@acedatacloud/skills 2026.622.3 → 2026.627.0

package/AGENTS.md

@@ -12,7 +12,6 @@ Skills are located in the `skills/` directory (also mirrored to `.agents/skills/
 - **fish-audio** — Text-to-speech and voice synthesis with Fish Audio
 
 ### AI Image Generation
-- **midjourney-image** — Generate, edit, blend, describe, and upscale images with Midjourney
 - **flux-image** — Generate and edit images with Flux (Black Forest Labs)
 - **seedream-image** — Generate and edit images with ByteDance Seedream
 - **nano-banana-image** — Generate and edit images with Google Gemini (NanoBanana)

package/README.md

@@ -12,7 +12,7 @@
 
 Compatible with **30+ AI coding agents** via the [agentskills.io](https://agentskills.io/) open standard: Claude Code, GitHub Copilot, Gemini CLI, OpenAI Codex, Cursor, Roo Code, Goose, and more.
 
-## Available Skills (30)
+## Available Skills (29)
 
 ### AI Music & Audio
 
@@ -26,7 +26,6 @@ Compatible with **30+ AI coding agents** via the [agentskills.io](https://agents
 
 | Skill | Description |
 |-------|-------------|
-| [midjourney-image](skills/midjourney-image/) | Generate, edit, blend, describe, and upscale images with Midjourney |
 | [flux-image](skills/flux-image/) | Generate and edit images with Flux (Black Forest Labs) |
 | [seedream-image](skills/seedream-image/) | Generate and edit images with ByteDance Seedream |
 | [nano-banana-image](skills/nano-banana-image/) | Generate and edit images with Google Gemini (NanoBanana) |
@@ -92,7 +91,7 @@ export ACEDATACLOUD_API_TOKEN="your-token-here"
 npx skills add acedatacloud/skills --skill '*' --yes
 
 # Install specific skills only
-npx skills add acedatacloud/skills --skill suno-music --skill midjourney-image --yes
+npx skills add acedatacloud/skills --skill suno-music --skill flux-image --yes
 ```
 
 The [`skills` CLI](https://github.com/vercel-labs/skills) auto-detects your agent and installs to the correct path.
@@ -197,7 +196,6 @@ Skills provide **knowledge** (when to use, parameters, gotchas). MCP servers pro
 | Skill | MCP Server | Install | Hosted Endpoint |
 |-------|-----------|---------|-----------------|
 | suno-music | [mcp-suno](https://pypi.org/project/mcp-suno/) | `pip install mcp-suno` | `https://suno.mcp.acedata.cloud/mcp` |
-| midjourney-image | [mcp-midjourney](https://pypi.org/project/mcp-midjourney/) | `pip install mcp-midjourney` | `https://midjourney.mcp.acedata.cloud/mcp` |
 | google-search | [mcp-serp](https://pypi.org/project/mcp-serp/) | `pip install mcp-serp` | `https://serp.mcp.acedata.cloud/mcp` |
 | flux-image | [mcp-flux-pro](https://pypi.org/project/mcp-flux-pro/) | `pip install mcp-flux-pro` | `https://flux.mcp.acedata.cloud/mcp` |
 | luma-video | [mcp-luma](https://pypi.org/project/mcp-luma/) | `pip install mcp-luma` | `https://luma.mcp.acedata.cloud/mcp` |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@acedatacloud/skills",
-  "version": "2026.622.3",
+  "version": "2026.627.0",
   "description": "Agent Skills for AceDataCloud AI services — music, image, video generation, LLM chat, web search. Compatible with Claude Code, GitHub Copilot, Gemini CLI, OpenAI Codex, and 30+ AI coding agents.",
   "keywords": [
     "agent-skills",
@@ -12,7 +12,6 @@
     "openai-codex",
     "ai-tools",
     "suno",
-    "midjourney",
     "flux",
     "luma",
     "sora",

package/skills/_shared/mcp-servers.md

@@ -8,7 +8,6 @@ Each AceDataCloud service has a corresponding MCP server that provides tool-use
 |-------|---------|-----------------|
 | suno-music | `pip install mcp-suno` | `https://suno.mcp.acedata.cloud/mcp` |
 | producer-music | — | — |
-| midjourney-image | `pip install mcp-midjourney` | `https://midjourney.mcp.acedata.cloud/mcp` |
 | google-search | `pip install mcp-serp` | `https://serp.mcp.acedata.cloud/mcp` |
 | flux-image | `pip install mcp-flux-pro` | `https://flux.mcp.acedata.cloud/mcp` |
 | luma-video | `pip install mcp-luma` | `https://luma.mcp.acedata.cloud/mcp` |

package/skills/acedatacloud-api/SKILL.md

@@ -112,7 +112,7 @@ Error response format:
 | Category | Services | Base Path |
 |----------|----------|-----------|
 | **AI Chat** | GPT, Claude, Gemini, Kimi, Grok | `/v1/chat/completions` |
-| **Image Gen** | Midjourney, Flux, Seedream, NanoBanana | `/midjourney/*`, `/flux/*`, etc. |
+| **Image Gen** | Flux, Seedream, NanoBanana | `/flux/*`, `/seedream/*`, etc. |
 | **Video Gen** | Luma, Sora, Veo, Kling, Hailuo, Seedance, Wan | `/luma/*`, `/sora/*`, etc. |
 | **Music Gen** | Suno, Producer, Fish Audio | `/suno/*`, `/producer/*`, `/fish/*` |
 | **Search** | Google Search (web/images/news/maps) | `/serp/*` |

package/skills/bilibili/scripts/bilibili.py

@@ -25,6 +25,7 @@ from __future__ import annotations
 import argparse
 import gzip
 import hashlib
+import html as _html
 import ipaddress
 import json
 import os
@@ -287,7 +288,7 @@ _IMG_SKIP = ("hdslb.com", "bilibili.com", "biliimg.com")
 # matches both HTML <img src="..."> and markdown ![alt](...)
 # <img ... src = "..."> / '...'  (tolerates spaces and either quote style)
 _HTML_IMG = re.compile(r"""(<img\b[^>]*?\bsrc\s*=\s*)(["'])(https?://[^"']+)(\2)""", re.IGNORECASE)
-_MD_IMG = re.compile(r"!\[([^\]]*)\]\((https?://[^)\s]+)\)")
+_MD_IMG = re.compile(r"!\[([^\]]{0,500})\]\((https?://[^)\s]{0,2000})\)")  # bounded: no O(n²) on "![" * N
 UPCOVER = "https://api.bilibili.com/x/article/creative/article/upcover"
 
 
@@ -429,6 +430,121 @@ def rehost_images(jar, csrf, content):
     return content
 
 
+# ── Markdown → HTML (专栏 body is HTML; convert if the caller passed Markdown) ──
+
+# Bounded so a single token can't scan to EOF — keeps these linear (no O(n²)
+# backtracking on malformed input like "![" * N). Real alt < 500 / URL < 2000.
+_IMG_RE = re.compile(r"!\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)")
+_LINK_RE = re.compile(r"(?<!!)\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)")
+
+
+def _looks_like_markdown(s):
+    # Three-way classification (no full parser): block-level HTML ⇒ already an
+    # HTML document → pass through (even if it also contains markdown-looking
+    # text like **bold**); else markdown markers ⇒ render; else inline-only HTML
+    # ⇒ pass through; else plain text ⇒ render (md_to_html just wraps/escapes it).
+    s = s or ""
+    block_html = re.search(
+        r"</?(?:p|div|h[1-6]|ul|ol|li|table|thead|tbody|tr|td|th|blockquote|"
+        r"pre|figure|figcaption|section|article|header|footer|nav|aside|hr|"
+        r"main|details|summary)\b",
+        s, re.I,
+    )
+    if block_html:
+        return False
+    # ^-anchored (re.M) + horizontal-only whitespace + bounded {0,N} repeats so
+    # the scan can't backtrack across newlines or to EOF (no quadratic scanning).
+    has_md = re.search(
+        r"^#{1,6}\s|!\[[^\]]{0,500}\]\([^)]{0,2000}\)|^[ \t]*[-*+]\s|^[ \t]*\d+\.\s"
+        r"|\[[^\]]{0,500}\]\([^)]{0,2000}\)|`[^`]{1,500}`|\*\*[^*]{1,500}\*\*",
+        s, re.M,
+    )
+    if has_md:
+        return True
+    inline_html = re.search(
+        r"</?(?:a|strong|em|b|i|u|s|span|code|br|img|small|mark|sup|sub|"
+        r"video|audio|iframe)\b",
+        s, re.I,
+    )
+    return not inline_html
+
+
+def _attr_url(u):
+    # Strip C0 controls/space FIRST, then allow only http/https/mailto — otherwise
+    # `\x01javascript:` would slip past a literal scheme check yet be re-normalised
+    # to javascript: by the browser. Finally neutralise attribute-breaking quotes.
+    u = re.sub(r"[\x00-\x20\x7f]", "", u or "")
+    m = re.match(r"(?i)([a-z][a-z0-9+.\-]*):", u)
+    if m and m.group(1).lower() not in ("http", "https", "mailto"):
+        return "#"
+    return u.replace('"', "%22").replace("'", "%27")
+
+
+def _alt(s):
+    return (s or "").replace('"', "&quot;").replace("'", "&#x27;")
+
+
+def _inline_md(t):
+    t = _html.escape(t, quote=False).replace("\x00", "")
+    # Stash code spans FIRST so emphasis/link/image markup inside `…` stays
+    # literal (e.g. `**x**` must render as text, not bold).
+    spans = []
+
+    def _stash(m):
+        spans.append("<code>" + m.group(1) + "</code>")
+        return f"\x00{len(spans) - 1}\x00"
+
+    t = re.sub(r"`([^`]+)`", _stash, t)
+    t = _IMG_RE.sub(lambda m: f'<img src="{_attr_url(m.group(2))}" alt="{_alt(m.group(1))}">', t)
+    t = _LINK_RE.sub(lambda m: f'<a href="{_attr_url(m.group(2))}">{m.group(1)}</a>', t)
+    t = re.sub(r"\*\*([^*]+)\*\*", r"<strong>\1</strong>", t)
+    t = re.sub(r"(?<!\*)\*([^*\n]+)\*(?!\*)", r"<em>\1</em>", t)
+    t = re.sub(r"\x00(\d+)\x00", lambda m: spans[int(m.group(1))], t)
+    return t
+
+
+def _md_block_to_html(b):
+    b = b.strip("\n")
+    if not b.strip():
+        return None
+    f = b.lstrip()
+    m = re.match(r"(#{1,6})\s+(.*)", f)
+    if m:
+        lvl = len(m.group(1))
+        return f"<h{lvl}>{_inline_md(m.group(2).strip())}</h{lvl}>"
+    if f.startswith(">"):
+        inner = re.sub(r"^>\s?", "", b, flags=re.M).replace("\n", " ")
+        return "<blockquote><p>" + _inline_md(inner) + "</p></blockquote>"
+    if re.match(r"[-*+]\s+", f):
+        items = [_inline_md(re.sub(r"^[-*+]\s+", "", ln)) for ln in b.split("\n") if ln.strip()]
+        return "<ul>" + "".join(f"<li>{x}</li>" for x in items) + "</ul>"
+    if re.match(r"\d+\.\s+", f):
+        items = [_inline_md(re.sub(r"^\d+\.\s+", "", ln)) for ln in b.split("\n") if ln.strip()]
+        return "<ol>" + "".join(f"<li>{x}</li>" for x in items) + "</ol>"
+    only = re.fullmatch(r"!\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)", f)
+    if only:
+        return f'<img src="{_attr_url(only.group(2))}" alt="{_alt(_html.escape(only.group(1), quote=False))}">'
+    return "<p>" + _inline_md(b.replace("\n", " ").strip()) + "</p>"
+
+
+def md_to_html(src):
+    src = (src or "").strip()
+    out = []
+    # Pull fenced code out FIRST (it may itself contain blank lines) so the
+    # blank-line block splitter below can never tear a ``` … ``` fence apart.
+    for i, part in enumerate(re.split(r"(?ms)^(```.*?(?:\n```[ \t]*$|\Z))", src)):
+        if i % 2 == 1:  # fenced code segment
+            code = re.sub(r"\A```[^\n]*\n?", "", part)
+            code = re.sub(r"\n?```[ \t]*\Z", "", code)
+            out.append("<pre><code>" + _html.escape(code) + "</code></pre>")
+            continue
+        for b in re.split(r"\n[ \t]*\n", part):  # blank-line split (no cross-newline \s*)
+            block = _md_block_to_html(b)
+            if block:
+                out.append(block)
+    return "\n".join(out)
+
+
 def cmd_publish(jar, args):
     if not args.title:
         die("--title is required")
@@ -461,6 +577,11 @@ def cmd_publish(jar, args):
     if not args.no_rehost_images:
         content = rehost_images(jar, csrf, content)
 
+    # 专栏 content is HTML; if the caller passed Markdown, render it (else the
+    # article shows literal `##`/`![]()` and runs together as one paragraph).
+    if _looks_like_markdown(content):
+        content = md_to_html(content)
+
     ref = "https://member.bilibili.com/"
     base = {
         "title": args.title, "content": content, "csrf": csrf,

package/skills/csdn/scripts/csdn.py

@@ -28,6 +28,7 @@ import base64
 import gzip
 import hashlib
 import hmac
+import html as _html
 import ipaddress
 import json
 import mimetypes
@@ -286,7 +287,7 @@ def cmd_article(jar, args):
 # Signed signature → Huawei OBS multipart → csdnimg.cn URL. Mirrors our
 # PlatformPublisher csdn.py flow.
 
-_MD_IMG = re.compile(r"!\[([^\]]*)\]\((https?://[^)\s]+)\)")
+_MD_IMG = re.compile(r"!\[([^\]]{0,500})\]\((https?://[^)\s]{0,2000})\)")  # bounded: no O(n²) on "![" * N
 _IMG_SKIP = ("csdnimg.cn", "csdn.net")
 
 
@@ -416,6 +417,126 @@ def rehost_images(jar, content):
     return _MD_IMG.sub(repl, content)
 
 
+# ── Markdown → HTML (CSDN renders the `content` HTML field, not the source) ──
+
+# Bounded so a single token can't scan to EOF — keeps these linear (no O(n²)
+# backtracking on malformed input like "![" * N). Real alt < 500 / URL < 2000.
+_IMG_RE = re.compile(r"!\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)")
+_LINK_RE = re.compile(r"(?<!!)\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)")
+
+
+def _attr_url(u):
+    # Strip C0 controls/space FIRST, then allow only http/https/mailto — otherwise
+    # `\x01javascript:` would slip past a literal scheme check yet be re-normalised
+    # to javascript: by the browser. Finally neutralise attribute-breaking quotes.
+    u = re.sub(r"[\x00-\x20\x7f]", "", u or "")
+    m = re.match(r"(?i)([a-z][a-z0-9+.\-]*):", u)
+    if m and m.group(1).lower() not in ("http", "https", "mailto"):
+        return "#"
+    return u.replace('"', "%22").replace("'", "%27")
+
+
+def _alt(s):
+    # alt text is already &/</>-escaped by the line escape; only the attribute
+    # quote can still break out, so neutralise it.
+    return (s or "").replace('"', "&quot;").replace("'", "&#x27;")
+
+
+def _inline_md(t):
+    t = _html.escape(t, quote=False).replace("\x00", "")
+    # Stash code spans FIRST so emphasis/link/image markup inside `…` stays
+    # literal (e.g. `**x**` must render as text, not bold).
+    spans = []
+
+    def _stash(m):
+        spans.append("<code>" + m.group(1) + "</code>")
+        return f"\x00{len(spans) - 1}\x00"
+
+    t = re.sub(r"`([^`]+)`", _stash, t)
+    t = _IMG_RE.sub(lambda m: f'<img src="{_attr_url(m.group(2))}" alt="{_alt(m.group(1))}">', t)
+    t = _LINK_RE.sub(lambda m: f'<a href="{_attr_url(m.group(2))}">{m.group(1)}</a>', t)
+    t = re.sub(r"\*\*([^*]+)\*\*", r"<strong>\1</strong>", t)
+    t = re.sub(r"(?<!\*)\*([^*\n]+)\*(?!\*)", r"<em>\1</em>", t)
+    t = re.sub(r"\x00(\d+)\x00", lambda m: spans[int(m.group(1))], t)
+    return t
+
+
+def _md_block_to_html(b):
+    b = b.strip("\n")
+    if not b.strip():
+        return None
+    f = b.lstrip()
+    m = re.match(r"(#{1,6})\s+(.*)", f)
+    if m:
+        lvl = len(m.group(1))
+        return f"<h{lvl}>{_inline_md(m.group(2).strip())}</h{lvl}>"
+    if f.startswith(">"):
+        inner = re.sub(r"^>\s?", "", b, flags=re.M).replace("\n", " ")
+        return "<blockquote><p>" + _inline_md(inner) + "</p></blockquote>"
+    if re.match(r"[-*+]\s+", f):
+        items = [_inline_md(re.sub(r"^[-*+]\s+", "", ln)) for ln in b.split("\n") if ln.strip()]
+        return "<ul>" + "".join(f"<li>{x}</li>" for x in items) + "</ul>"
+    if re.match(r"\d+\.\s+", f):
+        items = [_inline_md(re.sub(r"^\d+\.\s+", "", ln)) for ln in b.split("\n") if ln.strip()]
+        return "<ol>" + "".join(f"<li>{x}</li>" for x in items) + "</ol>"
+    only = re.fullmatch(r"!\[([^\]]{0,500})\]\(([^)\s]{0,2000})\)", f)
+    if only:
+        return f'<img src="{_attr_url(only.group(2))}" alt="{_alt(_html.escape(only.group(1), quote=False))}">'
+    return "<p>" + _inline_md(b.replace("\n", " ").strip()) + "</p>"
+
+
+def md_to_html(src):
+    """Minimal stdlib Markdown→HTML for the platforms whose body is HTML.
+    Covers headings, paragraphs, standalone images, links, bold/italic/code,
+    fenced code, blockquotes, and ordered/unordered lists."""
+    src = (src or "").strip()
+    out = []
+    # Pull fenced code out FIRST (it may itself contain blank lines) so the
+    # blank-line block splitter below can never tear a ``` … ``` fence apart.
+    for i, part in enumerate(re.split(r"(?ms)^(```.*?(?:\n```[ \t]*$|\Z))", src)):
+        if i % 2 == 1:  # fenced code segment
+            code = re.sub(r"\A```[^\n]*\n?", "", part)
+            code = re.sub(r"\n?```[ \t]*\Z", "", code)
+            out.append("<pre><code>" + _html.escape(code) + "</code></pre>")
+            continue
+        for b in re.split(r"\n[ \t]*\n", part):  # blank-line split (no cross-newline \s*)
+            block = _md_block_to_html(b)
+            if block:
+                out.append(block)
+    return "\n".join(out)
+
+
+def _looks_like_markdown(s):
+    # Three-way classification (no full parser): block-level HTML ⇒ already an
+    # HTML document → pass through (even if it also contains markdown-looking
+    # text like **bold**); else markdown markers ⇒ render; else inline-only HTML
+    # ⇒ pass through; else plain text ⇒ render (md_to_html just wraps/escapes it).
+    s = s or ""
+    block_html = re.search(
+        r"</?(?:p|div|h[1-6]|ul|ol|li|table|thead|tbody|tr|td|th|blockquote|"
+        r"pre|figure|figcaption|section|article|header|footer|nav|aside|hr|"
+        r"main|details|summary)\b",
+        s, re.I,
+    )
+    if block_html:
+        return False
+    # ^-anchored (re.M) + horizontal-only whitespace + bounded {0,N} repeats so
+    # the scan can't backtrack across newlines or to EOF (no quadratic scanning).
+    has_md = re.search(
+        r"^#{1,6}\s|!\[[^\]]{0,500}\]\([^)]{0,2000}\)|^[ \t]*[-*+]\s|^[ \t]*\d+\.\s"
+        r"|\[[^\]]{0,500}\]\([^)]{0,2000}\)|`[^`]{1,500}`|\*\*[^*]{1,500}\*\*",
+        s, re.M,
+    )
+    if has_md:
+        return True
+    inline_html = re.search(
+        r"</?(?:a|strong|em|b|i|u|s|span|code|br|img|small|mark|sup|sub|"
+        r"video|audio|iframe)\b",
+        s, re.I,
+    )
+    return not inline_html
+
+
 def cmd_publish(jar, args):
     if not args.title:
         die("--title is required")
@@ -450,8 +571,12 @@ def cmd_publish(jar, args):
     status_code = 2 if args.draft_only else 0
     body = {
         "title": args.title,
+        # markdowncontent = the editor source; content = the RENDERED HTML CSDN
+        # actually displays. Sending raw markdown as `content` shows literal
+        # `##`/`![]()` and collapses newlines — it MUST be HTML. Already-HTML
+        # callers are passed through untouched (don't double-escape).
         "markdowncontent": content,
-        "content": content,
+        "content": md_to_html(content) if _looks_like_markdown(content) else content,
         "readType": "public",
         "status": status_code,
         "categories": "",

package/skills/midjourney-image/SKILL.md

@@ -1,170 +0,0 @@
----
-name: midjourney-image
-description: Generate, edit, blend, upscale, and describe images with Midjourney via AceDataCloud API. Use when creating AI images from text prompts, editing existing images, generating 2x2 grids, upscaling, creating variations, blending multiple images, reverse-prompting from images, or generating video from images. Supports versions 5.2 through 8.
-license: Apache-2.0
-metadata:
-  author: acedatacloud
-  version: "1.0"
-compatibility: Requires ACEDATACLOUD_API_TOKEN in .env file (see _shared/authentication.md). Optionally pair with mcp-midjourney for tool-use.
----
-
-# Midjourney Image Generation
-
-Generate and manipulate AI images through AceDataCloud's Midjourney API.
-
-> **Setup:** See [authentication](../_shared/authentication.md) for token setup.
-
-## Quick Start — Generate an Image
-
-```bash
-curl -X POST https://api.acedata.cloud/midjourney/imagine \
-  -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
-  -H "Content-Type: application/json" \
-  -d '{"prompt": "a futuristic city at sunset, cyberpunk style --ar 16:9", "callback_url": "https://api.acedata.cloud/health"}'
-```
-
-> **Async:** See [async task polling](../_shared/async-tasks.md). Poll via `POST /midjourney/tasks` with `{"id": "..."}`.
-
-## Generation Modes
-
-| Mode | Speed | Cost | Best For |
-|------|-------|------|----------|
-| `fast` | Fast | Standard | Most tasks (default) |
-| `relax` | Slow | Cheaper | Batch generation |
-| `turbo` | Fastest | Premium | Time-sensitive work |
-
-## Midjourney Versions
-
-| Version | Notes |
-|---------|-------|
-| `8` | Latest, best quality |
-| `7` | Great quality, fast |
-| `6.1` | Stable, well-tested |
-| `6` | Previous generation |
-| `5.2` | Legacy |
-
-## Core Workflows
-
-### 1. Generate Images (Imagine)
-
-Creates a 2x2 grid of 4 image variations.
-
-```json
-POST /midjourney/imagine
-{
-  "prompt": "a serene mountain lake at dawn, photorealistic --ar 16:9 --v 7",
-  "mode": "fast",
-  "translation": true,
-  "split_images": true
-}
-```
-
-Set `translation: true` to auto-translate non-English prompts. Set `split_images: true` to get individual images besides the grid.
-
-### 2. Upscale / Vary / Pan / Zoom
-
-After generating a grid, use transform actions on individual images:
-
-```json
-POST /midjourney/imagine
-{
-  "action": "upscale1",
-  "image_id": "grid-image-id"
-}
-```
-
-**Available actions:**
-- `upscale1`–`upscale4`: Upscale individual quadrant
-- `variation1`–`variation4`: Create variation of a quadrant
-- `variation_subtle` / `variation_strong`: Subtle/strong variation of full image
-- `reroll`: Re-generate with same prompt
-- `zoom_out_2x` / `zoom_out_1_5x`: Zoom out
-- `pan_left` / `pan_right` / `pan_up` / `pan_down`: Extend canvas
-
-### 3. Edit an Image
-
-Modify an existing image using a text prompt, optionally with a mask.
-
-```json
-POST /midjourney/edits
-{
-  "image_url": "https://example.com/photo.jpg",
-  "prompt": "add a rainbow in the sky",
-  "mode": "fast"
-}
-```
-
-### 4. Blend Images
-
-Combine 2–5 images into a new composition.
-
-```json
-POST /midjourney/imagine
-{
-  "action": "blend",
-  "image_urls": [
-    "https://example.com/image1.jpg",
-    "https://example.com/image2.jpg"
-  ]
-}
-```
-
-### 5. Describe an Image (Reverse Prompt)
-
-Get AI-generated text descriptions of an image (returns 4 options).
-
-```json
-POST /midjourney/describe
-{"image_url": "https://example.com/photo.jpg"}
-```
-
-### 6. Generate Video from Image
-
-Create a video with a reference image and text prompt.
-
-```json
-POST /midjourney/videos
-{
-  "image_url": "https://example.com/photo.jpg",
-  "prompt": "the city comes alive with moving traffic",
-  "resolution": "720p"
-}
-```
-
-## Prompt Parameters
-
-Append these to your prompt text:
-
-| Parameter | Example | Description |
-|-----------|---------|-------------|
-| `--ar` | `--ar 16:9` | Aspect ratio |
-| `--v` | `--v 7` | Midjourney version |
-| `--q` | `--q 2` | Quality (0.25, 0.5, 1, 2) |
-| `--s` | `--s 750` | Stylization (0–1000) |
-| `--c` | `--c 50` | Chaos/variety (0–100) |
-| `--no` | `--no text, watermark` | Negative prompt |
-| `--seed` | `--seed 12345` | Reproducible generation |
-
-## API-Level Parameters (Billing Impact)
-
-These top-level fields on `POST /midjourney/imagine` affect billing and are separate from inline prompt parameters:
-
-| Parameter | Type | Description |
-|-----------|------|-------------|
-| `version` | string | Midjourney version (`"8"`, `"7"`, `"6.1"`, etc.) — used for billing calculation |
-| `hd` | boolean | Enable HD 2K resolution (V8 only) — costs 4× GPU time |
-| `quality` | string | Quality level: `".25"`, `".5"`, `"1"`, `"2"`, `"4"` — quality `"4"` is V8 only and costs 4× GPU time |
-| `style_reference` | boolean | Whether prompt uses `--sref` style references (V8: costs 4× GPU time) |
-| `moodboard` | boolean | Whether prompt uses moodboard image references (V8: costs 4× GPU time) |
-
-## Gotchas
-
-- Imagine returns a **2x2 grid** — use upscale/variation actions to work with individual images
-- Use `split_images: true` to also receive individual cropped images alongside the grid
-- Prompt parameters (`--ar`, `--v`, etc.) go **inside the prompt string**, not as separate fields
-- `translation: true` auto-translates Chinese/other languages to English before sending to Midjourney
-- Video generation requires a reference `image_url` — it cannot generate from text alone
-- Available transform actions depend on the image — check `available_actions` in the response
-- Get the seed with `POST /midjourney/seed` using the image_id for reproducible results
-
-> **MCP:** `pip install mcp-midjourney` | Hosted: `https://midjourney.mcp.acedata.cloud/mcp` | See [all MCP servers](../_shared/mcp-servers.md)