jimeng-cli-text2image — 即梦文生图 CLI 执行

Execute text-to-image generation via the dreamina CLI tool. This skill handles command construction, credit management, task submission, polling, and result retrieval.

This skill does NOT craft prompts. Prompt crafting is handled by jimeng-prompt-text2image. Use that skill first to create the prompt, get user approval, then activate this skill for CLI execution.

When to use this skill

Use this skill when:

• The user has an approved prompt and wants to generate the image
• The user asks to "generate this image" or "run this prompt"
• The user needs to check credits, query task status, or download results
• The user mentions dreamina CLI commands explicitly
• The user wants to manage sessions, list tasks, or upscale images

Do NOT use this skill for:

• Writing or improving prompts → use jimeng-prompt-text2image
• Image-to-image CLI → use jimeng-cli-image2image
• Video CLI → use jimeng-cli-text2video / jimeng-cli-image2video

Core Execution Flow

1. CHECK   → dreamina user_credit                  # Always first — check credits
2. SUBMIT  → dreamina text2image --prompt="..." --ratio=X:Y --poll=0  # Async → get submit_id
3. POLL    → Agent 每 ~5 秒手动调用 query_result --submit_id=<id> 检查 gen_status
4. RETRIEVE → gen_status="success" → 提取结果并报告

定时查询 SOP（智能体行为规范，非 shell 死循环）

提交生成任务后，应由智能体（AI）负责任务状态查询，而非用 while true shell 脚本阻塞终端：

步骤

Step 1 — 提交任务（一次 terminal 调用）：

dreamina text2image --prompt="..." --ratio=16:9 --poll=0

→ 解析输出中的 submit_id，记录下来。

Step 2 — 智能体周期查询（多次 terminal 调用，每次单独）：

每 ~5 秒执行一次: dreamina query_result --submit_id=<submit_id>

根据返回的 gen_status 做分支判断：

> 禁止在 terminal 中使用 while true; sleep 5; ... 死循环。应由智能体在多次对话轮次中调用 query_result，每次独立发起 terminal 调用。

示例（智能体自身逻辑）

# Turn 1
terminal: dreamina text2image --prompt="..." --poll=0
→ parse: submit_id = "abc-123"
→ tell user: "已提交任务，submit_id=abc-123，5秒后检查结果"

# Turn 2 (after ~5s)
terminal: dreamina query_result --submit_id=abc-123
→ gen_status = "querying"
→ tell user: "任务处理中，5秒后再检查"

# Turn 3 (after ~5s)
terminal: dreamina query_result --submit_id=abc-123
→ gen_status = "success"
→ extract result_url, report to user

> 原理: 每次单独调用 terminal 而非 shell 死循环，智能体可以在每次查询结果后做灵活判断（重试、报告进度、超时处理），且不阻塞终端资源。

How to use this skill

Step 1: Verify prerequisites

Before any generation command, verify the CLI environment:

1a. Check CLI is installed:

dreamina -h

If not installed:

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

1b. Check login status and credits:

dreamina user_credit

This is the definitive self-check. If it returns JSON with credit info, login and environment are working.

If login needed:

dreamina login              # Opens browser for OAuth (default)
dreamina login --debug      # Debug mode — prints callback URL for troubleshooting
dreamina login --headless   # QR code mode — scan with Douyin app

Additional account commands:

dreamina relogin            # Switch to a different account
dreamina logout             # Clear credentials (keeps config.toml and tasks.db)

1c. Check config file (if login issues persist):

ls -la ~/.dreamina_cli/

> ⚠️ Note: The official docs reference config.toml, credential.json, and tasks.db at ~/.dreamina_cli/, but these files may not exist after login — especially in Docker environments. Auth state can be stored ephemerally. If ls shows only SKILL.md and version.json, login is still valid if dreamina user_credit returns your balance. See the dreamina-cli skill's references/docker-compatibility.md for Docker-specific setup info.

Step 2: Ensure prompt is ready

This skill requires a production-ready prompt. The prompt should already be:

• Crafted using jimeng-prompt-text2image methodology
• Reviewed and approved by the user
• In Chinese (即梦 works best with Chinese prompts)

If no prompt exists, activate jimeng-prompt-text2image first.

Step 3: Map prompt parameters to CLI arguments

Take the prompt's recommendations and map them to CLI parameters:

Load references/parameter-reference.md for the complete parameter map.

Step 4: Execute the generation

Recommended (async + 智能体周期查询):

dreamina text2image --prompt="<prompt>" --ratio=16:9 --model_version=5.0 --resolution_type=4k --poll=0

→ 解析获取 submit_id，智能体随后每 ~5 秒调用 query_result 检查状态。

Quick poll (for fast tasks, auto 1s polling):

dreamina text2image --prompt="<prompt>" --ratio=16:9 --model_version=5.0 --resolution_type=4k --poll=30

With session (organized projects):

dreamina session create "project-name"  # First time
dreamina text2image --prompt="<prompt>" --session=<session_id> --ratio=16:9 --poll=0
# Then follow 5s polling SOP using submit_id

Step 5: Handle results

After generation completes:

• Confirm the gen_status is "success"
• If using the 5s 智能体周期查询 → 每次 query_result 返回后智能体根据 gen_status 做分支判断
• If using --poll=N and it completes → output is available immediately
• If still "querying" after reasonable wait time → ask user if they want to continue waiting

Step 6: Handle errors

Common errors and responses — see references/workflow-patterns.md for detailed patterns.

Key Parameters Quick Reference

Poll behavior: When --poll=N is set, the CLI polls every 1 second for up to N seconds.

• If task completes within N seconds → final result returned immediately
• If N seconds elapse without completion → returns a "querying" status intermediate result; use dreamina query_result --submit_id= to check later

Essential CLI Commands

# Credit & auth
dreamina user_credit                    # Check credit balance
dreamina login                          # Login (OAuth browser)
dreamina login --headless               # Login (QR code)

# Generation
dreamina text2image --prompt="..."      # Basic generation
dreamina text2image --prompt="..." --ratio=16:9 --model_version=5.0 --resolution_type=4k --poll=30

# Task management
dreamina query_result --submit_id=<id>               # Check task status
dreamina query_result --submit_id=<id> --download_dir=./downloads  # Query + download
dreamina list_task                                   # List all tasks
dreamina list_task --gen_status=success              # Filter by status
dreamina list_task --submit_id=<id>                  # Filter by submit_id

# Session management
dreamina session create "name"          # Create project session
dreamina session list                   # List sessions

# Utility
dreamina image_upscale --image ./x.png --resolution_type=4k  # Upscale image

Model Selection Guide

Load references/model-guide.md for detailed model selection guidance.

FAQ (from official documentation)

Q: Login succeeds but generation commands still fail?

A: Two checks: (1) verify ~/.dreamina_cli/config.toml exists and is correct; (2) run dreamina user_credit — if this fails, login/config is the issue. Don't test generation commands until user_credit works.

Q: Browser login flow is stuck?

A: Use dreamina login --debug to print detailed debug info including the callback URL. Provide this output for troubleshooting.

Q: Async task shows no final result for a long time?

A: Use --poll=N (recommend --poll=30) for auto-waiting. If timeout, save the submit_id from the intermediate result and query manually: dreamina query_result --submit_id=.

Q: How to switch or re-login to another account?

A: Run dreamina relogin — it clears existing login state and starts a new login flow.

Q: How to completely clear local login info?

A: Run dreamina logout — this clears credential.json only. config.toml and tasks.db are preserved on native installs, but these files may be absent in Docker setups.

Gotchas

1. Always check credits first — never run a generation without dreamina user_credit. Warn user if balance is low. If user_credit fails, do NOT proceed to generation — fix login/config first
2. Prompt goes through prompt skill first — don't craft prompts here. Route to jimeng-prompt-text2image for all prompt writing
3. ~/.dreamina_cli/ is the config directory — may contain config.toml (settings), credential.json (login), tasks.db (history) after native (non-Docker) login. In Docker setups these files may be absent. Don't delete this directory
4. --poll polls every 1 second — using --poll=30 means up to 30 polling attempts, not a 30-second fixed wait. Timeout returns "querying" status, not failure
5. Model 5.0 is the safe default — latest flagship, best quality. Don't hardcode but default here unless the prompt recommends a specific version
6. Resolution must match model capability — 1k only for 3.x, 2k for all, 4k for 4.0+
7. Chinese prompts produce best results — 即梦 is optimized for Chinese. If the user's prompt is in English, suggest translating
8. 不要写 shell 死循环做任务轮询 — 提交任务时用 --poll=0 获取 submit_id，然后由智能体在对话轮次中每 ~5 秒手动调用一次 query_result。禁止 while true; sleep 5; 阻塞终端。智能体需根据每次返回的 gen_status 做分支判断（继续等/报告结果/通知超时）

Available Resources