anime_porn_image_prompts

Repository: MingSafeR/anime_porn_image_prompts
Author / Contact: MingSafeR — xinhao601@gmail.com
License: Apache License 2.0 (Apache-2.0)
Content Type: Adult / NSFW generation prompts (text only). This repository contains prompts designed to synthesize adult (explicit) anime/photo images. It does not include images.

Critical warning: This repository contains prompts intended to generate adult/explicit imagery (NSFW). Do not use these prompts to generate content involving minors, non-consensual acts, sexual violence, or any illegal or unethical content. Users are responsible for all legal, ethical, and platform compliance.

---

1. Overview (Read before use)

• The repository’s primary files must follow a strict, uniform JSON format to allow pipeline ingestion, validation, and training.
• Each record in the primary prompt file MUST contain exactly three fields (strings): prompt, negative_prompt, and style. No extra fields in the main prompt file are permitted.
• style only accepts the values: "anime" or "photo". This controls recommended quality tokens and negative prompt adjustments.
• If you need to preserve extra metadata (seed, id, generator version, notes), store those in a separate sidecar file such as prompts_meta.json or inside a metadata/ directory — do not add extra keys to the main prompts file.

---

2. Required format (mandatory)

Every prompt record must be one JSON object with exactly these three keys:

{"prompt": "<positive prompt string>", "negative_prompt": "<negative prompt string>", "style": "<anime|photo>"}

• prompt — non-empty string. Contains the positive generation prompt (description, actions, appearance, scene, quality tails, etc.).
• negative_prompt — string (may be empty). Contains negative tokens to suppress unwanted features (watermarks, text, low quality, bad anatomy, etc.).
• style — string, must be either "anime" or "photo".

Examples

JSONL (single-line example):

{"prompt":"consenting adult subject, lingerie, close-up, soft lighting, high detail, anime style, explicit genitals visible","negative_prompt":"worst quality, low quality, watermark, signature, text, bad anatomy","style":"anime"}

JSON (array example):

[
  {"prompt":"...", "negative_prompt":"...", "style":"anime"},
  {"prompt":"...", "negative_prompt":"...", "style":"photo"}
]

---

3. Filenames & recommended conventions

• Recommended naming pattern: prompts_<topic>_<YYYYMMDD>.jsonl or prompts_<topic>_<YYYYMMDD>.json.
• For partitioned uploads: prompts_<topic>_part01.jsonl, prompts_<topic>_part02.jsonl, etc.
• Use UTF-8 encoding for all files.

---

4. Prompt composition guidance (recommended, not optional)

While the prompt free-form text is flexible, follow this constructive structure to improve clarity and downstream handling:

1. Consent and adult marker (recommended): include consenting adult subject or similar to signal consent.
2. Subject attributes: age descriptor limited to adults (e.g., 25+), body type, ethnicity (if relevant), breast/butt descriptors (use responsibly).
3. Facial expression / emotion: e.g., ecstatic expression, smiling.
4. Action / pose: prefer canonical terms (oral sex, vaginal penetration, cowgirl position, masturbation). Use controlled vocabulary where possible.
5. Clothing / outfit: lingerie, bikini, maid, kimono — be mindful of cultural sensitivity.
6. Scene / camera / lighting: close-up, soft cinematic rim light, 85mm f/1.4, bokeh background.
7. Quality tails: include quality tokens depending on style (see Section 6).
8. Avoid placing watermark/text suppression in prompt; use negative_prompt for that.

Example anime template (for reference only):

consenting adult subject, 25+, slim, long silver hair, close-up, lingerie, ahegao expression, explicit genitals visible, front view, close-up, masterpiece, best quality, ultra-detailed, clean lineart, no watermark, no text

Example photo template:

consenting adult subject, 30+, curvy, brunette, on bed, soft cinematic rim light, 85mm f/1.4, shallow depth of field, photorealistic, ultra-detailed skin, highres, no watermark, no text

---

5. Negative prompt recommendations

Use negative_prompt to suppress unwanted artifacts or styles. Core negative tokens:

worst quality, low quality, lowres, blurry, out of focus, motion blur, camera shake, bad composition,
jpeg artifacts, compression artifacts, watermark, signature, username, logo, text, caption,
bad anatomy, bad hands, extra digits, pixelated, censor

Additional negatives for anime:

bad shading, messy lineart, speech bubble, duplicate limbs

Additional negatives for photo:

white balance issue, over-sharpened, harsh shadow, flat lighting

---

6. Quality tails (recommended tokens)

Pick quality tails according to style.

Anime quality tokens:

masterpiece, best quality, ultra-detailed, highres, finely detailed eyes, 4k, clean lineart

Photo quality tokens:

photorealistic, ultra-detailed skin, highres, RAW, shallow depth of field, 85mm f/1.4

---

7. Standardization and normalization

• Canonicalize common synonyms (examples): blowjob → oral sex; pussy → vagina; cock → penis; cumshot → facial cumshot.
• Normalize whitespace and punctuation; lowercase for dedupe/uniqueness checks.
• De-duplicate prompts using a normalized key (e.g., SHA1 of normalized prompt, truncated) before publishing.

---

8. Validation rules (machine-enforced)

Validation must check:

1. JSON validity.

2. Each record ONLY contains keys prompt, negative_prompt, style. No extra keys.

3. All three fields are strings; prompt must be non-empty.

4. style must be one of anime or photo.

5. Block/flag if prompt contains disallowed categories:

   • Minor-related terms: e.g., underage, minor, child, teen, preteen, schoolgirl, loli, shota.
   • Non-consensual / coercive terms: e.g., rape, forced, non-consensual, drugged, abduct.
   • Real-person name patterns (optional strict mode): sequences like First Last or First Middle Last with capitalization indicating a real name — flag for manual review or reject if strict.

6. If any blocked keywords appear, fail validation and require manual correction.

---

9. Sidecar metadata (recommended)

If you require provenance fields, keep them in a separate file prompts_meta.json or metadata/<file>.json. Example entry:

{
  "id": "prompt_000001",
  "seed": 123456789,
  "generated_with": "tupanpromptscq v1.0",
  "created_at": "2025-09-16T08:00:00Z",
  "notes": "force_strong=true, dedupe=true"
}

Sidecar metadata preserves reproducibility and audit trails without breaking pipeline compatibility.

---

10. Quality control & sampling strategy (recommended)

• Initial acceptance sampling: randomly inspect ≥ 0.5% of the dataset, minimum 100 items, before publishing.
• Boundary classes (e.g., swimwear, lingerie, adult products): sample rate 2%–5% for closer scrutiny.
• Ongoing monitoring: sample 0.1%–0.5% of new uploads for regression/quality.
• Automated checks: length anomalies, duplicate rate, sexual token density; high-scoring anomalies flagged for manual review.
• Audit logging: store generation metadata and review outcomes for closed-loop improvements.

---

11. Change log (recommended structure)

Maintain a CHANGELOG.md or append version history here. Example:

v1.0.0 - 2025-09-16
- Initial release: three-field format, negative list, quality tails, validation rules.

v1.1.0 - 2025-10-01
- Added strict real-name detection option and expanded negative tokens.

---

12. Multi-line examples (several records)

{"prompt":"consenting adult subject, 25+, slim, long silver hair, lingerie, ahegao expression, explicit genitals visible, front view, close-up, masterpiece, best quality, ultra-detailed, clean lineart, no watermark, no text","negative_prompt":"worst quality, low quality, watermark, signature, text, bad anatomy, extra digits","style":"anime"}
{"prompt":"consenting adult subject, 30+, curvy, brunette, on bed, soft cinematic rim light, photorealistic, highres, shallow depth of field, 85mm f/1.4, no watermark, no text","negative_prompt":"worst quality, low quality, jpeg artifacts, watermark, logo, bad hands","style":"photo"}

---

13. Compliance & risk statement (practical)

• Absolutely do not include or imply minors. Automated regex checks exist, but they are not perfect — manual review is mandatory.
• Absolutely do not generate or publish non-consensual sexual content. The prompt generator and validators include keyword blocking for this, but human oversight is required.
• Real-person prompts: avoid generating or publishing content referencing private individuals or non-consenting persons. If a real-person workflow is intended, require documented consent and legal review.
• Audit trail: retain seed, prompt id, generator version, and timestamps for each generation to enable post-hoc accountability.
• This README is not legal advice. For commercial or public distribution, consult legal counsel.

---

14. Citation (for research / model cards)

If you use these prompts in research, evaluation, or a model card, please include:

Prompts from MingSafeR/anime_porn_image_prompts (© MingSafeR). Licensed Apache-2.0. Contact: xinhao601@gmail.com

---

15. Owner contact

Owner: MingSafeR Email: xinhao601@gmail.com

---

Quick reference cheat-sheet

• Core format: {"prompt":"...","negative_prompt":"...","style":"anime"}
• Allowed styles: anime, photo
• Core negative tokens: worst quality, low quality, lowres, blurry, watermark, signature, text, bad anatomy, extra digits
• Anime quality tail: masterpiece, best quality, ultra-detailed, highres, clean lineart
• Photo quality tail: photorealistic, ultra-detailed skin, highres, RAW, shallow depth of field
• Blocked categories: minors terms, non-consensual terms, real-person name patterns (strict mode)

---

End of README — this document is the authoritative, highly-detailed README for anime_porn_image_prompts. Follow the stated format, validation rules, and compliance guidance strictly.

::contentReference[oaicite:0]{index=0}