Skip to main content

Media Generation

DesireCore can use configured media providers to generate images, generate videos, and improve existing images. Media generation is typically invoked by skills or agents within appropriate tasks.

Prerequisites

Enable a provider for the relevant service type in Settings > Compute Service:

Service typeUse
image_genText-to-image, image-to-image, multi-image fusion
video_genText-to-video, image-to-video, first/last-frame video
image_understanding / ocrPost-generation review and image understanding

Official cloud compute, third-party API keys, and custom OpenAI-compatible providers can all serve as backends depending on the supplier's capabilities.

GenerateImage

GenerateImage creates or edits images. Common capabilities include:

  • Generate images from text prompts
  • Use one or more reference images for image-to-image or multi-image fusion
  • Adjust style, composition, size, and detail
  • Auto-save output and display it in the chat

Image Generation Provider Comparison

ProviderModelSpecialty
MiniMaximage-01 / image-01-liveStrong single-subject consistency; style transfer (manga, genki, medieval, watercolor)
VolcengineSeedream 4.0True image-to-image / inpainting / multi-image fusion (up to 10 reference images)
DesireCore Cloudwan2.7-image etc.Official cloud compute gateway, OpenAI Images API compatible
newapiRouted by model namenew-api gateway relay, multi-vendor support

Usage Examples

Describe the image you want directly in conversation:

"Generate a cyberpunk city nightscape with neon lights reflecting on rain-soaked streets."

To generate based on a reference image:

"Using this product photo as reference, generate a white-background e-commerce hero image." (attach the image)

GenerateVideo

GenerateVideo creates videos. Common capabilities include:

  • Text-to-video: generate video from a text description
  • Image-to-video: use an image as the first frame
  • First/last-frame constraints: specify both frames to control camera movement
  • Multi-image subject reference: maintain character/object consistency
  • Reference video/audio: use existing media to control camera style

Video Generation Provider Comparison

ProviderModelSupported inputs
MiniMaxHailuo-02First frame / first-last frame / S2V multi-image subject
VolcengineSeedanceMulti-image + first-last frame + reference video/audio
DashScopeHappyHorsei2v first frame / r2v multi-image
DesireCore Cloudhappyhorse-1.1 / doubao-seedance-2.0Official cloud compute, model-based selection
Asynchronous Generation

Video generation typically takes several minutes. After submission, processing runs in the background; the result is displayed in the conversation automatically once complete—you do not need to wait.

BeautifyImage

BeautifyImage optimizes existing images, for example:

  • Centering and cropping
  • Removing excess borders
  • Quality enhancement or size adjustment
  • Output-size protection to prevent abnormally large files

Approvals and Costs

Media generation may consume API quota or account credits. When a paid provider is used, reference material is uploaded, or local files are written, the system requests confirmation based on risk and permission policies.

Best Practices

  • Use a vision model when you need to "understand" an image; image generation models are for creating images
  • Choose a chat model for conversation; image/video generation models do not appear in the regular model selector
  • For production assets, have the agent review the output with a visual understanding tool to check text, composition, and obvious errors