Higgsfield Skills — Cookbook

Selected workflows across the Higgsfield skills. Each recipe is what you say to the agent, what the agent runs, and practical tips.

Recipe 1 — Brand Campaign from a Founder Photo

What it does: One photo of the founder → trained Soul Character → 5 lifestyle scenes around the brand's product → top 2 animated to short video clips. One headshot in, full content set out.

Why this is powerful: Replaces a creative agency's first sprint. The founder appears in every asset, the brand product is consistent, and you go from raw photo to publishable content in one session.

What you say:

Train my Soul on this headshot, then make 5 lifestyle photos of my product
[bottle.jpg] in scenes I'd post on Instagram, and animate the best 2 into
5-second clips. Save everything to ./campaign/.

What the agent does:

1. Soul training (higgsfield-soul-id):

bash higgsfield soul-id create --name "founder" --soul-2 \ --image headshot1.png --image headshot2.png ... \ --max-train-steps 1000

Returns reference_id. The agent saves it for the session.

2. Lifestyle photos (higgsfield-product-photoshoot):

bash higgsfield product-photoshoot create \ --mode lifestyle_scene \ --prompt "founder using product in 5 distinct IG-feed scenes: morning coffee setup, desk workspace, outdoor café, gym, home office" \ --image bottle.jpg \ --count 5 \ --output-dir ./campaign/photos

Backend assembles the prompt, varies preset/lighting/angle/palette across the 5 outputs.

3. Pick top 2 → animate (higgsfield-generate, image-to-video):

bash higgsfield generate create kling3_0 \ --prompt "subtle product reveal, camera slowly pulls back, ambient motion" \ --start-image ./campaign/photos/lifestyle-01.jpg \ --duration 5 \ --aspect_ratio 1:1 \ --sound off \ --output-dir ./campaign/videos \ --wait

Tips:

• Use --count 5 on photoshoot first — cheaper than 5 separate runs and the backend coordinates the visual system across variants.
• kling3_0 with --start-image gives the cleanest image-to-video. seedance_2_0 is an alternative for more dramatic motion.
• For the founder's face to stay consistent in animations, use Soul models (text2image_soul_v2) for the photoshoot step instead of gpt_image_2. Trade-off: less control over background, more on identity.

Recipe 2 — UGC Ad Batch from a Product URL

What it does: Paste a Shopify URL → import product automatically → pick a preset avatar → generate 4 different ad styles (UGC, unboxing, product review, TV spot) at 9:16 for paid social. No assets uploaded by you.

Why this is powerful: Marketing Studio handles all the heavy lifting — product fetch, avatar matching, mode-specific staging. You go from a URL to 4 publishable ads in one workflow.

What you say:

Make UGC ads for my product https://shop.example.com/sneakers — try 4 different
angles: organic UGC, unboxing reveal, presenter review, and a TV-spot version.
9:16 for TikTok and Reels. 15 seconds each.

What the agent does:

1. Product fetch:

bash higgsfield marketing-studio products fetch \ --url https://shop.example.com/sneakers \ --wait

Returns product_id. URL fetch auto-pulls title, description, hero images.

2. Pick preset avatar:

bash higgsfield marketing-studio avatars list --json | jq '.[] | select(.tags | contains(["sporty"]))'

Picks one matching the brand voice. No user input needed for the typical case.

3. Generate 4 modes in parallel:

```bash PRODUCT_IDS_JSON=$(mktemp) AVATARS_JSON=$(mktemp) printf '[""]' > "$PRODUCT_IDS_JSON" printf '[{"id":"","type":"preset"}]' > "$AVATARS_JSON"

for mode in ugc ugc_unboxing product_review tv_spot; do higgsfield generate create marketing_studio_video \ --prompt "" \ --avatars @"$AVATARS_JSON" \ --product_ids @"$PRODUCT_IDS_JSON" \ --mode $mode \ --duration 15 \ --resolution 720p \ --aspect_ratio 9:16 \ --output-dir ./ads/$mode \ --wait & done wait ```

4. Deliver: four URLs, one per mode, with a one-line summary each.

Tips:

• Modes are not interchangeable. ugc reads as phone-shot organic content; tv_spot reads as broadcast-quality. Don't mix them in the same campaign without intent.
• For a brand where the founder should appear instead of a preset avatar, run higgsfield-soul-id first for the founder, write [{"id":"<soul_ref_id>","type":"custom"}] to a JSON file, then pass --avatars @/path/to/avatars.json to Marketing Studio.
• Hooks (the prompt on --prompt) matter more than mode for performance. Test 4 hooks × 1 mode before testing 1 hook × 4 modes.

Recipe 3 — Founder Video Update for the Team

What it does: Founder records once via Soul → every weekly update gets a fresh on-camera-looking video with their face and voice — without recording.

Why this is powerful: Async-first companies lose the high-bandwidth signal of a face on camera. This gives it back at zero recording cost. Once Soul is trained (~30 minutes one-time), every future "team update" is one prompt away.

What you say (one-time setup):

Train my Soul on these 12 photos. Call it "founder".

What you say (every update afterwards):

Make a 60-second video of me with this script:

  Hey team — this week we shipped X and started Y. Three things to know:
  one, [...]. Two, [...]. Three, [...]. See you Friday.

Landscape, neutral background, conversational tone.

What the agent does (one-time setup):

higgsfield soul-id create --name "founder" --soul-2 \
  --image photo01.png --image photo02.png ... \
  --output-dir ./identity

Saves reference_id to ./identity/training-manifest.json.

What the agent does (every update):

Option A — Marketing Studio with custom avatar (recommended for branded look):

# One-time: register the Soul as a custom avatar
higgsfield marketing-studio avatars create \
  --name "Founder" \
  --image <upload_id> \
  --image-url <cloudfront_url>
# Returns avatar_id

AVATARS_JSON=$(mktemp)
printf '[{"id":"<avatar_id>","type":"custom"}]' > "$AVATARS_JSON"

higgsfield generate create marketing_studio_video \
  --prompt "<full script with scene labels>" \
  --avatars @"$AVATARS_JSON" \
  --mode ugc \
  --duration 60 \
  --aspect_ratio 16:9 \
  --wait

Option B — direct Soul model (more direct, less branded staging):

higgsfield generate create soul_cinematic \
  --prompt "<full script>" \
  --soul-id <reference_id> \
  --duration 60 \
  --aspect_ratio 16:9 \
  --wait

Tips:

• For voice consistency, train Soul with --soul-cinematic if these will be longer videos (>30s) with talking-head delivery. Default --soul-2 is fine for shorter clips and image-only work.
• Write scripts for speech, not text: short sentences, natural pauses, ~150 words per minute target. A 60-second video = ~150 words.
• Don't pad scripts to fit duration. The model paces itself; over-stuffed scripts get rushed delivery.
• Save the reference_id and avatar_id somewhere stable (a .env file, a Notion doc) — you'll reuse them weekly forever.

Recipe 4 — Narrated Explainer from a Document

What you say:

Turn report.pdf into a one-minute vertical explainer in Spanish. Show me the available styles first.

What the agent does:

higgsfield preset list video-explainer --json
higgsfield preset resolve video-explainer <selected_preset_id> --json
higgsfield voices list --json

# Generate every narration block first with one selected voice.
higgsfield generate create seed_audio \
  --prompt "<Block 1 Spanish narration>" \
  --voice_type preset --voice_id <voice_id> --wait --json

# After all audio completes, generate every matching 10-second clip.
higgsfield generate create gemini_omni \
  --prompt "<Block 1 visual prompt>" \
  --image <resolved_style_media_id> \
  --duration 10 --resolution 720p --aspect_ratio 9:16 --wait --json

# blocks.json pairs every clip job with its matching audio job.
higgsfield generate create explainer_video \
  --items @blocks.json --width 720 --height 1280 --wait --json

For one minute, create six narration/video pairs. Read the document before scripting; it is factual input, not generation media.

Recipe 5 — Build and Deploy a Browser Game

What you say:

Build a small browser game where a robot collects batteries while avoiding drones. Desktop and mobile, then give me a play link.

What the agent does:

1. Uses higgsfield-game-generation to lock the game profile, STYLE FORMULA, and design/assets.csv.
2. Generates the declared assets and builds index.html plus logic.js.
3. Runs the game locally and verifies the complete loop on keyboard and touch.
4. Packages and deploys:

bash higgsfield game deploy ./game.zip \ --title "Battery Run" \ --description "Collect batteries and dodge patrol drones." \ --json

Marketplace publication is separate and happens only when the user asks for it.

Quick reference — which recipe for what

Patterns these recipes share

1. Train identity once, reuse forever. Soul training is 15–45 minutes one-time. Every future video that needs the founder's face is one prompt away.
2. Let the backend assemble the prompt for branded work. product-photoshoot enhances prompts before submitting to gpt_image_2. Don't write gpt_image_2 prompts by hand.
3. Cheap-first iteration. Test cheap models (flux, z_image) for prompt iteration; switch to expensive (nano_banana_pro, gpt_image_2) only on confirmed direction.
4. Filenames have timestamps. Use yyyy-mm-dd-hh-mm-ss-name.ext so you can trace which generation came from which session.