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 type | Use |
|---|---|
image_gen | Text-to-image, image-to-image, multi-image fusion |
video_gen | Text-to-video, image-to-video, first/last-frame video |
image_understanding / ocr | Post-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
| Provider | Model | Specialty |
|---|---|---|
| MiniMax | image-01 / image-01-live | Strong single-subject consistency; style transfer (manga, genki, medieval, watercolor) |
| Volcengine | Seedream 4.0 | True image-to-image / inpainting / multi-image fusion (up to 10 reference images) |
| DesireCore Cloud | wan2.7-image etc. | Official cloud compute gateway, OpenAI Images API compatible |
| newapi | Routed by model name | new-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
| Provider | Model | Supported inputs |
|---|---|---|
| MiniMax | Hailuo-02 | First frame / first-last frame / S2V multi-image subject |
| Volcengine | Seedance | Multi-image + first-last frame + reference video/audio |
| DashScope | HappyHorse | i2v first frame / r2v multi-image |
| DesireCore Cloud | happyhorse-1.1 / doubao-seedance-2.0 | Official cloud compute, model-based selection |
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