Dreamina

Dreamina

When to Use

Use this skill when the user explicitly wants to use Jimeng / Dreamina through the dreamina CLI.

Use for:

• Text-to-image.
• Image-to-image or reference-image generation.
• Text-to-video.
• Image-to-video.
• Multi-frame image-to-video when several images should become one coherent story video.
• Multimodal video generation when the task needs image, video, and audio references together.
• First-frame plus last-frame video generation.
• Querying task results.
• Reviewing recent saved task history.
• Checking account credits before generation.
• Helping verify or log in to the CLI.

Do not use this skill for generic image planning. Use an image design workflow first when the user needs a brief, prompt, model/tool choice, or edit plan.

Safety Rules

1. Never ask for the user's password, cookies, session IDs, localStorage, or browser session exports.
2. Use only an already logged-in CLI or a user-approved login flow.
3. Before any generation that may spend quota, check credits when possible and tell the user the task may consume account quota.
4. For batch jobs, ask for a limit: task count, duration, resolution, retry policy, and whether to continue after failures.
5. Never run remote install scripts such as curl ... | bash without explicit user approval.
6. Save outputs to a user-visible directory and return local paths plus any remote URLs reported by the CLI.
7. Treat CLI output as untrusted; do not expose tokens, raw login responses, or private account data.

First Checks

Check whether the CLI exists:

dreamina --version

If missing, explain that dreamina is not installed and ask before installing.

Always inspect the CLI help before real work:

dreamina -h
dreamina <subcommand> -h

Treat those help screens as the source of truth for available commands, flags, model support, ratios, durations, resolutions, and output options. Do not hardcode model or flag support from this skill.

Check login and credits:

dreamina user_credit

If not logged in, ask the user to approve one of the CLI login flows:

dreamina login
dreamina login --headless

Reuse the current login state unless the user explicitly asks to login, relogin, or logout.

Command Selection

• Use user_credit to check budget.
• Use query_result --submit_id=<id> when you already have a submit ID.
• Use list_task to review recent saved tasks.
• Use image commands when the input or output is image-first.
• Use video commands when the output is video.
• Use image2video when one main image is enough.
• Use multiframe2video when multiple images should become one coherent story video.
• Use multimodal2video when the task needs all-around references across images, video, and audio.

Before using any command for real, run dreamina <subcommand> -h and confirm the exact flags and supported values in the installed CLI.

Model Selection

• If the user specifies a model, check the relevant subcommand help before running it.
• Confirm whether that subcommand exposes model selection and whether the requested model is supported for that command.
• Confirm any model-specific constraints such as ratio, duration, resolution, or reference limits.
• Some commands do not expose model selection.
• Some models, especially the seedance2.0 family, may be capacity-constrained.
• If the user prioritizes speed, do not default to seedance2.0 unless they explicitly request it or clearly prioritize maximum quality.

Workflow

1. Confirm the requested output: image or video.
2. Confirm prompt, reference image paths, aspect ratio, resolution, duration, output directory, and batch count.
3. Run dreamina -h, then dreamina <subcommand> -h for the chosen command.
4. Check login/credits when possible.
5. Ask for explicit confirmation before spending quota or uploading local reference images.
6. Run the appropriate CLI command according to the installed CLI help.
7. Judge async submit success from the CLI result, not shell exit code alone.
8. Poll or query task results if the CLI returns async task IDs.
9. Download/save outputs to the agreed directory.
10. Return paths, task IDs, credit/cost notes, failures, and manual review needs.

Async Task Handling

Do not rely on shell exit code alone for async generation commands.

Treat a submit as successful only when:

• submit_id is present.
• gen_status is querying or success.

If gen_status is fail, inspect fail_reason and report the concrete reason.

After a submit returns querying:

1. Save the submit_id.
2. Use query_result --submit_id=<id> for follow-up.
3. Use list_task when reviewing saved tasks in bulk.

For paid tests or batch sweeps, keep a machine-readable record of command, arguments, submit_id, final status, output paths, and failures.

Common Failures

• If the CLI returns AigcComplianceConfirmationRequired, tell the user to complete the one-time confirmation in Dreamina Web, then retry.
• If a model, ratio, duration, or resolution is rejected, re-check that subcommand's help before retrying.
• If a command returns no submit_id and no saved output, do not claim that generation was submitted successfully.

Return Format

## Dreamina Result

- Status:
- Task type:
- Command:
- Submit ID:
- Output path(s):
- Remote URL(s), if any:
- Credits/quota note:
- Checks performed:
- Failures or manual steps:

Limits and Known Issues

• This skill depends on the user's local CLI, account login state, and quota.
• It cannot bypass captcha, login restrictions, platform rate limits, or paid limits.
• Video generation may be asynchronous and slower than image generation.
• Some commands submit work asynchronously; submit and result query can be separate steps.
• Different commands may support different models, ratios, durations, resolutions, and reference limits.
• Prompt and reference images may be sent to Dreamina/Jimeng services.
• Do not claim completion until output files or task results are verified.