🛡️ Overseer AI Moderation Engine

Status: Model Loaded & Online

A private, serverless computer vision API powered by nsfwjs and TensorFlow.js.

\n\n

Quick Start

POST /

Send the raw image binary (ArrayBuffer) as the request body.

\nfetch(\"${url.origin}\", {\n  method: \"POST\",\n  headers: { \"Content-Type\": \"application/octet-stream\" },\n  body: imageBinaryData\n})

\n\n

Response Format

Returns a JSON array of classifications with probability scores (0.0 to 1.0).

\n[\n  { \"className\": \"Neutral\", \"probability\": 0.92 },\n  { \"className\": \"Drawing\", \"probability\": 0.05 },\n  { \"className\": \"Sexy\", \"probability\": 0.02 },\n  { \"className\": \"Porn\", \"probability\": 0.01 },\n  { \"className\": \"Hentai\", \"probability\": 0.00 }\n]

\n\n

Legacy Support

GET /?img=[URL]

Scans an image via a publicly accessible URL (Note: May fail on some CORS-restricted domains).

\n \n \n `;\n return new Response(docsHtml, {\n headers: { \"Content-Type\": \"text/html\" },\n });\n }\n\n // ==========================================\n // 3. CORE PROCESSING LOGIC\n // ==========================================\n let arrayBuffer: ArrayBuffer;\n\n if (req.method === \"POST\") {\n arrayBuffer = await req.arrayBuffer();\n } else if (targetImageUrl) {\n const res = await fetch(targetImageUrl);\n arrayBuffer = await res.arrayBuffer();\n } else {\n throw new Error(\"Invalid request state.\");\n }\n\n if (!arrayBuffer || arrayBuffer.byteLength === 0) {\n throw new Error(\"Empty image payload.\");\n }\n\n const image = await Jimp.read(Buffer.from(arrayBuffer));\n const { width, height, data } = image.bitmap;\n\n const numChannels = 3;\n const numPixels = width * height;\n const values = new Int32Array(numPixels * numChannels);\n\n for (let i = 0; i < numPixels; i++) {\n values[i * numChannels + 0] = data[i * 4 + 0];\n values[i * numChannels + 1] = data[i * 4 + 1];\n values[i * numChannels + 2] = data[i * 4 + 2];\n }\n\n const imageTensor = tf.tensor3d(\n values,\n [height, width, numChannels],\n \"int32\",\n );\n const predictions = await model.classify(imageTensor);\n imageTensor.dispose();\n\n return Response.json(predictions, {\n headers: { \"Access-Control-Allow-Origin\": \"*\" },\n });\n } catch (error) {\n return Response.json({ error: error.message }, {\n status: 500,\n headers: { \"Access-Control-Allow-Origin\": \"*\" },\n });\n }\n}","deleted","label","imgcheck",[102],"permalink","https://www.val.town/x/metamysteries8/antinsfw/code/main.ts","extension","actionData","errors"]