Dreamina CLI

Dreamina CLI

Use this skill when you need Dreamina（即梦） image or video generation, task querying, task listing, credit checks, or session management.

即梦 is the Chinese product name of Dreamina. If the user says 即梦, treat it as Dreamina and use this skill.

Prefer the packaged Python wrapper scripts in scripts/ over calling raw dreamina directly. The wrappers add path validation, command normalization, structured JSON output, --dry-run, and consistent error handling.

Treat dreamina -h and dreamina <subcommand> -h as the final source of truth when a specific flag or supported value is unclear.

Default workflow

1. If you need capability discovery, run python3 ./scripts/list_capabilities.py --format json.
2. Choose the narrowest wrapper script for the user request.
3. For expensive or ambiguous requests, run the wrapper with --dry-run first to inspect the exact CLI arguments.
4. Run the real wrapper command.
5. Read the wrapper JSON response instead of relying on shell exit code alone.
6. For async generation, keep the returned submit_id and use query_result.py if the user wants follow-up polling.

Script selection

• Image generation:
  • text2image.py
  • image2image.py
  • image_upscale.py
• Video generation:
  • text2video.py
  • image2video.py
  • frames2video.py
  • multiframe2video.py
  • multimodal2video.py
• Task and account operations:
  • query_result.py
  • list_task.py
  • user_credit.py
• Session operations:
  • login.py
  • login_checklogin.py
  • logout.py
  • relogin.py
  • session.py
• Capability discovery:
  • list_capabilities.py
• CLI version check:
  • version.py

Output contract

Every wrapper returns JSON with:

• ok: boolean
• command: wrapper command name
• cli_args: fully expanded command array
• data: parsed CLI payload on success
• error and details: normalized failure payload on failure

For generation wrappers, treat the submit as valid only when:

• submit_id is present
• gen_status is querying or success

If a generation wrapper returns gen_status=fail, the wrapper converts that to ok: false.

Safety rules

• Reuse the current login session unless the user explicitly asks to login, logout, or relogin.
• When you do need to invoke login or relogin, prefer using the --headless flag (e.g., python3 scripts/login.py --headless) to print OAuth Device Flow authorization material. After the user authorizes in the browser, use python3 scripts/login_checklogin.py --device-code <device_code> --poll 30.
• Some models may require a one-time Dreamina Web confirmation. If the CLI reports AigcComplianceConfirmationRequired, tell the user to finish that web-side authorization and retry.
• Seedream 5.0 Pro is currently exposed by the CLI as --model_version=5.0Pro on text2image and image2image.
• 5.0Pro supports resolution_type values 1k, 2k, and 4k; image generation defaults to model 5.0, not 5.0Pro.
• resolution_type is required for image generation/upscale, and video_resolution is required for every video generation command.
• text2image and image2image support paired --width / --height; custom dimensions cannot be combined with --ratio.
• Image generation wrappers support --generate-num / --generate_num for CLI generate_num in the range 1-10.
• seedance2.0mini is a supported Seedance 2.0 video model in current CLI help; treat it as 4-15s and 720p-only unless future dreamina <subcommand> -h output says otherwise.
• seedance2.5 supports 480p/720p and 4-30 second output. Its multimodal mode accepts audio-only input and uses higher model-specific input limits.
• Prefer small, reviewable generation batches.
• Keep track of submit_id for all async tasks.
• For supported commands, do not bypass the wrapper scripts unless the user explicitly asks for raw CLI execution.

References

• For wrapper inventory, parameters, and examples, read references/commands.md.
• For Vibe Coding IDE usage and portability guidance for OpenClaw or DeerFlow style agent runtimes, read references/integration.md.