Implementation:Ollama Ollama Imagegen Engine Main

Knowledge Sources	Ollama
Domains	Image Generation, CLI
Last Updated	2025-02-15 00:00 GMT

Overview

Standalone CLI binary for direct model inference, supporting text generation, image generation, and multimodal models with GPU profiling.

Description

The main.go file in cmd/engine provides a standalone CLI for testing and running models directly without the Ollama server. It parses flags for model path, prompt, generation parameters (max-tokens, temperature, top-p, top-k), image generation parameters (width, height, steps, seed, output path), and utility options (list tensors, CPU profiling, GPU capture, wired memory limit). The main function detects model type and dispatches to the appropriate pipeline: Z-Image or FLUX.2 for image generation, and Gemma3/Llama/GPT-OSS for text generation. It supports multi-image input for editing, negative prompts for CFG, TeaCache for fast inference, and fused QKV for non-quantized models.

Usage

Used for development and testing of MLX models outside the Ollama server framework, supporting GPU trace capture and profiling.

Code Reference

Source Location

• Repository: Ollama
• File: x/imagegen/cmd/engine/main.go
• Lines: 1-294

Signature

func main()

// CLI flags:
// -model        Model directory
// -prompt       Text prompt
// -max-tokens   Max tokens for text generation
// -width        Image width
// -height       Image height
// -steps        Denoising steps
// -seed         Random seed
// -output       Output image path
// -zimage       Z-Image generation mode
// -flux2        FLUX.2 Klein generation mode
// -teacache     Enable TeaCache
// -gpu-capture  GPU trace capture path

Import

import (
    "github.com/ollama/ollama/x/imagegen"
    "github.com/ollama/ollama/x/imagegen/models/flux2"
    "github.com/ollama/ollama/x/imagegen/models/zimage"
)

I/O Contract

Inputs

Name	Type	Required	Description
-model	string	Yes	Path to model directory
-prompt	string	Yes	Text prompt for generation
-output	string	No	Output file path (default "output.png")

Outputs

Name	Type	Description
stdout	text	Generated text or status messages
output file	PNG	Generated image file

Usage Examples

// Command line usage:
// go run ./x/imagegen/cmd/engine -model /path/to/zimage -zimage -prompt "a cat" -output cat.png
// go run ./x/imagegen/cmd/engine -model /path/to/llama -prompt "Hello" -max-tokens 100

Related Pages

• Principle:Ollama_Ollama_ImageGeneration