Skip to main content
Kronk — open source AI tooling for Go Local inference. Fewer moving parts. Explore Kronk
Ardan Labs home
Contact Us
Ardan Labs Logo

Let's Talk

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Image Generation Experimental

Local image generation for Go

Native stable-diffusion.cpp bindings for local image and video generation. Text-to-image, image-to-image, ControlNet, ADetailer, AnimateDiff, and Real-ESRGAN upscaling — with pure-Go PNG and JPEG I/O, no CGo, and no cloud dependency. OpenAI-compatible /v1/images/generations and /v1/images/edits on the Kronk model server.

Full Details View on GitHub
Malina — local image generation for Go
kronk — malina example
git clone https://github.com/ardanlabs/kronk.gitcd kronkmake example-malina✓ installed stable-diffusion.cpp libraries✓ downloaded sd-1.5 model bundle- dimensions       : 512x512- output           : malina.png
  • stable-diffusion.cpp
  • Stable Diffusion 1.5
  • SDXL
  • ControlNet
  • ADetailer
  • AnimateDiff
  • Wan2.2 S2V
  • Real-ESRGAN
  • Metal
  • CUDA
  • Vulkan
  • ROCm
  • PNG
  • JPEG
  • Motion-JPEG AVI
  • OpenAI images API
  • Kronk model server
  • Go SDK

Why Malina

Built for production-minded Go teams

Malina brings local image generation into the Kronk ecosystem so Go apps can create and transform images on infrastructure they control — with pure-Go I/O and no cloud dependency. The public API is experimental and still evolving.

Talk to an Engineer
01

Generate on your stack

Run image generation where your application runs — not behind another vendor API.

02

No CGo required

Pure-Go PNG and JPEG I/O plus Motion-JPEG AVI muxing keeps the integration path simpler for Go services and tools.

03

Familiar API direction

OpenAI-compatible /v1/images/generations and /v1/images/edits on the Kronk model server, with advanced workflows available in the Go SDK.

04

Fits the Kronk model

Same local-first philosophy as Kronk and Bucky: own the runtime, control cost, keep data private.

What you can do

Image workflows Malina is built for

Use the Go SDK inside your process for the full surface area, or create and edit images through Kronk’s OpenAI-compatible image APIs.

  1. Step 01

    Text-to-image

    Generate images from prompts with curated Stable Diffusion, SDXL, and multi-file diffusion bundles.

  2. Step 02

    Image-to-image

    Transform an existing PNG or JPEG with a prompt and strength — through the SDK or /v1/images/edits.

  3. Step 03

    ControlNet & ADetailer

    Canny edge conditioning for composition control, plus face detection and refinement for portraits.

  4. Step 04

    Video & upscaling

    AnimateDiff frame sequences, Wan2.2 S2V speech-to-video, Real-ESRGAN upscaling, and Motion-JPEG AVI output.

API surface

OpenAI-compatible, running locally

Malina powers Kronk’s image endpoints, so clients that already speak the OpenAI images API can generate and edit images on your own hardware.

ControlNet, ADetailer, video generation, and upscaling are available through the Malina Go SDK.

Explore the Malina Manual
  1. Generations /v1/images/generations Create images from a prompt with an installed model bundle and get base64 image data back.
  2. Edits /v1/images/edits Transform a source PNG or JPEG with a prompt and strength using multipart form fields.
  3. Progress events GET /v1/images/events Follow model loading and generation progress as server-sent events.

Capabilities

What you get

stable-diffusion.cpp

Native bindings for local diffusion models without standing up a separate generation service first.

Curated model bundles

Pull complete workflows — SD 1.5, ControlNet, AnimateDiff, Real-ESRGAN, ADetailer, SDXL, and more — with validated manifests.

Pure-Go media I/O

PNG and JPEG handling in Go, plus Motion-JPEG AVI muxing so image and video pipelines stay inside your application.

SDK + model server

Use Malina in-process, or create and edit images through Kronk’s OpenAI-compatible image endpoints.

Hardware accelerated

Metal, CUDA, Vulkan, and ROCm backends depending on host — with automatic library detection and install.

Local by default

No cloud round-trips for generation — useful for air-gapped, regulated, or cost-sensitive environments.

Models

Pull a complete workflow

Curated bundles from the catalog — pull what you need with the Kronk CLI.

sd-1.5

~4.3 GB

Stable Diffusion 1.5 checkpoint — the default path for first image generation

controlnet-canny-sd1.5

~2.3 GB

Quantized SD 1.5 plus Canny ControlNet for composition-constrained generation

realesrgan-x4-anime

~18 MB

Compact Real-ESRGAN 4× anime upscaler for enlarging generated frames

adetailer-face-yolov8n

~1.6 GB

Face detection and inpainting refinement for portrait workflows

animatediff-sd1.5

~2.4 GB

AnimateDiff motion module for temporally conditioned video sequences

sdxl-base-1.0

~6.9 GB

SDXL Base 1.0 checkpoint for higher-resolution generation

Get started

Run Malina with Kronk

Install Kronk, pull stable-diffusion.cpp libraries and an SD 1.5 bundle, then run the Malina example — or generate images through the model server once it is up.

brew install ardanlabs/kronk/kronk# or: go install github.com/ardanlabs/kronk/cmd/kronk@latest

Platforms

Where Malina runs

Compatible stable-diffusion.cpp library bundles are downloaded per host. Use the CLI as the source of truth for supported combinations on your install.

macOS

CPU
arm64
GPU
Metal

Linux

CPU
amd64
GPU
Vulkan ROCm

Windows

CPU
amd64
GPU
CUDA Vulkan ROCm

Related tooling

More in the Kronk ecosystem

All AI Tooling

Kronk

Text, vision, embedding, and reranking through llama.cpp and yzma — the foundation of the stack.

Explore Kronk

Bucky

Local speech-to-text through whisper.cpp — transcriptions, translations, streaming, and diarization.

Explore Bucky

Build AI systems you can own

Talk with Ardan Labs about local inference, private AI infrastructure, and production systems beyond API prototypes.

Talk to an Engineer Explore AI Engineering

Leverage our experience. Get what you need.

Let's Talk