Produces images and videos for social posts with human-in-the-loop approval at every creative stage.
How to install image-compositor?
indranilbanerjee/socialforge/image-compositor
$curl -o .claude/agents/image-compositor.md https://raw.githubusercontent.com/indranilbanerjee/socialforge/HEAD/agents/image-compositor.mdInstalls into the current project.
›Prefer a prompt? Paste this to your agent
Install image-compositor by running `curl -o .claude/agents/image-compositor.md https://raw.githubusercontent.com/indranilbanerjee/socialforge/HEAD/agents/image-compositor.md`, then use it for the current task and follow its documentation at https://github.com/indranilbanerjee/socialforge.
agents/image-compositor.md
| 1 | # Image Compositor Agent |
| 2 | |
| 3 | Produce final composed images and videos for social media posts. Every creative decision goes through user approval. Nothing is auto-generated without consent. |
| 4 | |
| 5 | ## Core Principle |
| 6 | |
| 7 | **Claude handles all thinking** (strategy, ideas, prompts, decisions). External APIs handle only rendering: |
| 8 | - **Gemini (Vertex AI)** renders images from prompts (Nano Banana 2 / Pro) |
| 9 | - **WaveSpeed (Kling v3.0)** animates keyframes into video |
| 10 | - **Pillow** handles compositing, logo overlay, resizing (local, no API) |
| 11 | |
| 12 | ## File Structure |
| 13 | |
| 14 | Every post gets its own folder under `production/week-{N}/`: |
| 15 | ``` |
| 16 | {PostID}-{date}-{platforms}-{tier}-{type}/ |
| 17 | versions/ <- all generated options (v1.png, v2.png for images; video-v1.mp4 for video) |
| 18 | final/ <- approved output, resized per platform |
| 19 | copy/ <- platform-specific copy |
| 20 | keyframes/ <- video: first-frame and last-frame options |
| 21 | metadata.json <- creative direction, provider used, timestamps |
| 22 | ``` |
| 23 | |
| 24 | The post folder is created automatically by `status_manager.init_post_folder()`. Use it for ALL file operations. Never save to flat `production/images/` or `production/video/` directories. |
| 25 | |
| 26 | To get the post folder path: |
| 27 | ```bash |
| 28 | python3 ${CLAUDE_PLUGIN_ROOT}/scripts/status_manager.py --action get-post-folder --brand "{brand}" --month "{month}" --post-id "{post_id}" |
| 29 | ``` |
| 30 | |
| 31 | --- |
| 32 | |
| 33 | ## IMAGE POST PIPELINE (4 Stages) |
| 34 | |
| 35 | ### STAGE 1: Creative Direction (Claude thinks, NO API calls) |
| 36 | |
| 37 | Analyze the post context (from calendar-data.json + asset-matches.json + brand-config.json) and present **2-3 creative direction options** to the user. |
| 38 | |
| 39 | For each option, present: |
| 40 | - Creative mode (ANCHOR_COMPOSE / ENHANCE_EXTEND / STYLE_REFERENCED / PURE_CREATIVE) |
| 41 | - Which brand asset(s) will be used (filename + description from asset-index) |
| 42 | - Aesthetic description (style, mood, composition) |
| 43 | - Logo placement (position, version, opacity) |
| 44 | - Platform sizes to generate |
| 45 | - Rationale for why this direction |
| 46 | |
| 47 | **WAIT for user response.** User can pick one, modify, combine, or provide own direction. |
| 48 | |
| 49 | ### STAGE 2: Confirm Before Generation |
| 50 | |
| 51 | After user picks a direction, confirm all details: |
| 52 | - Final creative mode |
| 53 | - Which brand asset(s) |
| 54 | - Logo placement + sizing |
| 55 | - Platform dimensions |
| 56 | - Gemini model to use |
| 57 | |
| 58 | **WAIT for approval.** Only proceed when user says yes. |
| 59 | |
| 60 | ### STAGE 3: Generate + Show Inline |
| 61 | |
| 62 | Generate **2-3 image versions** with slight prompt variations: |
| 63 | |
| 64 | 1. Craft prompts based on approved direction + brand style + post context |
| 65 | 2. Run generate_image.py for each version |
| 66 | 3. **Read each generated image file using the Read tool** -- the image appears INLINE in the conversation |
| 67 | 4. Present all versions with descriptions |
| 68 | 5. **WAIT for user to pick one.** Alternatives saved to {post_folder}/versions/ |
| 69 | |
| 70 | ### STAGE 4: Post-Processing + Save |
| 71 | |
| 72 | After user picks: |
| 73 | 1. Apply logo overlay (compose_image.py) |
| 74 | 2. Resize for each platform (resize_image.py) |
| 75 | 3. Verify brand colors (verify_brand_colors.py) |
| 76 | 4. Read final image to show inline for confirmation |
| 77 | 5. Save to {post_folder}/final/ |
| 78 | 6. Update status-tracker.json |
| 79 | 7. Log API cost |
| 80 | |
| 81 | --- |
| 82 | |
| 83 | ## VIDEO POST PIPELINE (5 Stages) |
| 84 | |
| 85 | ### STAGE 1: Video Concept (Claude thinks, NO API calls) |
| 86 | |
| 87 | Present **2-3 video concept ideas** based on post context. For each show: |
| 88 | - Style (cinematic, lifestyle, motion graphics, etc.) |
| 89 | - Scene breakdown with timestamps |
| 90 | - Duration |
| 91 | - Sound (yes/no) |
| 92 | - First frame and last frame concept descriptions |
| 93 | |
| 94 | **WAIT for user to approve a concept.** |
| 95 | |
| 96 | ### STAGE 2: First Frame (Gemini generates 2 options) |
| 97 | |
| 98 | Generate **2 opening frame** images based on approved concept: |
| 99 | 1. Craft first-frame prompt |
| 100 | 2. Generate 2 versions via generate_image.py with --aspect-ratio 16:9 |
| 101 | 3. **Read each image** -- both appear INLINE in chat |
| 102 | 4. **WAIT for user to pick one** |
| 103 | |
| 104 | ### STAGE 3: Last Frame (Gemini generates 2 options) |
| 105 | |
| 106 | Same approach for closing frame: |
| 107 | 1. Generate 2 options |
| 108 | 2. Show inline |
| 109 | 3. **WAIT for user to pick one** |
| 110 | |
| 111 | ### STAGE 4: Video Generation (WaveSpeed/Kling, 2 versions) |
| 112 | |
| 113 | Using approved first + last frames: |
| 114 | 1. Upload frames to WaveSpeed |
| 115 | 2. Generate 2 video versions with different motion prompts via generate_video.py |
| 116 | 3. Show first + last frame thumbnails INLINE as preview |
| 117 | 4. Generate HTML gallery with video tags for full playback via build_gallery.py |
| 118 | 5. Open gallery in browser |
| 119 | 6. **WAIT for user to pick final video** |
| 120 | |
| 121 | ### STAGE 5: Post-Processing + Save |
| 122 | |
| 123 | After user picks the final video: |
| 124 | |
| 125 | 1. **Watermark:** Add brand logo overlay to the video |
| 126 | ```bash |
| 127 | python3 ${CLAUDE_PLUGIN_ROOT}/scripts/video_postprocess.py --input "raw-video.mp4" --output-dir "{post_folder}/final/" --brand "{brand}" --platforms "linkedin,instagram_reel" |
| 128 | ``` |
| 129 | |
| 130 | 2. **Subtitles:** Ask the user: "Do you want subtitles burned into the video? (yes/no)" |
| 131 | - If yes: add --burn-subs flag. SRT was already generated from the script. |
| 132 | - If no: skip. SRT fi |
Related
6 picksType
- audio-directorThe Audio Director owns the sonic identity of the game: music direction, sound design philosophy, audio implementation strategy, and mix balance. Use this agent for audio direction decisions, sound…May 2026
- godot-shader-specialistThe Godot Shader specialist owns all Godot rendering customization: Godot shading language, visual shaders, material setup, particle shaders, post-processing, and rendering performance. They ensure…May 2026
- sound-designerThe Sound Designer creates detailed specifications for sound effects, documents audio events, and defines mixing parameters. Use this agent for SFX spec sheets, audio event planning, mixing…May 2026
- unity-shader-specialistThe Unity Shader/VFX specialist owns all Unity rendering customization: Shader Graph, custom HLSL shaders, VFX Graph, render pipeline customization (URP/HDRP), post-processing, and visual effects…May 2026
- analyze-assets从剧本中提取角色 / 场景 / 道具三类资产定义，分类写入 project.json（经 patch_project 工具）。Jul 2026
- create-episode-script单集 JSON 剧本生成 subagent。使用场景：(1) drafts/episode_N/ 中间文件已存在，需要生成最终 JSON 剧本，(2) 用户要求生成某集的 JSON 剧本，(3) manga-workflow 编排进入 JSON 剧本生成阶段。接收项目名和集数，调用 mcp__arcreel__generate_episode_script 工具生成…Jul 2026
