IMA Video AI — Video Generator

Aliases: 万/Wan → Wan 2.6 · 可灵O1 → kling-video-o1 · 海螺2.3 → MiniMax-Hailuo-2.3 · Veo → veo-3.1-generate-preview · Ima Sevio 1.0 → ima-pro · Ima Sevio 1.0-Fast → ima-pro-fast

Use --list-models --task-type <text_to_video|image_to_video|...> when unsure.

Video Modes (task_type)

If ima-knowledge-ai is installed, read references/video-modes.md and visual-consistency.md when user needs continuity across shots or references a previous image.

Visual Consistency (IMPORTANT)

• Text-only generation cannot reliably keep the same character/scene across runs.
• For “同一个角色 / 续集 / 分镜”: use image modes with the prior result (or reference image), not text_to_video alone.

Model Selection Priority

1. User explicit preference (saved in ima_prefs.json only when user clearly picks a model)
2. ima-knowledge-ai (if installed)
3. Fallback defaults (see SKILL-DETAIL.md for full table)

Script Usage

# Text to video
python3 {baseDir}/scripts/ima_video_create.py \
  --api-key $IMA_API_KEY \
  --task-type text_to_video \
  --model-id wan2.6-t2v \
  --prompt "a puppy runs across a sunny meadow, cinematic" \
  --user-id {user_id} \
  --output-json

# Image to video (URLs or local paths; script uploads locals)
python3 {baseDir}/scripts/ima_video_create.py \
  --api-key $IMA_API_KEY \
  --task-type image_to_video \
  --model-id wan2.6-i2v \
  --prompt "camera slowly zooms in" \
  --input-images https://example.com/photo.jpg \
  --user-id {user_id} \
  --output-json

# First–last frame
python3 {baseDir}/scripts/ima_video_create.py \
  --api-key $IMA_API_KEY \
  --task-type first_last_frame_to_video \
  --model-id kling-video-o1 \
  --prompt "smooth transition" \
  --input-images https://example.com/first.jpg https://example.com/last.jpg \
  --user-id {user_id} \
  --output-json

Sending Results to User

video_url = json_output["url"]
message(action="send", media=video_url, caption="✅ 视频生成成功！\n• 模型：[Name]\n• 耗时：[X]s\n• 积分：[N pts]\n\n🔗 原始链接：[url]")

Never download to a local path for media — use the HTTPS URL from the API.

UX Protocol (Brief)

1. Pre-generation: model name · estimated time range · credits
2. Progress: poll ~8s; update user every 30–60s; cap % at 95 until done
3. Success: send media=video_url, then optional text with link for copy/share
4. Failure: plain-language reason + 1–2 alternate models — never raw API errors. Full error table in SKILL-DETAIL.md.

Never say to users: script names, endpoints, attribute_id, internal field names.

Environment

Base URL: https://api.imastudio.com
Headers: Authorization: Bearer $IMA_API_KEY · x-app-source: ima_skills · x_app_language: en
Image upload (when needed): imapi.liveme.com (same provider; see SKILL-DETAIL.md).

Core Flow

1. GET /open/v1/product/list?app=ima&platform=web&category=<task_type> → attribute_id, credit, model_version, form_config
2. Image tasks: ensure public HTTPS URLs (script handles local upload)
3. POST /open/v1/tasks/create → task_id
4. POST /open/v1/tasks/detail → poll every 8s, timeout up to ~40 min as documented in detail file

MANDATORY: Always query product list first; wrong or stale attribute_id causes create failures.

User Preference Memory

Path: ~/.openclaw/memory/ima_prefs.json
Save when user explicitly chooses a default model; clear when they ask for “推荐 / 自动 / 最好的”. Do not save auto-picked models as preference.

Polling & Timing (summary)

Sora 2 Pro (brief)

Strict safety: avoid people, celebrities, and IP in prompts; prefer landscapes/abstract/safe subjects — details in SKILL-DETAIL.md.