Replicate Provider

Replicate provides a marketplace to run any open-source model through a simple API. Perfect for experimentation and accessing diverse models including image generation, video, and specialized models.

Setup

Installation

import (
    "github.com/digitallysavvy/go-ai/pkg/ai"
    "github.com/digitallysavvy/go-ai/pkg/providers/replicate"
)

Configuration

provider := replicate.New(replicate.Config{
    APIKey: os.Getenv("REPLICATE_API_KEY"),
})

model, err := provider.LanguageModel("meta/meta-llama-3-70b-instruct")

Get API Key

export REPLICATE_API_KEY=r8_...

Available Models

Language Models

Model ID	Context	Price	Best For
meta/meta-llama-3-70b-instruct	8K	$0.65/1M	General purpose
mistralai/mixtral-8x7b-instruct-v0.1	32K	$0.30/1M	Balanced

Image Generation

Model ID	Quality	Price	Best For
stability-ai/sdxl	High	$0.0119/run	Images
black-forest-labs/flux-pro	Excellent	$0.055/run	Professional
black-forest-labs/flux-schnell	Good	$0.003/run	Fast

Video Generation

Model ID	Quality	Price	Best For
stability-ai/stable-video-diffusion	Medium	$0.10/run	Video from image

provider.VideoModel(modelID) accepts both forms Replicate uses:

• Unversioned — "owner/model" (no :version). Requests go to POST /models/{owner}/{model}/predictions, which always runs the model's latest version. This is now supported; previously only versioned IDs worked.
• Versioned — "owner/model:version-hash". Requests go to POST /predictions with version in the body, pinning an exact version.

videoModel, err := provider.VideoModel("stability-ai/stable-video-diffusion")
if err != nil {
    log.Fatal(err)
}

result, err := ai.GenerateVideo(ctx, ai.GenerateVideoOptions{
    Model:  videoModel,
    Prompt: ai.VideoPrompt{Text: "A boat sailing across a calm lake"},
})

VideoModel also implements provider.VideoModelStarter / VideoModelStatusChecker for the async ai.ExperimentalStartVideo / ai.ExperimentalGetVideoStatus calls.

Examples

Basic Text Generation

package main

import (
    "context"
    "fmt"
    "log"
    "os"

    "github.com/digitallysavvy/go-ai/pkg/ai"
    "github.com/digitallysavvy/go-ai/pkg/providers/replicate"
)

func main() {
    ctx := context.Background()
    provider := replicate.New(replicate.Config{
        APIKey: os.Getenv("REPLICATE_API_KEY"),
    })

    model, err := provider.LanguageModel("meta/meta-llama-3-70b-instruct")
    if err != nil {
        log.Fatal(err)
    }

    result, err := ai.GenerateText(ctx, ai.GenerateTextOptions{Model: model, Prompt: "Explain Replicate"})
    if err != nil {
        log.Fatal(err)
    }
    fmt.Println(result.Text)
}

Image Generation

imageModel, err := provider.ImageModel("black-forest-labs/flux-pro")
if err != nil {
    log.Fatal(err)
}

result, err := ai.GenerateImage(ctx, ai.GenerateImageOptions{
    Model:  imageModel,
    Prompt: "A cyberpunk city at night",
    Size:   "1024x1024",
})

fmt.Printf("Image bytes: %d\n", len(result.Images[0].Data))

Best Practices

1. Model Discovery

   • Browse Replicate marketplace for models
   • Try different models easily
   • Great for prototyping

2. Cost Management

   • Pay-per-run pricing
   • Monitor usage carefully
   • Cache results when possible

3. Use Cases

   • Experimentation and testing
   • Specialized model access
   • Image/video generation

Workflow Serialization

Replicate image models can cross a workflow boundary with providerutils.SerializeModel / DeserializeModel. See Provider Serialization for the mechanism; video models are not yet serializable.

See Also

• Replicate Documentation
• Model Marketplace