8000 flowkit/skills at main · nlminh/flowkit · GitHub
[go: up one dir, main page]

Skip to content

Latest commit

 

History

History
 
 

skills

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Skills — Flow Kit

Workflow skills for AI agents and humans. Each skill is a step-by-step recipe.

Pipeline (run in order)

# Skill File Description
1 fk-create-project fk-create-project.md Create project + entities + video + scenes
2 fk-gen-refs fk-gen-refs.md Generate reference images for all entities
3 fk-gen-images fk-gen-images.md Generate scene images with character refs
4 fk-gen-videos fk-gen-videos.md Generate videos from scene images
5 fk-concat fk-concat.md Download + merge all scene videos

Advanced Video

Skill File Description
fk-gen-chain-videos fk-gen-chain-videos.md Auto start+end frame chaining for smooth transitions
fk-insert-scene fk-insert-scene.md Multi-angle shots, cutaways, close-ups
fk-creative-mix fk-creative-mix.md Analyze story + suggest all techniques combined

Reference

Skill File Description
fk-camera-guide fk-camera-guide.md Camera angles, movements, lighting, DOF for cinematic video prompts

Utilities

Skill File Description
fk-status fk-status.md Full project dashboard + next action
fk-fix-uuids fk-fix-uuids.md Repair any CAMS... media_ids to UUID format

Cross-Tool Compatibility

Skills live in skills/ — the single source of truth. Run setup.py to generate configs for your AI tool:

python setup.py           # Interactive setup
python setup.py --tool all  # Generate for all tools
python setup.py sync      # Re-sync after adding skills
python setup.py clean     # Remove generated configs
Tool Generated Config Instruction File Invocation
Claude Code .claude/commands/fk:<name>.md (stubs) CLAUDE.md (committed) /fk:<name>
Gemini CLI .gemini/commands/fk/<name>.toml GEMINI.md (generated) /fk:<name>
Codex CLI — AGENTS.md (generated) Read skills/fk:<name>.md

Adding a new skill: Create skills/fk:<name>.md, then run python setup.py sync.


Full Example: "Luna the Space Cat" (3 scenes, vertical)

This walkthrough shows exactly how an AI agent uses skills to go from idea to final video.

1. /fk-create-project

Agent asks: What's your project? Characters, locations, key objects, scenes, orientation?

User says: "Luna is a small white cat astronaut exploring a candy planet. 3 scenes: she lands on the planet, discovers a chocolate river, and plants a flag on a gummy bear mountain. Vertical, Pixar 3D."

Agent thinks: Break the story into visual elements that repeat across scenes:

Element entity_type Why it needs a reference description (appearance only)
Luna character Main character in all 3 scenes Small white cat with big blue eyes, wearing a tiny orange space suit with a round glass helmet. Fluffy tail sticking out. Pixar-style 3D.
Candy Planet Surface location Ground/environment in all scenes Alien planet surface made of colorful hard candies, lollipop trees, cotton candy clouds, pastel pink and purple sky. Pixar-style 3D.
Chocolate River location Key location in scene 2 Wide flowing river of dark melted chocolate with marshmallow rocks along the banks, steam rising, rich brown tones. Pixar-style 3D.
Gummy Bear Mountain location Key location in scene 3 Tall mountain made of giant translucent gummy bears stacked together, glowing from inside, rainbow colors. Pixar-style 3D.

Agent writes scene prompts (action + environment, NO character appearance):

Scene Prompt character_names Chain
1 "Luna steps out of a small rocket onto Candy Planet Surface. First footprint in candy dust. Wide shot, dramatic landing moment. Pixar 3D, cinematic." Luna, Candy Planet Surface ROOT
2 "Luna kneels at the edge of Chocolate River on Candy Planet Surface, dipping a paw in. Surprised expression, licking the paw. Warm lighting reflected off chocolate. Pixar 3D." Luna, Candy Planet Surface, Chocolate River CONTINUATION ← 1
3 "Luna plants a small flag on top of Gummy Bear Mountain, triumphant pose. Candy Planet Surface stretching to the horizon below. Epic wide angle, sunset glow through gummy bears. Pixar 3D." Luna, Candy Planet Surface, Gummy Bear Mountain CONTINUATION ← 2

Agent executes:

# Create project (gets projectId from Google Flow)
curl -X POST http://127.0.0.1:8100/api/projects \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Luna the Space Cat",
    "story": "Luna is a small white cat astronaut who lands on a candy planet. She discovers a chocolate river and plants a flag on a gummy bear mountain.",
    "characters": [
      {"name": "Luna", "entity_type": "character", "description": "Small white cat with big blue eyes, wearing a tiny orange space suit with round glass helmet. Fluffy tail. Pixar-style 3D.", "voice_description": "Soft curious childlike voice full of wonder, with a slight purring undertone"},
      {"name": "Candy Planet Surface", "entity_type": "location", "description": "Alien planet surface made of colorful hard candies, lollipop trees, cotton candy clouds, pastel pink and purple sky. Pixar-style 3D."},
      {"name": "Chocolate River", "entity_type": "location", "description": "Wide flowing river of dark melted chocolate with marshmallow rocks along banks, steam rising. Pixar-style 3D."},
      {"name": "Gummy Bear Mountain", "entity_type": "location", "description": "Tall mountain made of giant translucent gummy bears stacked together, glowing from inside, rainbow colors. Pixar-style 3D."}
    ]
  }'
# → project_id: "p-xxx"

# Create video
curl -X POST http://127.0.0.1:8100/api/videos \
  -H "Content-Type: application/json" \
  -d '{"project_id": "p-xxx", "title": "Luna Episode 1"}'
# → video_id: "v-xxx"

# Create 3 scenes
# Scene 1: prompt (image) + video_prompt (8s sub-clip with camera + dialogue)
curl -X POST http://127.0.0.1:8100/api/scenes -H "Content-Type: application/json" \
  -d '{
    "video_id": "v-xxx", "display_order": 0,
    "prompt": "Luna steps out of a small rocket onto Candy Planet Surface. First footprint in candy dust. Wide shot, dramatic landing. Pixar 3D.",
    "video_prompt": "0-3s: Wide crane down shot, Luna emerges from rocket hatch onto Candy Planet Surface. Luna gasps \"Wow!\" 3-6s: Low angle tracking shot, Luna takes first steps on candy ground, looking around in wonder. Luna says \"Everything is made of candy!\" 6-8s: Wide establishing shot, Luna small in frame against vast Candy Planet Surface landscape, cotton candy clouds. Silence, gentle wind.",
    "character_names": ["Luna", "Candy Planet Surface"],
    "chain_type": "ROOT"
  }'
# → scene_id: "s-1"

curl -X POST http://127.0.0.1:8100/api/scenes -H "Content-Type: application/json" \
  -d '{
    "video_id": "v-xxx", "display_order": 1,
    "prompt": "Luna kneels at the edge of Chocolate River on Candy Planet Surface, dipping a paw in. Surprised expression. Warm lighting. Pixar 3D.",
    "video_prompt": "0-3s: Over-the-shoulder shot, Luna kneels at Chocolate River edge on Candy Planet Surface, steam rising. Luna asks \"What is this?\" 3-5s: Close-up Luna paw dipping into chocolate, slow motion. Luna tastes it, eyes widen. Luna says \"It is chocolate! Real chocolate!\" 5-8s: Medium shot, Luna jumps up with arms raised, Chocolate River flowing behind. Luna shouts \"This is the best planet ever!\"",
    "character_names": ["Luna", "Candy Planet Surface", "Chocolate River"],
    "chain_type": "CONTINUATION", "parent_scene_id": "s-1"
  }'
# → scene_id: "s-2"

curl -X POST http://127.0.0.1:8100/api/scenes -H "Content-Type: application/json" \
  -d '{
    "video_id": "v-xxx", "display_order": 2,
    "prompt": "Luna plants a small flag on top of Gummy Bear Mountain. Candy Planet Surface below. Sunset glow through gummy bears. Pixar 3D.",
    "video_prompt": "0-3s: Low angle push in, Luna climbs last steps of Gummy Bear Mountain, determination on face. 3-5s: Medium shot, Luna plants flag on summit, triumphant pose. Luna says \"I did it! Planet Candy is claimed!\" 5-8s: Epic crane up to wide bird eye view, Luna tiny on Gummy Bear Mountain peak, Candy Planet Surface stretching to horizon, golden sunset through translucent gummy bears. Silence, majestic.",
    "character_names": ["Luna", "Candy Planet Surface", "Gummy Bear Mountain"],
    "chain_type": "CONTINUATION", "parent_scene_id": "s-2"
  }'
# → scene_id: "s-3"

Agent says: "Project created with 4 entities and 3 scenes. Run /fk-gen-refs p-xxx to generate reference images."


2. /fk-gen-refs p-xxx

Agent executes (one at a time, waits for each):

Generating Luna (character, portrait)...
  Request req-1 → PROCESSING → COMPLETED ✓  media_id: a1b2c3d4-...

Generating Candy Planet Surface (location, landscape)...
  Request req-2 → PROCESSING → COMPLETED ✓  media_id: e5f6a7b8-...

Generating Chocolate River (location, landscape)...
  Request req-3 → PROCESSING → COMPLETED ✓  media_id: c9d0e1f2-...

Generating Gummy Bear Mountain (location, landscape)...
  Request req-4 → PROCESSING → COMPLETED ✓  media_id: a3b4c5d6-...

All 4 references ready:
  Luna                  character      media_id=a1b2c3d4-...  ✓
  Candy Planet Surface  location       media_id=e5f6a7b8-...  ✓
  Chocolate River       location       media_id=c9d0e1f2-...  ✓
  Gummy Bear Mountain   location       media_id=a3b4c5d6-...  ✓

Run /fk-gen-images p-xxx v-xxx to generate scene images.

3. /fk-gen-images p-xxx v-xxx

Agent pre-checks: All 4 refs have UUID media_id ✓

Agent executes (one at a time):

Scene 1 (Luna + Candy Planet Surface): using 2 reference images
  Request req-5 → PROCESSING → COMPLETED ✓

Scene 2 (Luna + Candy Planet Surface + Chocolate River): using 3 reference images
  Request req-6 → PROCESSING → COMPLETED ✓

Scene 3 (Luna + Candy Planet Surface + Gummy Bear Mountain): using 3 reference images
  Request req-7 → PROCESSING → COMPLETED ✓

All 3 scene images ready:
  Scene 1: image=OK  media_id=f1e2d3c4-...
  Scene 2: image=OK  media_id=b5a6c7d8-...
  Scene 3: image=OK  media_id=e9f0a1b2-...

Run /fk-gen-videos p-xxx v-xxx to generate videos.

4. /fk-gen-videos p-xxx v-xxx

Agent pre-checks: All 3 scenes have UUID image media_id ✓

Agent executes (one at a time, 2-5 min each):

Scene 1 video: startImage=f1e2d3c4-...
  Request req-8 → PROCESSING → polling... [30s] [60s] [90s] → COMPLETED ✓

Scene 2 video: startImage=b5a6c7d8-..., endImage=f1e2d3c4-... (chained from scene 1)
  Request req-9 → PROCESSING → polling... [30s] [60s] [90s] [120s] → COMPLETED ✓

Scene 3 video: startImage=e9f0a1b2-..., endImage=b5a6c7d8-... (chained from scene 2)
  Request req-10 → PROCESSING → polling... [30s] [60s] [90s] → COMPLETED ✓

All 3 videos ready:
  Scene 1: video=OK  url=https://storage.googleapis.com/...
  Scene 2: video=OK  url=https://storage.googleapis.com/...
  Scene 3: video=OK  url=https://storage.googleapis.com/...

Run /fk-concat v-xxx to download and merge.

5. /fk-concat v-xxx

Getting output dir from API...
OUTDIR=output/luna_the_space_cat

Downloading scene 1... 2.1 MB → output/luna_the_space_cat/scenes/scene_1.mp4
Downloading scene 2... 3.4 MB → output/luna_the_space_cat/scenes/scene_2.mp4
Downloading scene 3... 2.8 MB → output/luna_the_space_cat/scenes/scene_3.mp4
Normalizing (720x1280, 24fps, h264) → output/luna_the_space_cat/norm/
Concatenating 3 scenes...

Done!
  output/luna_the_space_cat/luna_the_space_cat_final.mp4
  720x1280 (vertical)
  24 seconds
  8.3 MB

Result

A 24-second Pixar-style video of Luna the Space Cat:

  • Scene 1: Luna lands on the candy planet (8s)
  • Scene 2: Luna discovers the chocolate river and tastes it (8s)
  • Scene 3: Luna plants a flag on gummy bear mountain at sunset (8s)

Luna looks the same across all 3 scenes because her reference image was used as imageInputs in every scene generation. The candy planet surface is consistent too — same colors, same lollipop trees, same sky.


What the agent needed to know

  1. Break story into visual elements → characters, locations, assets (with entity_type)
  2. Description = appearance only → what it looks like, not what it does
  3. Scene prompt = action only → what happens, reference entities by name
  4. character_names = which refs to use → list every entity visible in that scene
  5. Pipeline order → refs MUST complete before images, images before videos
  6. One request at a time → 10s cooldown, wait for COMPLETED before next
0