higgsfield-generate

name	higgsfield-generate
description	Generate brand-quality images, videos, audio (voiceover / text-to-speech, voice cloning, and video dubbing — Seed Audio 1.0 plus ElevenLabs and other selectable TTS engines), and ads via Higgsfield (Marketing Studio: branded ads, avatars, products, hooks; Soul/Cinema/Kling models). Use when the user says "/omg-higgsfield-generate", "/higgsfield-generate", "generate an image", "make a video", "make a voiceover", "text to speech", "voice clone", "dub this video", "brand video", "UGC ad", "product demo", or "marketing studio". R-VISUAL-ID: the visual-generation engine; consumes Soul IDs from higgsfield-soul-id. Requires the higgsfield CLI, auth, and a paid plan.
triggers	["omg-higgsfield-generate","higgsfield-generate","generate image","make a video","make a voiceover","text to speech","voice clone","dub this video","dubbing","brand video","UGC ad","product demo","marketing studio"]
allowed-tools	["Bash","Read","Glob","Grep"]
domain	visual-identity
read_only	false
argument-hint	<what to generate>
source	higgsfield-ai/skills (github.com/higgsfield-ai/skills)
license	see upstream repo

OmegaOS skill — vendored from higgsfield-ai/skills (github.com/higgsfield-ai/skills). Triggers /omg-higgsfield-generate (and /higgsfield-generate). Governed by R-VISUAL-ID. Requires: the higgsfield CLI (installed at runtime), higgsfield auth login, and a paid plan.

Higgsfield Generate

Submit jobs to any Higgsfield model. Wraps the higgsfield CLI. Covers generic image/video gen, Marketing Studio (branded ads, avatars, products, hooks, settings), and, secondarily, Virality Predictor video scoring.

Step 0 — Bootstrap

Before any other command:

1. If higgsfield is not on $PATH, install it:

   curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh
   

2. If higgsfield account status fails with Session expired / Not authenticated, ask the user to run higgsfield auth login (interactive) and wait for confirmation.

UX Rules

1. Be concise. No raw IDs, no JSON dumps in chat. Print the media URL for generated assets, or the text summary for Virality Predictor.
2. No internal jargon. Don't narrate "calling higgsfield cost", "polling job".
3. Detect the user's language from the first message and reply in it. Technical args (--aspect_ratio 16:9) stay English.
4. Don't batch-ask. Pick a sane default model and ask one thing at a time only if genuinely missing.
5. Don't pre-estimate cost or optimize for cheaper models unless the user asks. Prefer the quality default first.
6. Pass --wait to generate create so the command blocks until done and prints the result URL itself. Avoid the two-step create → wait pattern.

Discovery guardrail

When looking for a Higgsfield feature/model, do not rely only on semantic search or CLI --help. First run an unfiltered model list, then inspect likely job_set_type names. If the user says a model exists but search returns no results, trust that signal and verify with the full model list before answering.

Virality Predictor is exposed as:

• Customer-facing name: Virality Predictor
• Technical job_set_type: brain_activity
• Category/output: text report. This is video-in/text-out analysis, not a text/chat generation model.
• Input: uploaded video
• Purpose: finished-video hook, attention, retention, and virality analysis

If the user says "analyze this video", "score this ad", "evaluate the hook", or similar, route to brain_activity even though it appears under text/analysis models. Classify by task intent and required input, not by output category alone.

Workflow — generic generation

1. Pick a model. Start with the core defaults unless the brief clearly needs a specialist:

   • GPT Image 2 → default image model for high-fidelity general generation, graphic design, UI, banners, typography, and on-image text.
   • Seedance 2.0 → default video model for serious motion, cinematic clips, multi-shot work, image-to-video, and 4–15s production-quality output. 12s is valid.
   • Nano Banana 2/Pro → default for character, cartoon, stylized, and reference-driven image work; use Pro for harder briefs.
   • Marketing Studio → default for ads, UGC, product demos, unboxing, TV spots, presenter videos, and brand/product workflows.

   Image:

   • Brand product visual (Pinterest pin, lifestyle, hero banner, ad pack, virtual try-on) → use higgsfield-product-photoshoot instead. NOT this skill.
   • Generated product concept / packaging / can / bottle with brand name or label text → GPT Image 2.
   • Branded ad image with avatar + product (Marketing Studio shape) → Marketing Studio Image (see Marketing Studio below)
   • Aesthetic UGC / fashion editorial / lifestyle character → Soul 2.0
   • Cinematic still frame → Soul Cinema
   • Highly characterful creative persona (text-only, distinctive) → Soul Cast
   • Locations / environments / no-people scenes → Soul Location (best in class)
   • Vector illustrations OR face edit + complex scene swap → Seedream 4.5
   • Soul Character (reference id from higgsfield-soul-id) → Soul 2.0 for stills, Soul Cinema for cinematic
   • Character or cartoon-style work → Nano Banana 2; step up to Nano Banana Pro on hard cases
   • Fast and cheap iteration → Z Image
   • Default for everything else → GPT Image 2. Graphic design, UI, banners, typography, and high-fidelity general generation.

   Video:

   • All advertising / commercial / branded ad video → Marketing Studio (see Marketing Studio below)
   • Default all-purpose serious video (multi-shot, consistent identity, motion-heavy, image-to-video, 4–15s requests) → Seedance 2.0. SOTA. Do not downgrade to Seedance 1.5 just because its duration enum is easier to read; validate Seedance 2.0 first.
   • Single-plane scene without strong dynamics, cheaper than Seedance 2.0 → Kling 3.0
   • Cheap clean shot without cuts, only when the user asks for cheaper/budget output → Seedance 1.5 Pro
   • Cinema-grade highest fidelity → Cinema Studio Video 3.0
   • Cheap with strong physics, no audio needed → Minimax Hailuo
   • Fast batch / volume → Veo 3.1 Lite

   Video analysis:

   • Rate a finished video's hook, virality potential, attention, retention, or distraction risk → Virality Predictor (brain_activity). This is a video analysis model that returns a text score/report, not a generated media asset.

   For the actual --model ID to pass to higgsfield generate create, run higgsfield model list --json | jq to map display names to IDs. See references/model-catalog.md for the full table.

2. Pass media inputs straight to flags. Media flags accept a local file path or a UUID. CLI auto-uploads paths and auto-detects job vs upload for UUIDs. No need to pre-upload. Each model declares accepted roles (image, start_image, end_image, video, audio) — see references/media-inputs.md.

3. Validate quickly. If unsure of params, run higgsfield model get <jst> --json once and pass only what's needed. Validate the preferred model before falling back to an older one. Use schema defaults otherwise. The server returns adjustments for non-fatal coercions (e.g. aspect_ratio=99:99 → closest match) and a structured error for invalid declared-param values.

4. Submit and wait in one shot. higgsfield generate create <jst> [--prompt "..."] [media flags] [param flags] --wait. Blocks until terminal status and prints the result on stdout. Tunables: --wait-timeout 20m (default 10m), --wait-interval 5s (default 3s). Virality Predictor does not need a prompt; pass --video.

5. Deliver. For generated media, send the URL plus a one-line summary (model, duration if video). For Virality Predictor, deliver the scores, business interpretation, and the Open report link. Do not surface .glb, .bin, or region-table internals in normal chat output.

To inspect or rerun later, higgsfield generate list --json and higgsfield generate get <id> --json work for retrospection. higgsfield generate wait <id> is still available if you ever need to rejoin a job started without --wait.

Media flags

Flag	Purpose	Models that accept it
--image <path-or-id>	reference image	most image models, seedance_2_0, veo3, marketing_studio_video
--start-image <path-or-id>	first frame for image-to-video transitions	kling3_0, kling2_6, veo3_1, seedance_2_0, marketing_studio_video
--end-image <path-or-id>	last frame for transitions	kling3_0, seedance_2_0, marketing_studio_video
--video <path-or-id>	reference or analyzed video	seedance_2_0, brain_activity
--audio <path-or-id>	reference audio (lipsync, soundtrack match)	seedance_2_0 (use this, NOT --generate-audio)

Each flag accepts either a local file path (auto-uploaded) or a UUID (upload id from higgsfield upload create, or a previous job id). Each model declares its own role set via MEDIA_ROLES. See references/media-inputs.md for the full table.

Common params

Flags pass through to model schema. Use higgsfield model get <jst> to discover.

higgsfield generate create gpt_image_2 --prompt "neon city at dusk" --aspect_ratio 16:9 --resolution 2k --wait
higgsfield generate create nano_banana_2 --prompt "anime character concept, expressive pose" --image ./ref.png --wait
higgsfield generate create seedance_2_0 --prompt "camera dollies in" --start-image ./first.png --duration 12 --wait
higgsfield generate create text2image_soul_v2 --prompt "..." --soul-id <soul_ref_id> --quality 2k --wait
higgsfield generate create brain_activity --video ./ad.mp4 --wait

For machine-readable output (chained pipelines, agent context), add --json. With --wait --json you get the final job object array. Without --wait, you get the job IDs. Virality Predictor stores raw analysis and render artifacts in the job params, but the default text output should stay to scores plus Open report.

Stdin prompt: echo "..." | higgsfield generate create z_image --wait.

Soul image quality: for text2image_soul_v2 and soul_cinematic, pass --quality 1.5k or --quality 2k. These are UI-facing tiers; the backend maps them to 720p/1080p and model-specific dimensions from the selected --aspect_ratio. soul_location has no quality selector; it uses fixed dimensions per aspect ratio.

Marketing Studio

Branded image/video gen: avatars + products + optional setup hooks/settings + ad-style modes. Use models marketing_studio_video and marketing_studio_image.

Concepts

• Avatar — presenter face. Curated preset (browse higgsfield marketing-studio avatars list) or custom (uploaded photos via higgsfield marketing-studio avatars create). For UGC modes, an avatar is optional if the brief clearly mentions a person; the backend can create a Soul Character automatically. Pass an avatar when the user wants a specific presenter.
• Product — brand item with title + reference images. Imported from URL (higgsfield marketing-studio products fetch --url ...) or created from uploaded images (higgsfield marketing-studio products create).
• Webproduct — App Store / web page version. Auto-routes when fetching App Store URLs.
• Hook — reusable opening angle / ad hook. Browse with higgsfield marketing-studio hooks list. Hook text is prepended to the user's prompt; it does not replace --prompt.
• Setting — reusable environment / scene context. Browse with higgsfield marketing-studio settings list.
• Ad reference — reusable inspiration video that can be bound to an avatar and/or product. Created from an uploaded video (--video-input <upload_id>) or a previous generation job (--job <job_id>). Browse with higgsfield marketing-studio ad-references list. See references/marketing-ad-references.md.
• Brand kit — captures a brand's identity (name, logo, hero images, colours, fonts, tone) for reuse across image generations. Created by handing in a website URL (higgsfield marketing-studio brand-kits fetch --url https://… --wait). See references/marketing-brand-kits.md.
• Ad format — presets that drives the visual structure of a generated image (headline, bullet-points, etc.). Read-only, browse with higgsfield marketing-studio ad-formats list. Required input for dtc-ads generate.

Discovery commands

Use these exact list commands when the user asks what already exists:

higgsfield marketing-studio avatars list --json
higgsfield marketing-studio products list --json
higgsfield marketing-studio hooks list --json
higgsfield marketing-studio settings list --json
higgsfield marketing-studio ad-references list --json
higgsfield marketing-studio brand-kits list --json
higgsfield marketing-studio ad-formats list --json

--hook_id and --setting_id are supported by marketing_studio_video only; do not pass them to marketing_studio_image.

UX rules (additional)

• One question per phase. Don't ask product+avatar+mode upfront.
• Two ad approaches are mutually exclusive. Either the user gives an ad reference video (reference-driven) or picks hook/setting blocks (composed-from-blocks) — never both. If the user has an ad reference selected, do not offer hook/setting; if hook/setting are picked, do not offer to attach an ad reference.
• Ad reference source. The only valid inputs are a local video file (uploaded via higgsfield upload create ... --video) or a prior video job. If the user provides anything else, ask for a local file.
• dtc-ads ad format is mandatory. Always ask the user to pick from ad-formats list. There is no auto-default — both the CLI and server reject calls without --format-id.
• dtc-ads optional inputs. Suggest avatars, products, and reference media when the brief calls for them; only attach what the user picks.

Workflow — quick ad video

1. Get product.
   • Existing product → higgsfield marketing-studio products list --json
   • URL → higgsfield marketing-studio products fetch --url <url> --wait (polls until import done)
   • Local images → higgsfield upload create <photo>... then higgsfield marketing-studio products create --title "..." --image <id>... Capture product id. When using --hook_id, strongly prefer passing --product_ids; hooks are designed to pivot into a product and work poorly without product context.
2. Pick avatar if needed.
   • Default: higgsfield marketing-studio avatars list and pick a preset matching the brand voice.
   • Custom: higgsfield marketing-studio avatars create --name "..." --image <upload_id>. For UGC modes, you may omit --avatars when no specific presenter is required and the brief mentions a person; the backend can synthesize a Soul Character.
3. Optionally pick setup items.
   • Hook: higgsfield marketing-studio hooks list --json
   • Setting: higgsfield marketing-studio settings list --json Pass selected IDs as --hook_id <hook_id> and --setting_id <setting_id> for marketing_studio_video only. Do not copy the hook's prompt into --prompt unless the user explicitly wants to reinforce the same wording.
4. Pick mode if needed. Default is ugc; --mode is not required just because --hook_id is present. Other current slugs: ugc_how_to, ugc_unboxing, product_showcase, product_review, tv_spot, wild_card, ugc_virtual_try_on, virtual_try_on. Hook/setting are valid only for ugc, ugc_how_to, ugc_unboxing, product_review, ugc_virtual_try_on — do not pass --hook_id / --setting_id with the other modes. See references/marketing-modes.md.
5. Generate (one-shot).

   PRODUCT_IDS_JSON=$(mktemp)
   AVATARS_JSON=$(mktemp)
   printf '["<product_id>"]' > "$PRODUCT_IDS_JSON"
   printf '[{"id":"<avatar_id>","type":"preset"}]' > "$AVATARS_JSON"
   
   higgsfield generate create marketing_studio_video \
     --prompt "..." \
     --avatars @"$AVATARS_JSON" \
     --product_ids @"$PRODUCT_IDS_JSON" \
     --mode ugc \
     --duration 15 \
     --resolution 720p \
     --aspect_ratio 9:16 \
     --wait
   

   Add --hook_id <hook_id> and/or --setting_id <setting_id> when a setup hook/setting was selected. product_ids and avatars are JSON arrays; pass them via @/path/to/file.json. Do not pass a bare UUID to --product_ids. Resolution is 480p or 720p. Aspect ratio is one of auto/21:9/16:9/4:3/1:1/3:4/9:16. --generate-audio true is supported here (unlike seedance_2_0). --wait blocks until done; bump --wait-timeout 30m for longer ad runs.
6. Deliver. URL + one-line summary (mode, duration).

Click-to-Ad shortcut (URL-driven)

When the user gives a product URL and wants a marketing video in one go:

# 1. Trigger fetch (returns the product id, import runs in the background)
higgsfield marketing-studio products fetch --url https://shop.example.com/sneakers --wait

# 2. Generate the marketing video against the same URL — backend reuses the entity
higgsfield generate create marketing_studio_video \
  --url https://shop.example.com/sneakers \
  --mode ugc \
  --duration 15 \
  --aspect_ratio 9:16 \
  --wait

Backend dedupes by URL, so repeated runs reuse the existing entity instead of re-fetching.

Workflow — marketing image

Same as above but use marketing_studio_image model:

higgsfield generate create marketing_studio_image \
  --prompt "..." \
  --aspect_ratio 1:1 \
  --resolution 2k \
  --wait

Virality Predictor video scoring

Use Virality Predictor (brain_activity) when the user wants to evaluate a finished video as a business creative: hook strength, virality potential, attention, retention, or how well the content/product holds focus and minimizes distraction. Treat "Virality Predictor" as the customer-facing feature name; brain_activity is only the CLI/job_set_type.

higgsfield generate create brain_activity --video ./creative.mp4 --wait

The result is text, not a generated image/video. Report the overall score, peak hook second, sustain score, strongest/weakest regions, and report URL if present. Interpret it as an objective attention proxy for creative testing: higher Visual/Auditory/Language/Attention scores suggest stronger stimulus and focus; lower Default Mode is better because it suggests less mind-wandering.

The CLI prints an Open report URL like https://<app-domain>/apps/virality-predictor?resultJobId=<job_id>. Send that URL for the visual report. Raw artifact URLs such as brain_example_url, vertexMapBinaryUrl, and vertexMapUrl are implementation details; mention them only when the user asks for raw data or implementation details.

Good final shape:

Overall score: 44/100
Peak hook: 49% at 1s
Sustain: 89%
Strongest region: Visual Cortex
Risk: Default Mode is high, which can indicate mind-wandering.

Open report: <report_url>

Audio generation

Higgsfield generates audio, not only images and video — voiceover / text-to-speech, voice cloning, sound, and video dubbing. This is generation (audio out); it is distinct from the --audio reference input used for Seedance lipsync (audio in — see references/media-inputs.md). Voiceover is powered by Seed Audio 1.0 and, via text2speech_v2, selectable engines including ElevenLabs. Same submit-and-wait shape as everything else: higgsfield generate create <job_set_type> [flags] --wait.

Apply the discovery guardrail here too — list the audio roster before answering. Note that dubbing and voice_change are listed under type: video (they rewrite a video's voice track), so classify by intent, not by the output category:

higgsfield model list --json | jq '.[] | select(.type=="audio")'   # audio-typed models
higgsfield model get <job_set_type> --json                          # exact params for one model

Audio models (runtime-verified, CLI v0.1.40)

Job	job_set_type	Required / key params	Notes
Voiceover / TTS — Seed Audio	seed_audio	--prompt (req); --format wav|mp3|pcm|ogg_opus; --sample_rate 8000…48000; --speech_rate/--pitch_rate/--loudness_rate	Seed Audio 1.0 — default high-quality TTS from a script.
Voiceover / TTS — multi-engine	text2speech_v2	--model elevenlabs|minimax|seed_speech|vibe_voice|cozy_voice (req); --prompt (req); --voice_id (req); --voice_type preset|element (req)	ElevenLabs = --model elevenlabs — an engine option here, not a separate model id.
Voiceover — Inworld	inworld_text_to_speech	--prompt (req); --voice (req)	Inworld TTS.
Sound / SFX / ambience	mirelo_text_to_audio	--prompt (req); --duration (req)	Mirelo text-to-audio.
Music	sonilo_music	run higgsfield model get sonilo_music	Sonilo Music.
Dubbing (localize a video)	dubbing	--video <path-or-id> (req); --target_language <code> (req)	Re-voices a finished video in another language. type: video. 18 languages today (see below).
Voice cloning / transfer	voice_change	--video <path-or-id> (req); --voice_id (req); --voice_type preset|element	Swaps the speaking voice in a video for a consistent brand voice. type: video.
Transcription (audio→text)	speech2text	--audio <path-or-id> (req)	Speech-to-text.

Media inputs (schema fields input_video / input_audio) are passed with the standard media flags --video / --audio — a local path (auto-uploaded) or an upload/job id, same as the rest of the CLI. If a flag is rejected, confirm the accepted shape with higgsfield model get <job_set_type> --json.

Examples (verified flags only)

# Voiceover from a script (Seed Audio 1.0) → mp3
higgsfield generate create seed_audio --prompt "Welcome to the future of marketing." --format mp3 --wait

# Voiceover via the ElevenLabs engine (voice_id + voice_type required)
higgsfield generate create text2speech_v2 --model elevenlabs --voice_type preset --voice_id <voice_id> --prompt "..." --wait

# Dub a finished video into French
higgsfield generate create dubbing --video ./ad.mp4 --target_language fra --wait

# Re-voice a video with a brand voice (voice transfer)
higgsfield generate create voice_change --video ./ad.mp4 --voice_type preset --voice_id <voice_id> --wait

Dubbing languages

higgsfield model get dubbing enumerates 18 target_language codes in CLI v0.1.40: eng, cmn, fra, hin, ita, jpn, kor, por, rus, tur, spa, deu, ara, pol, ind, fil, swe, fin. Higgsfield markets dubbing in 50+ languages — treat 50+ as the platform claim and these 18 as what this CLI version exposes today; re-run model get dubbing as the roster grows.

Getting a voice_id

text2speech_v2 and voice_change require a voice_id (with voice_type preset or element/custom). CLI v0.1.40 surfaces no voices list command — source the id from the Higgsfield voice library / app (preset) or your cloned element voice. Don't guess a voice_id.

When to reach for audio

You can now produce the whole asset in one stack (R-VISUAL-ID): generate the visual (image → video) first, then voiceover (seed_audio / text2speech_v2) for narration, voice_change to re-voice a clip in a consistent brand voice, or dubbing to localize a finished ad into another language. Live audio runs spend credits and need higgsfield auth login + a paid plan; OmegaOS ships only this skill markdown and never auto-installs the CLI.

Errors

• Missing required params: prompt → user gave no prompt; ask for it.
• Missing required params: medias on brain_activity / Virality Predictor → pass exactly one video via --video <path-or-id>.
• Invalid values: aspect_ratio=99:99 (allowed: ...) → bad enum; pick from allowed.
• Unknown params: foo → schema doesn't accept that flag; check higgsfield model get <jst>. If this happens for hook_id or setting_id, the selected model/job_set_type does not support Marketing Studio setup items.
• Session expired → higgsfield auth login.

See references/troubleshooting.md for more.

Reference docs

Load on demand:

• references/model-catalog.md — picking the right model for the task
• references/prompt-engineering.md — writing prompts that work
• references/media-inputs.md — image/video/audio reference flows and Virality Predictor video analysis
• references/audio-generation.md — audio generation: voiceover/TTS, voice cloning, dubbing (audio out; distinct from the --audio reference input)
• references/troubleshooting.md — common errors and fixes
• references/marketing-avatars.md — preset vs custom avatars
• references/marketing-products.md — URL fetch vs manual product create
• references/marketing-setup-items.md — hooks/settings discovery and usage
• references/marketing-ad-references.md — ad reference videos (create/list/get)
• references/marketing-brand-kits.md — brand kits (fetch from URL, list, get)
• references/marketing-dtc-ads.md — DTC Ads Engine (dtc-ads generate)
• references/marketing-modes.md — every Marketing Studio mode