IMA Image AI — Image Generator

Image Generation Modes

Visual Consistency (IMPORTANT)

If user mentions "same character", "series", "multi-shot", or continues from a previous generation:

• Do NOT use text_to_image (will produce different-looking results)
• Use image_to_image with previous result as reference
• Read ima-knowledge-ai/references/visual-consistency.md if available

Pre-Check: Knowledge Base

If ima-knowledge-ai is installed, read before generating:

1. ima-knowledge-ai/references/visual-consistency.md — if multi-shot or character continuity needed

If not installed: use this SKILL's model table and defaults.

Model Selection Priority

1. User preference (if explicitly stated) → highest priority
2. ima-knowledge-ai recommendation (if installed)
3. Fallback defaults:

User Input Parsing

Size/Resolution: 512/1K/2K/4K → via attribute_id for Nano Banana series Aspect ratio: 16:9/9:16/4:3/3:4/1:1 → SeeDream 4.5 or Nano Banana series (Midjourney only 1:1) Budget: 最便宜→Nano Banana2 (4pts) · 最好→Nano Banana Pro (4K) or SeeDream 4.5

Script Usage

# Text to image
python3 {baseDir}/scripts/ima_image_create.py \
  --api-key $IMA_API_KEY \
  --task-type text_to_image \
  --model-id doubao-seedream-4.5 \
  --prompt "a cute puppy running on grass" \
  --user-id {user_id} \
  --output-json

# Image to image (accepts URLs and local file paths)
python3 {baseDir}/scripts/ima_image_create.py \
  --api-key $IMA_API_KEY \
  --task-type image_to_image \
  --model-id doubao-seedream-4.5 \
  --prompt "turn into oil painting style" \
  --input-images https://example.com/photo.jpg \
  --user-id {user_id} \
  --output-json

# With aspect ratio (SeeDream 4.5 or Nano Banana)
python3 {baseDir}/scripts/ima_image_create.py \
  --api-key $IMA_API_KEY \
  --task-type text_to_image \
  --model-id doubao-seedream-4.5 \
  --prompt "beautiful landscape" \
  --extra-params '{"aspect_ratio": "16:9"}' \
  --user-id {user_id} \
  --output-json

Sending Results to User

# ✅ CORRECT: Use remote URL directly for inline image display
message(action="send", media=image_url, caption="✅ 图片生成成功！\n• 模型：[Name]\n• 耗时：[X]s\n• 积分：[N pts]\n\n🔗 原始链接：[url]")

# ❌ WRONG: Never download to local file (shows as attachment, not rendered)

UX Protocol (Brief)

1. Pre-generation: "🎨 开始生成图片… 模型：[Name]，预计[X~Y]秒，消耗[N]积分"
2. Progress: Every 15-30s: "⏳ 正在生成中… [P]%" (cap at 95% until API returns success)
3. Success: Send image via media=image_url + include link in caption for sharing
4. Failure: Natural language error + suggest alternative models. Never show technical errors to users. See SKILL-DETAIL.md for full error translation table.

Never say to users: script names, API endpoints, attribute_id, technical parameter names. Only: model name · time · credits · result · status.

Midjourney Limitations

Midjourney has fixed 1:1 aspect ratio (1024×1024 only). If user asks for 16:9 etc. with "MJ", recommend SeeDream 4.5 or Nano Banana series instead.

Environment

Base URL: https://api.imastudio.com Headers: Authorization: Bearer $IMA_API_KEY · x-app-source: ima_skills · x_app_language: en

Core Flow

1. GET /open/v1/product/list?app=ima&platform=web&category=<task_type> → get attribute_id, credit, model_version, form_config
2. [image_to_image only] Upload images or pass local paths to script
3. POST /open/v1/tasks/create → get task_id
4. POST /open/v1/tasks/detail → poll every 3-5s until resource_status==1

MANDATORY: Always query product list first. attribute_id is required — if 0 or missing, task fails.

User Preference Memory

Storage: ~/.openclaw/memory/ima_prefs.json

• Save when user explicitly says "用XXX" / "默认用XXX" / "always use XXX"
• Clear when user says "用最好的" / "推荐一个" / "自动选择"
• Never save auto-selected or fallback models as preferences

Model Capabilities