, you steer these with prompt descriptors (e.g., thin inked lines, hard cel shading), model/LoRA selection, and sampler/upscale settings. Use consistent phrasing and repeat key style descriptors to stabilize results across batches.

Use these as starting points; add subject, composition, and lighting details.

• Cel-shaded TV look: "clean thin lineart, hard cel shading, two-tone shadows, flat colors, minimal texture, crisp outlines"
• Shonen action: "dynamic pose, exaggerated perspective, speed lines, thick-to-thin lineweight, saturated colors, high contrast"
• Shojo romance: "elegant slim proportions, soft eyes, delicate eyelashes, pastel palette, gentle gradients, light bloom"
• Ghibli-inspired whimsy: "painterly backgrounds, warm natural palette, soft diffuse light, subtle linework, gentle texture"
• Mecha/tech: "precise panel lines, mechanical joints, glossy metal, rim light, blueprint accuracy, hard edges"
• 90s retro: "limited color, bold outlines, grainy cel, retro screentone, slight off-register look, VHS warmth"
• Seinen gritty: "muted colors, realistic anatomy, textured shading, film grain, subdued light, nuanced expressions"
• Chibi/super-deformed: "big head small body, thick outlines, simple shading, bright playful colors, minimal detail"

Character portrait: "[character concept], clean thin lineart, [substyle], [palette], [lighting], detailed eyes, smooth skin, uncluttered background"

Full-body character sheet: "turnaround sheet, front/side/back, neutral pose, consistent proportions, clean cel shading, white background, model sheet layout"

Action scene: "dynamic composition, [substyle], motion blur/speed lines, environmental FX (dust, sparks), strong contrast, cinematic framing"

Background plate: "[location], painterly foliage, simplified shapes, soft light, color harmony, gentle texture, anime background style"

Style lock (add to any): "cohesive style, consistent lineweight, two-tone shadows, no photorealism"

Models: Use anime-tuned checkpoints (e.g., SD 1.5 anime mixes or SDXL anime variants) for confident linework and cel shading. Add LoRAs for substyles (mecha panels, shojo eyes, retro halftone) and for clothing or hair specifics as needed.

Samplers: DPM++ 2M Karras or similar balanced samplers work well. Start 20–30 steps (1.5) or 25–35 (SDXL) and adjust by detail level.

CFG/Guidance: Moderate guidance usually preserves clean linework: 5–8 (1.5) or 3.5–6 (SDXL).

Resolution: Start 640–832 px short side for layout. Upscale later to 2–4×.

Clip/Color: If available, enable anime-friendly VAEs to avoid muddy colors. Clip-skip 2 (on 1.5-based stacks) can sharpen lines.

High-res pass: Use a small denoise (0.2–0.35) to keep line integrity while refining textures.

Reliable negatives reduce artifacts: "watermark, text, logo, signature, extra fingers, extra limbs, deformed hands, blur, lowres, jpeg artifacts, noisy lineart, messy shading".

Pitfalls:

• Over-detailed prompts muddy cel shading—favor fewer, stronger style cues.
• Excess denoise in upscaling melts linework—keep it low.
• Mixing multiple LoRAs at high weights causes style drift—limit to 1–2 core LoRAs.

Ratios: Use 3:4 or 4:5 for character showcase, 16:9 for establishing shots, and square for avatars. Keep headroom and limb crop points clean.

Color: For cel looks, restrict to a tight palette (base + shadow + accent). For painterly anime, favor harmonious triads and soft gradients.

Lines and tones: Specify thin vs. bold outlines and whether to use screentone/halftone. Repeat lineweight preferences in prompts and model notes.

1. Plan: Choose substyle and 3–5 visual anchors (linework, shading, palette, proportions, texture).
2. Draft: Low-res batch for framing and pose; pick top 1–2.
3. Detail: Re-run with refined prompt and substyle LoRA.
4. Upscale: 2–4× with low denoise; sharpen edges slightly.
5. Polish: Color balance, subtle film grain or paper texture; remove artifacts.
6. Consistency: Save prompts, seeds, and LoRA weights for series continuity.

Use this pass/fail list before publishing:

• Lines: consistent weight, no double edges
• Shading: clean cel cuts or intentional soft gradients
• Anatomy: correct hands, joints, and facial alignment
• Palette: cohesive and on-brief; no muddy mixes
• Background: matches character style; no unintended photoreal elements
• Artifacts: no watermarks, stray text, or compression noise