Skip to content
PixelDepot
Get an API key

Curated open-source image AI

The best open-source image AI, ready to use.

Thousands of powerful AI image projects are being released across GitHub and the open-source ecosystem. Finding the right one, understanding its license, testing its quality and deploying it securely takes time.

PixelDepot does that work for you. We discover promising projects, verify their commercial-use rights, benchmark and improve them, then provide them through fast, secure and consistent APIs.

We find it. We test it. We improve it. We deploy it. You call one API.

Get a free API keyExplore APIs
  • License-reviewed for commercial use
  • Tested and benchmarked
  • Securely hosted
  • One consistent API
curl https://api.pixeldepot.com/v1/face/analyze \ -H "Authorization: Bearer $PIXELDEPOT_KEY" \ -H "Content-Type: application/json" \ -d '{"image_url": "https://.../guest.jpg", "features": ["geometry","apparent_gender","age"]}'

Detected

face_1 · 0.9997
placeholder image

1536 × 2048
184 ms · 1 face

{ "status": "succeeded", "faces": [{ "confidence": 0.9997, "bounding_box": { "x": 390, "y": 706, "width": 781, "height": 780 }, "eye_distance": 368, "pose": { "yaw": -3.0, "pitch": -17.0, "roll": -1.0 }, "quality": { "score": 0.79, "usable": true }, "attributes": { "age": { "estimate": 26, "confidence": 0.72 }, "apparent_gender": { "label": "female" } } }], "meta": { "processing_ms": 184, "models": [ … ] } }

The problem

Open source is moving faster than anyone can evaluate it.

New image AI projects appear every day. But stars and impressive demos do not tell you:

  • Whether the model is truly production-ready.
  • Whether its weights and training data allow commercial use.
  • Whether it performs reliably outside its demo images.
  • How to install, optimize, scale and maintain it.

PixelDepot handles all of this.

How PixelDepot works

From open-source project to production API.

  1. 01

    Discover

    We search GitHub and the wider open-source AI ecosystem for image models worth shipping.

  2. 02

    Verify

    We review the code, model weights, training data and dependency licenses.

  3. 03

    Evaluate

    We benchmark accuracy, speed, hardware requirements and real-world performance.

  4. 04

    Improve

    When necessary, we train, fine-tune, optimize or combine models.

  5. 05

    Deploy

    We securely host the selected models and expose them through one consistent API.

The catalog

One API for the bestimage AI services.

Start with Face Analyze, then expand into face detection, background removal, restoration, upscaling, image transformation and other carefully selected services.

  • /v1/face/analyzeIn training

    Face Analyze

    Boxes, landmarks, eye distance, head pose and a real quality score. Age, apparent gender, glasses and lips behind feature flags.

    PixelDepot Face v1 · trained in-house

  • /v1/face/detectIn training

    Face Detect

    Geometry only, no attributes, no retention. The cheapest way to count and crop faces in a crowd shot.

    PixelDepot Face v1 · trained in-house

  • /v1/image/matteOn the roadmap

    Background Removal

    Alpha matte and cutout for product and portrait images, tuned for hair and event lighting.

    Under license review

  • /v1/image/enhanceOn the roadmap

    Restore & Upscale

    Denoise, sharpen and 2–4× upscale for print output from booth-quality captures.

    Under license review

  • /v1/image/transformOn the roadmap

    Image to Image

    Style and scene transforms with a fixed prompt contract, so results stay predictable across a run.

    Under license review

  • /v1/face/matchOn the roadmap

    Face Recognition

    Galleries and 1:N matching — gated behind biometric-privacy and commercial-licensing sign-off.

    Pending biometric review

Nothing ships until a human has signed off on all four licenses: code, weights, training data and dependencies.

Ten minutes from signup to first response.

  1. 01

    Create a key

    Sign up, name a project, reveal the secret once. Live and test keys are separate and revocable.

    pd_live_7e77c8deb5fc4b958c0…

  2. 02

    Hand us an image

    Post a URL, or take a short-lived signed upload URL and send bytes straight to storage — never through our web tier.

    POST /v1/assets/upload-url

  3. 03

    Read the JSON

    One synchronous call. Stable error codes, the request ID in every log line, and the exact model digest that produced the result.

    POST /v1/face/analyze → 200

The four-gate license review

Vision APIs you're actually allowed to ship.

"MIT on the repo" is not a license to ship. Most image models on GitHub carry a permissive code license and non-commercial weights trained on scraped data. We treat missing, research-only and unclear as blocked — and we publish the paperwork for every model you call.

  1. 01Repository code license
  2. 02Pretrained weight license
  3. 03Training-dataset rights inherited by the weights
  4. 04Bundled dependency and component licenses

Stateless by default

Face Analyze never creates an enrollment record. Inputs are deleted after 24 hours, or immediately after inference if you ask.

Your images stay yours

No customer image is ever used for training without a separate, explicit consent agreement. EXIF is stripped on ingest.

Apparent, not actual

Ancestry and gender outputs are visual classifications with calibrated confidence and an unknown outcome — never proof of identity, and never for hiring, credit, housing or policing.

Reproducible releases

Every response names the model versions and SHA-256 weight digests behind it, so you can pin a version and diff a regression.

Pay per image. Or run it yourself.

Provisional · pricing finalises at GA

  • Free

    $0/ month

    For evaluating the contract and diffing against your current provider.

    • 1,000 images / month
    • Test + live keys
    • Face Detect & Analyze
    • Community support
    Start free

  • Scale

    Most teams

    $0.0012/ image

    Metered per successful image, with volume tiers from 250k images a month.

    • Unlimited keys and projects
    • Every released service
    • Usage + latency dashboard
    • 99.9% availability target at GA, email support
    Get an API key

  • On-premises

    Custom/ annual

    The same Docker inference plane running inside your own network or venue.

    • Self-hosted GPU or CPU image
    • Air-gapped deployment
    • Model pinning + private builds
    • Named engineer at GA
    Talk to us

Use case · migration

Replacing Kairos? Start with Face Analyze.

Face Analyze returns every analytical field you were pulling from Enroll — without creating a recognition record for someone who just walked past a photo booth. Point your client at a new base URL, swap the auth header, keep your parsing code.

See the field mapTalk to us about volume

Kairos Enroll

PixelDepot v1

topLeftX / topLeftY / w / h
bounding_box
leftEyeCenter / rightEyeCenter
landmarks.left_eye / right_eye
eyeDistance
eye_distance
yaw / pitch / roll
pose
quality
quality.score + components
gender.type + confidences
attributes.apparent_gender
asian / black / hispanic / white
attributes.apparent_ancestry
glasses · lips
attributes.glasses · lips

Send your first image today.

Free tier, no card, 1,000 images a month. Bring your Kairos payloads and diff the responses side by side.

Create a free accountBook a migration call