jimeng-skill

Jimeng Skill

Use the Jimeng CLI to generate images or videos [Important concurrency limit]: Jimeng can run at most three queued or generating tasks in parallel!

Default preference order

Always prioritize user requirements; when the user has no requirements, use the following default preference order

Image generation

• Model preference order (model_version): 4.6 > 4.5 > 5.0 > 4.0
• Resolution (resolution_type): 2k

Video generation

• Model preference order (model_version): seedance2.0 > seedance2.0fast
• Resolution (video_resolution): 720p

General (image generation + video generation)

• Ratio preference order: 16:9 > 9:16 > 1:1

Initialization

1. First check status and credits

dreamina user_credit

If the command is not found, install it first:

curl -fsSL https://jimeng.jianying.com/cli | bash

If not logged in, you need to run dreamina login login. This command will launch a local browser to obtain login credentials.

2. View all features

dreamina --help

Image generation

Image generation is relatively fast; generate directly:

dreamina text2image --prompt "你的提示词" --ratio 16:9

After generation is complete, simply download it locally.

Video generation (asynchronous task + progress polling)

Video generation usually takes a long time. After submitting a task, it will return submit_id, and you need to poll at regular intervals to get task progress and queue status.

1. Submit a video generation task

View help for specific subcommands:

dreamina multimodal2video --help
# 或
dreamina text2video --help

After submitting the task, it will return submit_id and the current queue position. Please record submit_id for subsequent polling to query progress.

2. Query task status and poll

Use the following command to query task status at regular intervals:

dreamina query_result --submit_id <submit_id>

This command will return:

• Current task status
• Queue position/queue progress
• If completed, it will return the download URL of the generated result

3. Scheduled polling process

Since video generation takes a long time, you need to poll once per minute:

# 每分钟查询一次状态
while true; do
  echo "=== $(date) ==="
  dreamina query_result --submit_id <submit_id>
  echo
  sleep 60
done

After the task is completed, extract the download URL from the returned result, use curl to download it, and specify a semantic filename:

curl -o "output.mp4" "<download_url>"

4. Queue progress and waiting time estimation

Estimate based on the actual consumption speed of the queue:

• Record the time difference and queue position changes for each poll
• Calculatehow many positions the queue advances on average per minute
• Divide the current remaining queue position by the advance speed to getthe estimated number of minutes remaining until the queue ends
• Simply tell the user these two pieces of information

Universal reference

Image generation and video generation support multiple reference inputs

Reference reference images according to the order in which the generation parameter resources are passed in: the way to reference is to write 图1, 图2, 音频1, 视频1 in the prompt, for example

提取图1、图2、图3的相机,把背景换成白色,相机在一个白色桌子上,然后以相机为主体360°缓慢旋转,晰展示相机的正面侧面以及背面。

Generate from a local text prompt file

All image generation and video generation commands support reading the prompt from a local text file: ina multimodal2video --prompt "$(cat prompt.txt)"

## 从本地文本提示词文件生成

所有生图和生视频命令都支持从本地文本文件读取提示词：

```bash
dreamina text2image --prompt "$(cat prompt.txt)" --ratio 16:9
dreamina text2video --prompt "$(cat prompt.txt)"

Local prompt + multi-reference video generation

Command line example:

dreamina multimodal2video --image ./角色A.jpg --image ./角色B.jpg --image ./场景1.jpg --prompt "角色A参考图1，角色B参考图2，场景1参考图3。 $(cat prompt.txt)"

Commonly used subcommands (prefer using these)

• dreamina text2image: Generate an image based on the prompt
• dreamina image2image: Generate an image from multiple reference images
• dreamina text2video: Text-to-video
• dreamina multimodal2video: All-purpose reference-to-video generation (supports image reference + text prompt)
• dreamina query_result: Query asynchronous task status (required for video generation)
• dreamina list_task: List submitted tasks and their statuses
• dreamina user_credit: Check remaining credits

Download resources

• Use curl to download and specify a semantic filename:

For example

curl -o "semantic-filename.mp4" "<download_url>"

Error handling

If a generation error occurs, you must tell the user the specific error message and guide the user to resolve the error.