Niu Lai Translator / 牛来低模转译器

Core principle: 保留大关系，主动放弃精致还原；让低质量来自整套生产能力，而不是后期滤镜。

Rebuild the source as a scene made by a sincere but technically limited early-3D team. Keep the image readable and structurally related to the source while making modeling, eyes, textures, asset variety, lighting, and rendering visibly crude.

Workflow

Track this sequence:

- [ ] 1. Confirm that an input image exists; request one if absent
- [ ] 2. Read explicit parameters and infer the rest from the primitive defaults
- [ ] 3. Extract broad anchors: subject count, layout, pose, camera, color blocks
- [ ] 4. Set a deliberately low detail and fidelity budget
- [ ] 5. Build one source-based image-edit prompt in the required order
- [ ] 6. Generate/edit unless the user asked for prompts or options only
- [ ] 7. Inspect against the inverted quality gate; retry polished results
- [ ] 8. Return the image plus one compact treatment note

Interaction

• Require an input image unless the user explicitly asks for a new scene.
• Execute directly when an image is supplied without parameters or the user says “默认”, “你来判断”, or equivalent.
• If the user says “先别生成”, “给方案”, or “让我选”, do not generate. Offer at most three distinct directions.
• Ask at most one question only when a missing decision materially changes output.
• Use an image-edit tool that receives the source image. Include every target image through the tool's supported reference input.
• Treat this as reconstruction, not compression, pixelation, or a polygon overlay.

Analyze broad anchors internally

Identify without reporting every item unless asked:

• Subjects: count, broad type, silhouette, large hairstyle, garment category, dominant color blocks, large props, pose, gesture, and approximate gaze.
• Composition: crop, camera height, perspective, horizon, relative scale, spacing, foreground/midground/background, and occlusion.
• Scene: only the major terrain, architecture, or stage masses required for the image to remain recognizable.
• Palette: three to six source colors worth retaining.
• Discardable detail: fine facial likeness, subtle expression, hair strands, embroidery, microtexture, unique background assets, small signage, and decoration.

Use primitive defaults

Use primitive_folk_cgi unless the user requests a gentler translation. Read references/style-system.md when selecting another preset.

preset: primitive_folk_cgi
reconstruction_strength: extreme
anchor_lock: strict
identity_lock: medium
detail_budget: very_low
geometry: primitive_low_poly
face_geometry: clumsy_asymmetric
gaze_quality: stiff_misaligned
texture_resolution: very_low
texture_repetition: obvious
asset_reuse: heavy
background_detail: sparse
lighting: naive_single_light
post_effects: subtle_capture_noise
text_mode: none
ratio: source_ratio

These are intentional defaults. High facial fidelity, rich backgrounds, unique textures, attractive eyes, or professional lighting are failures unless requested.

Preserve anchors, not polish

• Strictly preserve subject count, broad left-to-right arrangement, major pose, gesture, crop, camera, occlusion, scene category, and dominant color blocks.
• Preserve large identity anchors such as species, hair mass, clothing category, and major props. Do not preserve every facial detail by default.
• Permit proportions, facial topology, gaze alignment, garment detail, and surface accuracy to become visibly crude while the scene remains legible.
• Raise identity_lock to high only when the user requests likeness or the task is identity-sensitive; reduce reconstruction strength if necessary.
• Do not change a real person's apparent age, ethnicity, body category, pose, or relationship to others merely to create the style.
• Add no people, animals, horns, cow traits, weapons, props, logos, subtitles, interface, meme text, landmarks, or weather not evidenced by the source.

Rebuild all production layers

Apply all layers together:

1. Geometry: use few, large, awkward planes; hard normals; lumpy silhouettes; blocky joints; wedge hands; crude facial wedges; no elegant tessellation.
2. Faces and gaze: reuse simple eye construction; allow unequal eye size, slightly off-center pupils, imperfect gaze alignment, stiff lids, flat mouth slits, and frozen expressions. Keep intent readable, not professionally acted.
3. Materials: use tiny blurry diffuse maps, seams, UV stretching, mirrored details, inconsistent scale, and crude painted shadow information.
4. Repetition: visibly tile textures and reuse a very small library of trees, posts, rocks, buildings, props, hair cards, cloth maps, or eye assets.
5. Background: remove incidental detail. Repeat a few primitive assets with minimal variation; avoid rich set dressing or individually authored objects.
6. Lighting: use one blunt directional/overhead-front light plus weak flat ambient fill. Allow clipped highlights, muddy dark patches, hard low-resolution shadows, weak contact, and exposure mismatch between nearby subjects.
7. Rendering: use weak antialiasing, limited texture filtering, simple shadow maps, fine noise, mild color fringe, and modest screen-capture softness.

Do not use darkness, horror grading, VHS damage, heavy JPEG corruption, or mosaic as shortcuts. “Bad” must come primarily from limited scene production.

Build the edit prompt

Read references/prompt-blueprint.md for the full schema and reusable clauses. Construct prompts in this order:

1. Declare source-based primitive 3D reconstruction.
2. Lock broad anchors and explicitly release fine-detail fidelity.
3. Set primitive preset, extreme strength, and very-low detail budget.
4. Specify crude geometry, faces, eyes, hair, hands, and proportions.
5. Require very-low-resolution maps, obvious tiling, and heavy asset reuse.
6. Strip background detail to repeated scene primitives.
7. Specify naive lighting, uneven exposure, and basic render limitations.
8. State ratio, text behavior, and source-tailored prohibitions.

Prefer observable flaws over labels such as “ugly” or “bad quality”. Mention a named game, engine, or film only when the user asks; always translate the name into construction, material, lighting, and rendering properties.

Inspect and recover

Read references/quality-and-recovery.md before judging output. Retry whenever the result is too faithful, attractive, detailed, varied, or professionally lit. In particular, reject results where:

• faces retain polished animation acting or appealing aligned eyes;
• textures are clean, unique, high-resolution, or subtly varied;
• background objects are numerous, detailed, and individually modeled;
• lighting uses cinematic separation, rim light, soft bounce, or volumetric depth;
• geometry resembles fashionable designer low-poly art;
• the effect is only blur, pixelation, noise, or color grading.

On retry, change only the failed dimension and restate all anchor locks. Stop after two retries unless the user asks to continue.

Output

For a completed edit, return the generated image and one sentence naming the preset plus the main retained anchors. Include YAML only when requested.

For prompt-only requests, return one production prompt and one negative block.

Minimal invocation

/niu-lai-translator
启用牛来低模转译器
把这张图重制成制作能力很差的早期 3D 动画截图

skill: niu-lai-translator
preset: primitive_folk_cgi
reconstruction_strength: extreme
identity_lock: medium
asset_reuse: heavy
lighting: naive_single_light
ratio: source_ratio

不是做得像“低模艺术”，而是像真的只有少量模型、少量贴图、简单灯光和有限技术。