CCD JPEG Filter

# Wiggly CCD JPEG Filter Format

This runnable Repo preserves the exact prompt and five transformation examples
gathered from @skaigenerated. It keeps the source scene intact while changing
only the photographic rendering into a noisy, compressed late-2000s CCD JPEG
through Replicate.

Run from `v3`:

```bash
npm install
npm run format:skai-image -- smoke --format=ccd-jpeg-filter
npm run format:skai-image -- init --format=ccd-jpeg-filter --run=my-ccd-jpeg --input=/absolute/path/to/photo.jpg
npm run format:skai-image -- validate --format=ccd-jpeg-filter --run=my-ccd-jpeg
npm run format:skai-image -- estimate --format=ccd-jpeg-filter --run=my-ccd-jpeg
```

Generation requires `REPLICATE_API_TOKEN` and explicit approval:

```bash
npm run format:skai-image -- render --format=ccd-jpeg-filter --run=my-ccd-jpeg --approve-paid
```

View the output, record visual notes with
`inspect --visual-pass --review-notes="..."`, then use
`finalize --approve-final`.

Model routes:

- Economy: `google/nano-banana-2-lite`
- Default: `google/nano-banana-2`
- Premium alternative: `google/nano-banana-pro`
- Source model: `openai/gpt-image-2` at its supported 2:3 portrait ratio

No Replicate proof call was made while packaging this Repo. The five included
source examples and guide are the proof artifacts.

---
name: wiggly-ccd-jpeg-filter
description: Give one existing photo an authentic noisy, compressed late-2000s consumer-camera rendering while preserving its composition, subject, lighting, and colors using the exact gathered SKAI prompt.
---

# CCD JPEG Filter

Use the packaged runner and prompt. Do not call Replicate separately or create
another render path.

Ask one short question first:

> Which photo should I give the CCD JPEG look?

Use Nano Banana 2 by default, Nano Banana 2 Lite for economy, or Nano Banana
Pro for the premium route. Use GPT Image 2 for the source-model route.

```bash
npm run format:skai-image -- check --format=ccd-jpeg-filter
npm run format:skai-image -- init --format=ccd-jpeg-filter --run=<id> --input=<absolute-image-path>
npm run format:skai-image -- validate --format=ccd-jpeg-filter --run=<id>
npm run format:skai-image -- estimate --format=ccd-jpeg-filter --run=<id>
npm run format:skai-image -- render --format=ccd-jpeg-filter --run=<id> --approve-paid
npm run format:skai-image -- inspect --format=ccd-jpeg-filter --run=<id>
npm run format:skai-image -- inspect --format=ccd-jpeg-filter --run=<id> --visual-pass --review-notes="<specific observations>"
npm run format:skai-image -- finalize --format=ccd-jpeg-filter --run=<id> --approve-final
```

Hard rules:

- Preserve the original composition, framing, perspective, lighting, subject, and colors.
- Change only the photographic rendering; do not add, remove, or restage scene content.
- Require dense electronic sensor noise, realistic JPEG artifacts, reduced micro-detail, weak sharpening, and low dynamic range.
- Keep the result accidental and consumer-digital, never cinematic film grain or an obvious retro overlay.
- Validate before spending.
- Never generate without `--approve-paid`.
- Persist and resume the same prediction ID.
- Never exceed three attempts.
- View the actual output and record notes before `--approve-final`.
- Never print, store, or commit `REPLICATE_API_TOKEN`.

{
  "runtime": {
    "node": ">=22",
    "commands": ["npm"]
  },
  "providers": [
    {
      "name": "Replicate",
      "models": [
        "google/nano-banana-2-lite",
        "google/nano-banana-2",
        "google/nano-banana-pro",
        "openai/gpt-image-2"
      ]
    }
  ],
  "environment": [
    {
      "name": "REPLICATE_API_TOKEN",
      "secret": true,
      "requiredFor": ["render", "resume"]
    }
  ],
  "notes": [
    "All smoke and validation commands are local.",
    "No Replicate proof generation was purchased for this package.",
    "The gathered source examples are the visual proof.",
    "Paid generation requires explicit approval and is capped at three attempts."
  ]
}

{
  "required": [
    {
      "id": "photo",
      "type": "image",
      "formats": ["jpg", "png", "webp"],
      "minimumWidth": 512,
      "minimumHeight": 512,
      "maximumBytes": 26214400
    }
  ],
  "question": "Which photo should I give the CCD JPEG look?",
  "defaults": {
    "model": "nano-banana-2",
    "aspectRatio": "3:4",
    "outputCount": 1
  },
  "modelRoutes": {
    "economy": "google/nano-banana-2-lite",
    "default": "google/nano-banana-2",
    "premium": "google/nano-banana-pro",
    "sourceModel": "openai/gpt-image-2"
  }
}

{
  "progress": "Prepare -> Validate -> Transform -> Inspect",
  "attemptCap": 3,
  "stages": [
    {
      "id": "prepare",
      "output": "A run pinned to the source photo, exact gathered prompt, and chosen model route.",
      "paid": false,
      "approvalRequired": false
    },
    {
      "id": "validate",
      "output": "The image and complete provider contract checked locally.",
      "paid": false,
      "approvalRequired": false
    },
    {
      "id": "transform",
      "output": "One scene-preserving CCD JPEG-style image from the selected Replicate model.",
      "paid": true,
      "approvalRequired": true
    },
    {
      "id": "inspect",
      "output": "Automatic file checks plus an actual visual review.",
      "paid": false,
      "approvalRequired": true
    }
  ]
}

{
  "sourceReference": {
    "creator": "@skaigenerated",
    "postUrl": "https://www.instagram.com/p/DbQ10eICvPo/",
    "resourceUrl": "https://docs.google.com/document/d/1VK39fnwcHtReiXfzf7lPN0OQLMqiW98Zz1JlTkdv66I/edit",
    "guide": "assets/source/guide.pdf",
    "exampleOutput": "assets/source/example-output.jpg",
    "hero": "assets/source/hero-with-reference.jpg",
    "referenceInput": "assets/source/reference-input.jpg",
    "carouselExamples": [
      "assets/source/example-output.jpg",
      "assets/source/example-02.jpg",
      "assets/source/example-03.jpg",
      "assets/source/example-04.jpg",
      "assets/source/example-05.jpg"
    ],
    "promptPath": "prompts/transform.txt",
    "verification": "The exact prompt, negative prompt, GPT Image 2 label, guide, five examples, and matching before insets were gathered through the creator's Android Instagram comment and DM flow and visually checked. The reference fixture is cropped directly from the creator's before inset for the matching lakeside example; no provider generation was run."
  },
  "fixture": {
    "path": "assets/source/reference-input.jpg",
    "purpose": "Creator-provided lakeside before inset used for local validation, hero comparison, and free smoke input only; no provider call was made."
  }
}

{
  "automatic": [
    "Exactly 1 decodable JPG, PNG, or WebP output",
    "At least 640x800 pixels",
    "Aspect ratio within 0.02 of the selected route",
    "At least 50 KB"
  ],
  "visual": [
    "The original composition, framing, perspective, lighting, subject, and colors remain recognizable",
    "Dense fine electronic noise covers smooth surfaces and shadows without reading as cinematic film grain",
    "JPEG blockiness, smearing, reduced micro-detail, and weak sharpening feel naturally accumulated",
    "Crushed noisy shadows, softly clipped highlights, and imperfect automatic exposure lower the dynamic range",
    "The result feels like a cheap late-2000s camera file repeatedly uploaded and reshared, not a clean modern image with a retro overlay"
  ],
  "finalization": {
    "automaticPassRequired": true,
    "visualPassRequired": true,
    "reviewNotesRequired": true,
    "explicitApprovalFlag": "--approve-final"
  }
}

{
  "purpose": "Judge scene preservation, dense electronic noise, realistic JPEG compression, soft consumer optics, imperfect exposure, and early-social-media authenticity against five gathered examples.",
  "examples": [
    {
      "id": "lakeside-speedboat",
      "title": "A clean lakeside snapshot becomes a compressed archive find",
      "imagePath": "assets/source/hero-with-reference.jpg",
      "model": "GPT Image 2",
      "whyItWorks": [
        "The person, boat interior, lake, mountain, and background speedboat remain in the same composition",
        "Dense sensor noise, cyan contamination, crushed shadows, and soft detail feel native to an old CCD camera"
      ]
    },
    {
      "id": "red-car-night",
      "title": "Night flash keeps the pose and car while degrading the file",
      "imagePath": "assets/source/example-02.jpg",
      "model": "GPT Image 2",
      "whyItWorks": [
        "The subject, black outfit, red car, and parking-lot framing remain unchanged",
        "Highlight clipping, hard flash, chroma noise, and JPEG smearing create believable cheap-camera rendering"
      ]
    },
    {
      "id": "phone-booth",
      "title": "A phone-booth portrait gets true early-social compression",
      "imagePath": "assets/source/example-03.jpg",
      "model": "GPT Image 2",
      "whyItWorks": [
        "The phone, suit, booth, graffiti, and original framing remain recognizable",
        "Texture smearing and noisy shadows degrade the image without turning into film grain"
      ]
    },
    {
      "id": "parking-lot-lighter",
      "title": "A simple flash portrait feels downloaded and reshared",
      "imagePath": "assets/source/example-04.jpg",
      "model": "GPT Image 2",
      "whyItWorks": [
        "The lighter, jacket, jewelry, subject, and open parking lot stay fixed",
        "Underexposed midtones, electronic grain, and weak sharpening feel accidental rather than stylized"
      ]
    },
    {
      "id": "airport-flowers",
      "title": "A daylight airport photo still reads through heavy compression",
      "imagePath": "assets/source/example-05.jpg",
      "model": "GPT Image 2",
      "whyItWorks": [
        "The flowers, cap, clothing, airplane, and tarmac framing remain intact",
        "Compression artifacts, soft optics, and noisy low-range exposure create an archived-internet finish"
      ]
    }
  ]
}

{
  "id": "ccd-jpeg-filter",
  "version": "1.0.0",
  "title": "CCD JPEG Filter",
  "description": "Turn one existing photo into an authentically noisy, compressed late-2000s consumer-camera image while preserving the original scene.",
  "status": "agent-ready",
  "owner": "Wiggly Studio",
  "source": {
    "creator": "@skaigenerated",
    "postUrl": "https://www.instagram.com/p/DbQ10eICvPo/",
    "resourceUrl": "https://docs.google.com/document/d/1VK39fnwcHtReiXfzf7lPN0OQLMqiW98Zz1JlTkdv66I/edit",
    "ctaKeyword": "AI",
    "modelShown": "GPT Image 2"
  },
  "outputs": [
    "1 authentic CCD JPEG-style image",
    "state.json with model and prediction provenance",
    "quality-report.json with automatic and visual inspection"
  ]
}

{
  "name": "wiggly-ccd-jpeg-filter-format-kit",
  "private": true,
  "version": "1.0.0",
  "type": "module",
  "engines": {
    "node": ">=22"
  },
  "scripts": {
    "check:kit": "node kit-smoke.mjs",
    "format:skai-image": "tsx scripts/skai-image-format.ts",
    "test": "tsx tests/skai-image-format-runner.test.ts --format=ccd-jpeg-filter && node kit-smoke.mjs"
  },
  "dependencies": {
    "image-size": "2.0.2",
    "replicate": "1.4.0",
    "tsx": "4.20.6"
  }
}

{
  "slug": "ccd-jpeg-filter",
  "version": "1.0.0",
  "promptPath": "prompts/transform.txt",
  "input": {
    "mode": "image",
    "formats": ["jpg", "png", "webp"],
    "minimumWidth": 512,
    "minimumHeight": 512,
    "maximumBytes": 26214400
  },
  "defaultModel": "nano-banana-2",
  "modelRoutes": {
    "nano-banana-2-lite": {
      "label": "Nano Banana 2 Lite",
      "lane": "economy",
      "model": "google/nano-banana-2-lite",
      "family": "nano-banana",
      "aspectRatio": "3:4",
      "outputFormat": "jpg",
      "costEstimate": "about $0.034 per output image at the current listed rate",
      "timeEstimate": "usually 5-60 seconds"
    },
    "nano-banana-2": {
      "label": "Nano Banana 2",
      "lane": "default",
      "model": "google/nano-banana-2",
      "family": "nano-banana",
      "aspectRatio": "3:4",
      "outputFormat": "jpg",
      "resolution": "1K",
      "costEstimate": "about $0.067 per 1K output image at the current listed rate",
      "timeEstimate": "usually 10-120 seconds"
    },
    "nano-banana-pro": {
      "label": "Nano Banana Pro",
      "lane": "premium",
      "model": "google/nano-banana-pro",
      "family": "nano-banana",
      "aspectRatio": "3:4",
      "outputFormat": "jpg",
      "resolution": "1K",
      "costEstimate": "about $0.15 per 1K output image at the current listed rate",
      "timeEstimate": "usually 30-180 seconds"
    },
    "gpt-image-2": {
      "label": "GPT Image 2",
      "lane": "source model",
      "model": "openai/gpt-image-2",
      "family": "gpt-image",
      "aspectRatio": "2:3",
      "outputFormat": "jpeg",
      "quality": "medium",
      "costEstimate": "current Replicate GPT Image 2 rate",
      "timeEstimate": "usually 20-120 seconds"
    }
  },
  "expectedOutputs": 1,
  "maximumAttempts": 3,
  "smokeInputPath": "assets/source/reference-input.jpg",
  "smokeExamplePath": "assets/source/example-output.jpg",
  "minimumOutputWidth": 640,
  "minimumOutputHeight": 800,
  "minimumOutputBytes": 51200,
  "manualReview": [
    "preserved composition, subject, lighting, and colors",
    "dense fine electronic sensor noise rather than cinematic film grain",
    "realistic JPEG artifacts and reduced micro-detail",
    "low dynamic range with noisy shadows and soft highlight clipping",
    "authentic accidental late-2000s consumer-camera rendering"
  ]
}