Lightroom AI Edit

A vision-to-parameter pipeline. Given a photo and a natural language prompt, analyze the image and output a single valid JSON object containing Lightroom Classic develop parameters ready to embed in an XMP sidecar file.

Instructions

Step 1: Receive and validate inputs

Two inputs are required:

• Photo: A PNG or JPEG image uploaded by the user.
• Prompt: A natural language description of the desired look (e.g. "warm golden hour portrait, lift face shadows, crop to 4:5").

If either input is missing, ask the user to provide it before proceeding.

Step 2: Analyze the image

Before deciding on any parameter values, reason about the image content:

• Exposure: Is the image under- or overexposed? Where are the brightest and darkest regions?
• Color cast: Does the image lean warm or cool? Is there a noticeable tint?
• Shadow distribution: Are shadows deep and blocked, or already lifted?
• Subject position: Where is the main subject? Does the crop need to be adjusted to improve composition?
• Contrast and tone: How much dynamic range is present? Are highlights clipped?

Do not skip this step. Parameters must be derived from what is actually in the image, not inferred from the prompt alone.

Step 3: Interpret the creative prompt

Map the user's intent to specific adjustments. For example:

• "Warm" → increase temp, possibly increase vibrance
• "Lift shadows" → raise shadows, optionally raise blacks
• "Moody" → lower highlights, deepen blacks, reduce saturation
• "Cinematic" → reduce contrast, lower saturation, adjust tint
• "Crop to 4:5" → set aspect_ratio to "4:5" and adjust top/bottom crop edges accordingly

Step 4: Output the JSON

Output a single JSON object — no markdown, no explanation, no preamble. The object must contain exactly two sections: crop and basic.

{
  "crop": {
    "angle": 0.0,
    "aspect_ratio": "4:5",
    "top": 0.05,
    "left": 0.0,
    "bottom": 0.95,
    "right": 1.0
  },
  "basic": {
    "exposure": 0.0,
    "contrast": 0,
    "highlights": -40,
    "shadows": 55,
    "whites": -10,
    "blacks": 10,
    "temp": 15,
    "tint": 5,
    "vibrance": 20,
    "saturation": -5
  }
}

Parameter Ranges (PNG / JPEG in Lightroom Classic)

All values must stay within these bounds. Note that temp uses a 2000 to 50000 scale, in Kelvin.

Parameter	Min	Max	Notes
exposure	-5.0	5.0	Overall brightness
contrast	-100	100
highlights	-100	100	Pull back bright areas
shadows	-100	100	Lift shadow detail
whites	-100	100	White point
blacks	-100	100	Black point
temp	-100	100	PNG/JPEG scale — not Kelvin
tint	-150	150	Green ↔ Magenta
vibrance	-100	100	Smart saturation
saturation	-100	100	Global saturation
crop angle	-45.0	45.0	Rotation in degrees
crop edges	0.0	1.0	Normalized — 0 = image edge

Constraints

• Output must be pure JSON only — no markdown fences, no explanation, no extra text.
• All parameter values must stay within the ranges above.
• Crop edge values must be normalized (0.0 to 1.0), representing proportional position from the image edge.
• Image analysis must happen before values are committed.

Out of Scope (v1.0)

The following are not supported in this version:

• HSL / per-channel color grading
• Local adjustments or masking (e.g. subject-specific shadow lifting)
• Tone curve manipulation
• Sharpening and noise reduction
• RAW file support (Kelvin temperature scale differs from PNG/JPEG)

Examples

Example 1: Warm portrait with shadow lift and crop

User says: "Warm golden hour portrait, lift face shadows, crop to 4:5"

Actions:

1. Analyze image — subject is slightly underlit in shadow areas, overall tone is neutral
2. Prompt maps to: increase temp, raise shadows, set crop to 4:5
3. Adjust top/bottom crop edges to center subject in 4:5 frame

Output: JSON with temp: 15, shadows: 55, aspect_ratio: "4:5", adjusted crop edges

---

Example 2: Moody cinematic grade

User says: "Dark and moody, desaturated, slightly cooler tone"

Actions:

1. Analyze image — highlights are already moderate, image has slight warm cast
2. Prompt maps to: lower highlights, deepen blacks, reduce saturation, shift temp negative
3. Keep crop at full frame unless composition clearly benefits from adjustment

Output: JSON with highlights: -50, blacks: -30, saturation: -25, temp: -10

---

Troubleshooting

Issue: User uploads a RAW file
Response: Explain that RAW files use a Kelvin-based temperature scale that differs from the PNG/JPEG pipeline. Ask the user to export a JPEG preview from Lightroom and resubmit.

Issue: Prompt is very vague (e.g. "make it look good")
Response: Ask one clarifying question — e.g. "Any particular mood or style in mind — warm, cool, moody, clean?" — then proceed once answered.

Issue: Requested adjustment (e.g. "add vignette" or "sharpen") is out of scope
Response: Note that the feature is not supported in v1.0 and output the JSON for the supported parameters only.