Media Generation

Generate images, music clips, and videos from text prompts using state-of-the-art AI models.

Yukkai can generate three types of media from text prompts: images, music, and videos. Each tool uses a different AI model and is designed for a specific creative purpose.

Overview

Tool Model Output Use Case
generate_image Google Gemini 3 Pro Image PNG image Illustrations, UI mockups, concept art
generate_music Google Lyria 3 MP3 audio Background music, soundtracks, jingles
generate_video ByteDance Seedance 1.5 Pro MP4 video Short animations, motion clips

generate_image

Generate an image from a text description. Supports both text-to-image (generate from scratch) and image-to-image (edit/transform a source image) modes.

Parameters

Parameter Type Required Description
prompt string Yes Image description (English preferred)
aspect_ratio string No 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 (default: 1:1)
image_size string No 1K (default), 2K, 4K
model string No Default: google/gemini-3-pro-image-preview
input_image_url string No Public URL of a source image for image-to-image mode

Modes

  • Text-to-image: Provide only a prompt. The model generates from scratch.
  • Image-to-image: Provide a prompt + input_image_url. The model edits, transforms, or creates a variation of the source image. If your image is local, upload it first to get a public URL.

Output

Images are saved to ~/YukkaiDocs/images/{Title}_{timestamp}.png and displayed automatically inline in the chat.

Example

generate_image(
  prompt: "A minimalist app icon with a blue gradient background and a white compass symbol",
  aspect_ratio: "1:1",
  image_size: "2K"
)

⚠️ Strict Rules

  1. Call generate_image exactly once per user request.
  2. After it returns, stop — do not call show_media, bash, web_browse, or generate_image again.
  3. The image is displayed automatically in the chat via base64 inline rendering.
  4. Only the first image variant is shown, even if the API returns multiple.
  5. If using image-to-image with a local image, call the upload tool first to get a public URL.

generate_music

Generate a music clip (30 seconds) or a full song from a text description using Google Lyria 3.

Parameters

Parameter Type Required Description
prompt string Yes Describe genre, mood, instruments, tempo, and optional lyrics
title string No Title for the file name
model string No google/lyria-3-clip-preview (30s clip, default) or google/lyria-3-pro-preview (full song)

Output

Audio is saved to ~/YukkaiDocs/music/{Title}_{timestamp}.mp3 and displayed with an inline audio player automatically.

Example

generate_music(
  prompt: "Upbeat lo-fi hip hop with jazz piano, warm vinyl crackle, 85 BPM, mellow afternoon vibe",
  title: "AfternoonLoFi"
)

⚠️ Strict Rules

  1. Call generate_music exactly once per user request.
  2. After it returns, stop — do not call show_media, bash, or any generate_* tool again.

generate_video

Generate a short video (4–8 seconds) from a text prompt. Async operation — takes 1–3 minutes to complete.

Parameters

Parameter Type Required Description
prompt string Yes Scene description, movement, mood, style
title string No Title for the file name
duration string No 4 (default) or 8 seconds
resolution string No 480p, 720p (default), 1080p
aspect_ratio string No 16:9 (default), 9:16, 4:3, 3:4, 1:1, 21:9
image_url string No Public URL of a keyframe image for image-to-video mode

Modes

  • Text-to-video: Provide only a prompt. The model generates from scratch.
  • Image-to-video: Provide a prompt + image_url. The model animates from the keyframe image. If your image is local, upload it first to get a public URL.

Output

Video is saved to ~/YukkaiDocs/videos/{Title}_{timestamp}.mp4 and displayed with an inline video player automatically.

Example

generate_video(
  prompt: "A drone shot over a misty forest at sunrise, camera slowly ascending through the treetops, warm golden light",
  duration: "8",
  resolution: "1080p",
  aspect_ratio: "16:9"
)

⚠️ Strict Rules

  1. Call generate_video exactly once per user request.
  2. After it returns, stop — do not call show_media, bash, or any generate_* tool again.
  3. If using image-to-video with a local image, call the upload tool first to get a public URL.

Tips

  • Be descriptive in prompts — include style, mood, lighting, colors, composition, and subject details for best results.
  • English prompts work best with all three models.
  • Image-to-image / image-to-video: Always upload local images first to obtain a public URL before passing it as a parameter.
  • All generated media is displayed automatically — you do not need to call any additional tool to show the result.
  • Only one call per tool per request — the strict rules enforce this to prevent duplicate generation and wasted credits.

Last updated: July 19, 2026