jimeng-cli-text2video — 即梦文生视频 CLI 执行

Execute text-to-video generation via the dreamina CLI tool using Seedance 2.0 models. This skill handles command construction, credit management, task submission, polling, and result retrieval.

This skill does NOT craft prompts. Video prompt crafting is handled by jimeng-prompt-text2video. Use that skill first, get user approval, then activate this skill for CLI execution.

When to use this skill

Use this skill when:

• The user has an approved video prompt and wants to generate the video
• The user asks to "generate this video" or "run this video prompt"
• The user needs to check credits, query video task status, or download results
• The user mentions dreamina text2video CLI commands explicitly

Do NOT use this skill for:

• Writing or improving video prompts → use jimeng-prompt-text2video
• Image-to-video CLI → use jimeng-cli-image2video
• Image generation CLI → use jimeng-cli-text2image

Core Execution Flow

1. CHECK   → dreamina user_credit                  # Always first — check credits
2. SUBMIT  → dreamina text2video --prompt="..." --duration=N --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 text2video --prompt="..." --duration=5 --ratio=16:9 --model_version=seedance2.0fast_vip --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独立调用 query_result。

示例（智能体自身逻辑）

# Turn 1
terminal: dreamina text2video --prompt="..." --poll=0
→ parse: submit_id = "abc-123"
→ tell user: "已提交视频任务，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: "视频生成中（通常2-15分钟），5秒后再检查"

# Turn 3 (after ~5s) ... repeat until success

How to use this skill

Step 1: Verify prerequisites

1a. Check CLI is installed:

dreamina -h

If not installed:

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

1b. Check login status and credits (mandatory before every generation):

dreamina user_credit

This is the definitive self-check. If it returns JSON with credit info, login and environment are working. Video generation consumes significantly more credits than images — warn user if balance is low.

If login needed:

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

Account management:

dreamina relogin            # Switch accounts
dreamina logout             # Clear credentials (keeps config.toml and tasks.db)

1c. Check config (if login issues):

ls -la ~/.dreamina_cli/

> ⚠️ Note: config.toml, credential.json, tasks.db may not exist — especially in Docker. Auth state can be ephemeral. Use dreamina user_credit as the definitive check.

Step 2: Ensure prompt is ready

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

• Be crafted using jimeng-prompt-text2video methodology
• Include explicit motion description and camera movement
• Be reviewed and approved by the user
• Have duration and ratio recommendations

If no prompt exists, activate jimeng-prompt-text2video first.

Step 3: Map prompt parameters to CLI arguments

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 text2video \
  --prompt="镜头缓缓推进，一个女孩在森林里缓步前行，裙摆摇曳，阳光透过树冠洒在她身上" \
  --duration=8 \
  --ratio=16:9 \
  --model_version=seedance2.0fast_vip \
  --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 generation (defaults, 5s poll fallback):

dreamina text2video --prompt="..." --poll=60
# Defaults: duration=5s, ratio=16:9, model=seedance2.0fast_vip, resolution=720P

High quality (slower):

dreamina text2video \
  --prompt="..." \
  --duration=10 \
  --model_version=seedance2.0 \
  --poll=120

Async (batch or non-blocking):

dreamina text2video --prompt="..." --poll=0
# Save submit_id, query later:
dreamina query_result --submit_id=<id>

Step 5: Handle results

• 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 --poll completes within timeout → result returned immediately
• If --poll times out → returns "querying" intermediate status with submit_id; use 智能体周期查询继续检查
• Report output file path and video details to the user

Step 6: Handle errors

Key Parameters Quick Reference

Poll behavior: --poll=N polls every 1 second for up to N seconds. If task completes within N → result returned. If N seconds elapse → "querying" intermediate result returned with submit_id.

Model Selection Guide

Recommendation (VIP 账户优先)：用户当前账户 VIP 等级为 maestro，应优先使用 VIP 通道以获得更快速度和更高并发：

• seedance2.0fast_vip — 首选：快速迭代，VIP 通道优先
• seedance2.0_vip — 高质量最终输出，VIP 通道优先
• seedance2.0fast — 备用（非 VIP 通道，标准速度）
• seedance2.0 — 备用高质量（非 VIP 通道）

Essential CLI Commands

# Credit & auth
dreamina user_credit                          # Check credit balance
dreamina login                                # Login (OAuth browser)
dreamina login --debug                        # Login with debug output

# Video generation
dreamina text2video --prompt="..." --duration=8 --poll=60
dreamina text2video --prompt="..." --duration=5 --ratio=9:16 --poll=60
dreamina text2video --prompt="..." --model_version=seedance2.0 --poll=120

# Task management
dreamina query_result --submit_id=<id>        # Check task status
dreamina query_result --submit_id=<id> --download_dir=./videos  # 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 "video-project"       # Create session
dreamina session list                         # List sessions

Duration-to-Action Guide

> Match action complexity to duration. A 5s clip with a 3-stage action plan will look rushed.

FAQ (from official documentation)

Q: Login succeeds but generation commands still fail?

A: (1) Verify ~/.dreamina_cli/config.toml exists; (2) Run dreamina user_credit — if this fails, login/config is the issue. Do NOT test generation commands until user_credit works.

Q: Browser login flow is stuck?

A: Use dreamina login --debug for detailed debug output.

Q: Async task shows no final result?

A: Use --poll=N for auto-waiting. If timeout, save submit_id from intermediate result and query: dreamina query_result --submit_id=.

Q: How to switch accounts?

A: dreamina relogin clears login state and starts new flow.

Gotchas

1. Always check credits first — video generation costs significantly more than images. Run dreamina user_credit before every generation session
2. Prompt goes through prompt skill first — don't craft video prompts here. Route to jimeng-prompt-text2video
3. --poll polls every 1 second — --poll=60 means up to 60 polling attempts. Timeout returns "querying" (not failure)
4. Video takes much longer than images — expect 2-15 minutes. Use --poll=120 for high-quality models
5. 720P is the only resolution for Seedance 2.0 — written as 720P (capital P in official docs). 1080p only available with legacy models
6. Match duration to action — don't cram a 3-stage narrative into a 5-second clip
7. Default model is seedance2.0fast — good quality, faster. Use seedance2.0 for final renders
8. Some models need web auth first — if AigcComplianceConfirmationRequired, authorize on dreamina website
9. VIP 账户优先使用 VIP 通道 — seedance2.0fast_vip 和 seedance2.0_vip 有独立 VIP 队列，速度更快、并发更高。账户 VIP 等级为 maestro，应默认为 seedance2.0fast_vip
10. 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 秒调用一次 query_result。禁止 while true; sleep 5; 阻塞终端。智能体需根据 gen_status 做分支判断（继续等/报结果/通知超时）
11. ~/.dreamina_cli/ directory — may contain config.toml, credential.json, tasks.db after native (non-Docker) login. In Docker setups, these files may be absent (auth stored ephemerally). Don't delete

Available Resources