EvoLink Media Agents

Overview

This example shows how to use the EvoLink provider inside HNO to generate text, images, and video via pkg/hno/models/evolink/*.

Setup

• Export env variables:

export EVO_API_KEY=sk-your-evolink-key
export EVO_BASE_URL=https://api.evolink.ai # optional, defaults to this value

Supported image models

All EvoLink image endpoints require an explicit model identifier. The provider currently exposes:

Model ID	Description	Typical use
gpt-4o-image	GPT-4o multi-modal image generation	High-fidelity creative renders, supports references + mask
doubao-seedream-4.0	Seedream 4.0 (Doubao)	Stylized illustration, anime aesthetics
gemini-2.5-flash-image	Nano Banana (Gemini Flash)	Fast drafts, low-latency ideation
qwen-image-edit	Qwen image editing	Mask-based edits, restoration
wan2.5-text-to-image	Wan2.5 TTI	Photorealistic text-to-image
wan2.5-image-to-image	Wan2.5 I2I	Reference-driven variation / style transfer

Use InvokeRequest.Extra["model"] or the model config default to switch among them.

Supported video models

All video endpoints also require a model value matching the EvoLink docs:

Model ID	Description	Typical use
veo3.1-fast	Veo 3.1 Fast text/image to video	Fast previews with Veo 3.1
sora-2	Sora-2 base model	High-fidelity cinematic clips
sora-2-pro	Sora-2 Pro	Extended capabilities + quality
wan2.5-text-to-video	Wan2.5 TTV	Photorealistic videos from prompts
wan2.5-image-to-video	Wan2.5 ITV	First-frame/reference driven videos
doubao-seedance-1.0-pro-fast	Seedance 1.0 Pro Fast	Stylized, dance-like motion (Doubao)

Video model quick tips:

• image_urls (array) provide reference frames. Wan2.5 ITV requires exactly one URL; Veo accepts up to two; Sora/Seedance accept at most one.
• duration depends on the model: Sora {10,15}, Wan2.5 {5,10}, Seedance 2–12. Leave it empty to use EvoLink defaults.
• quality applies to Sora-2-Pro (standard|high) and Seedance (720p|1080p).
• remove_watermark defaults to true on Sora variants—set the extra to false to keep the watermark.
• aspect_ratio has model-specific limits. Wan2.5 image-to-video derives its ratio from the input and ignores the field altogether.
• callback_url must be HTTPS; EvoLink POSTs completion/failure/cancel events there.

Text (Chat Completions)

package main

import (
  "context"
  "fmt"
  "os"

  evoText "github.com/rexleimo/agno-go/pkg/hno/models/evolink/text"
  "github.com/rexleimo/agno-go/pkg/hno/models"
  "github.com/rexleimo/agno-go/pkg/hno/types"
)

func main() {
  m, _ := evoText.New("evo-gpt-4o", evoText.Config{APIKey: os.Getenv("EVO_API_KEY"), BaseURL: os.Getenv("EVO_BASE_URL"), Temperature: 0.6})
  resp, _ := m.Invoke(context.Background(), &models.InvokeRequest{Messages: []*types.Message{ types.NewUserMessage("Write a 3-shot underwater city storyboard") }})
  fmt.Println(resp.Content)
}

Images (GPT-4O Images)

Allowed sizes: 1:1, 2:3, 3:2, 1024x1024, 1024x1536, 1536x1024. n ∈ {1,2,4}. Up to 5 references. If mask_url is provided, exactly one reference is required and the mask must be a PNG.

package main

import (
  "context"
  "fmt"
  "os"

  evoImg "github.com/rexleimo/agno-go/pkg/hno/models/evolink/image"
  "github.com/rexleimo/agno-go/pkg/hno/models"
  "github.com/rexleimo/agno-go/pkg/hno/types"
)

func main() {
  m, _ := evoImg.New("evo-gpt-4o-images", evoImg.Config{APIKey: os.Getenv("EVO_API_KEY"), BaseURL: os.Getenv("EVO_BASE_URL"), Size: "1:1", N: 2, Model: evoImg.ModelGPT4O})
  resp, _ := m.Invoke(context.Background(), &models.InvokeRequest{Messages: []*types.Message{ types.NewUserMessage("A futuristic city made of coral") }})
  fmt.Printf("task: %s, data: %#v\n", resp.ID, resp.Metadata.Extra["data"])
}

Video (Sora-2)

Sora accepts aspect ratios 16:9 or 9:16, durations 10 or 15 seconds, an optional single image_urls entry for image-to-video, and an optional HTTPS callback_url. Watermarks are removed by default; send remove_watermark=false when you need to keep them.

package main

import (
  "context"
  "fmt"
  "os"

  evoVid "github.com/rexleimo/agno-go/pkg/hno/models/evolink/video"
  "github.com/rexleimo/agno-go/pkg/hno/models"
  "github.com/rexleimo/agno-go/pkg/hno/types"
)

func main() {
  v, _ := evoVid.New("evo-sora-2", evoVid.Config{APIKey: os.Getenv("EVO_API_KEY"), BaseURL: os.Getenv("EVO_BASE_URL"), AspectRatio: "9:16", DurationSeconds: 15, Model: evoVid.ModelSora2})
  req := &models.InvokeRequest{
    Messages: []*types.Message{ types.NewUserMessage("A marine biologist swims with manta rays at dawn") },
    Extra: map[string]interface{}{
      "image_urls": []string{"https://assets.example.com/keyframes/manta.png"},
    },
  }
  resp, _ := v.Invoke(context.Background(), req)
  fmt.Printf("task: %s, status: %v\n", resp.ID, resp.Metadata.Extra["status"])
}

Notes

• Assets may expire after 24h; persist them if needed.
• EvoLink enforces strict moderation and HTTPS-only webhooks.
• Reference docs: GPT-4O images: https://docs.evolink.ai/en/api-manual/image-series/gpt-4o/gpt-4o-image-generation. Local Sora-2 HTML reference: /home/rex/下载/Sora-2 视频生成 - EvoLink.html.