jimeng-cli-image2video — 即梦图生视频 CLI 执行

Execute image-to-video generation via the dreamina CLI. This skill handles 4 sub-commands, routing by input type, credit management, file verification, and result retrieval.

This skill does NOT craft prompts. Prompt crafting is handled by jimeng-prompt-image2video.

When to use this skill

Use this skill when:

• The user has reference image(s) and an approved video prompt
• The user mentions dreamina image2video, frames2video, multiframe2video, multimodal2video
• The user says "让这张图动起来" or "把这些图做成视频"

Do NOT use for:

• Prompt crafting → jimeng-prompt-image2video
• Text-to-video CLI → jimeng-cli-text2video
• Image CLI → jimeng-cli-text2image / jimeng-cli-image2image

Mode Detection

Route to the correct sub-command based on input:

Core Execution Flow

1. CHECK   → dreamina user_credit                    # Always first
2. VERIFY  → Input files exist (ls -la)              # All modes need local files
3. DETECT  → Select correct sub-command              # Based on input count/type
4. SUBMIT  → Submit with --poll=0 (async → get submit_id)  # Then use 5s polling SOP
5. POLL    → Loop every 5s: dreamina query_result --submit_id=<id> until success/fail
6. RETRIEVE → Results saved/downloaded

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：

Step 1 — 提交任务（一次 terminal 调用）：

dreamina image2video --image ./photo.png --prompt="微风吹动头发" --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。

Mode 1: Single Image to Video (dreamina image2video)

Input: 1 image → animate with motion description

# Basic (VIP)
dreamina image2video --image ./photo.png --prompt="微风吹动头发和衣角，眼睛缓慢眨眼，镜头缓缓推近" --model_version=seedance2.0fast_vip --poll=0
# Then follow 5s polling SOP

# With duration
dreamina image2video --image ./photo.png --prompt="云层缓慢飘移，水面波纹扩散" --duration=8 --model_version=seedance2.0fast_vip --poll=0
# Then follow 5s polling SOP

# High quality (VIP)
dreamina image2video --image ./photo.png --prompt="..." --model_version=seedance2.0_vip --poll=0
# Then follow 5s polling SOP

Mode 2: First & Last Frame (dreamina frames2video)

Input: 2 images (start frame + end frame) → transition animation

# Basic (VIP)
dreamina frames2video --first ./start.png --last ./end.png --prompt="花瓣从含苞到完全盛开，外层先展开，内层随后逐层绽放" --model_version=seedance2.0fast_vip --poll=0
# Then follow 5s polling SOP

# With duration (VIP)
dreamina frames2video --first ./morning.png --last ./night.png --prompt="天空从橙色渐变到深蓝，城市灯光渐次亮起" --duration=10 --model_version=seedance2.0fast_vip --poll=0
# Then follow 5s polling SOP

Mode 3: Multi-Frame Story (dreamina multiframe2video)

Input: 2-20 images (storyboard/keyframes) → narrative video

# 2-image
dreamina multiframe2video --images ./a.png,./b.png --prompt="从清晨到正午，光影在建筑表面移动" --poll=0
# Then follow 5s polling SOP

# 3+ images with transition control
dreamina multiframe2video \
  --images ./frame1.png,./frame2.png,./frame3.png,./frame4.png \
  --prompt="故事板叙事：沉思→惊喜→急切→喜悦" \
  --transition-prompt="平滑过渡，镜头跟随情绪节奏" \
  --duration=15 \
  --poll=0
# Then follow 5s polling SOP

> Note: multiframe2video does NOT support --model_version or --video_resolution overrides.

Mode 4: All-Around Reference (dreamina multimodal2video)

Input: images + video + audio mix → synthesized video (strongest mode)

# Images only (VIP)
dreamina multimodal2video \
  --image ./person.png,./scene.png,./style.png \
  --prompt="图一人物特征配图二场景环境和图三的艺术风格。统一光源方向。色调协调。" \
  --model_version=seedance2.0fast_vip \
  --poll=0
# Then follow 5s polling SOP

# Images + audio (VIP)
dreamina multimodal2video \
  --image ./person.png \
  --audio ./voice.mp3 \
  --prompt="人物按照音频节奏自然说话，口型同步，伴随自然眨眼和微表情" \
  --model_version=seedance2.0fast_vip \
  --poll=0
# Then follow 5s polling SOP

# Full combo (images + video + audio, VIP)
dreamina multimodal2video \
  --image ./person.png,./outfit.png,./scene.png \
  --video ./movement_ref.mp4 \
  --audio ./music.mp3 \
  --prompt="人物特征+服装+场景合成。运动节奏参考视频。情绪基调跟随音频。" \
  --duration=12 \
  --model_version=seedance2.0fast_vip \
  --poll=0
# Then follow 5s polling SOP

> At least one of --image, --video, or --audio required. Max: 9 images + 3 videos + 3 audio.

FAQ (from official documentation)

Q: Login succeeds but generation commands fail?

A: Verify ~/.dreamina_cli/config.toml exists. Run dreamina user_credit as definitive self-check.

Q: Browser login stuck?

A: dreamina login --debug for detailed output.

Q: Async task no final result?

A: Use the 5s polling SOP above — it has no timeout cap, so it will keep checking until done.

Q: Switch accounts?

A: dreamina relogin. Clear credentials: dreamina logout.

Gotchas

1. Always check credits first — video consumes far more than images
2. Prompt goes through prompt skill first — route to jimeng-prompt-image2video for all prompt crafting
3. Prompt is incremental — describe only what MOVES, not static content from reference image
4. Mode detection is critical — 1 image vs 2 (frames2video) vs 2+ sequential (multiframe2video) — routing determines the sub-command and parameter names differ (--image vs --images vs --first/--last)
5. Seedance 2.0 = 720P — written as 720P (capital P). 1080P only on legacy 3.0/3.5pro
6. Use 5s polling SOP instead of --poll=N — --poll=N has a timeout; the 5s polling loop has no timeout and uses fewer API calls
7. multimodal2video max inputs — 9 images + 3 videos + 3 audio. Exceeding this errors
8. multiframe2video max 20 images — beyond this not supported
9. Some models need web auth — AigcComplianceConfirmationRequired → authorize on website
10. Parameter names vary by sub-command — image2video uses --image (singular); multiframe2video uses --images (plural); frames2video uses --first/--last
11. 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

Available Resources