generate-all-content-types

Higgsfield Generate All Content Types

Purpose

Drive higgsfield.ai — an AI-native creative suite — to generate every content type it offers from a signed-in account: still Images (Higgsfield Soul), Videos, and Audio (voiceover / voice-change), plus the secondary generators (Effects, Cinema Studio, 3D Jutsu, Canvas, Edit/Layers, Marketing Studio). Every generator shares the same shape: compose a prompt → pick a model/settings → click Generate. This skill documents that workflow, the credit cost of each generator, and — critically — the account/login gate that anonymous sessions hit. It is write / credit-consuming (each Generate spends account credits), so it requires real account credentials to complete end-to-end.

When to Use

• "Generate an image / video / voiceover on Higgsfield with our account."
• "Run one generation through each Higgsfield content type to smoke-test our subscription."
• A pipeline that produces marketing creative (image + video + audio) from prompts on Higgsfield.
• You need to know the credit cost and model options for each Higgsfield generator before committing an account's credits.

Workflow

Prerequisite — account credentials. Every generator requires being logged in; anonymous users can compose but cannot see or download output. Supply the account's login (Google / Apple / Microsoft OAuth, or email) via your credential store. No credentials were available in the run that produced this skill, so the post-login generation + download steps below are documented from the UI structure but were NOT executed end-to-end. Do not fabricate output — if login fails, stop and report the auth failure.

1. Open a stealth remote session

sid=$(browse cloud sessions create --keep-alive --verified --proxies \
  | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
export BROWSE_SESSION="$sid"

Both --verified and --proxies are required — higgsfield.ai fronts with Cloudflare + Cloudflare WAF + DataDome. A bare session risks a challenge/block.

2. Sign in first (do this once, before composing)

browse open "https://higgsfield.ai/ai/image?model=gpt_image_2" --remote
browse wait load --remote
# Dismiss the cookie banner: click "Accept All Cookies"
# Click the header "Login" (or "Sign in") button to open the auth dialog.

The auth dialog ("Welcome to Higgsfield") offers Continue with Google / Apple / Microsoft / Email. Complete whichever matches the stored account. New business-email signups are advertised 50 free credits. After login, the header shows the account avatar/credit balance instead of Login/Sign up — verify this before generating.

3. Generate — Image (/ai/image)

browse open "https://higgsfield.ai/ai/image?model=gpt_image_2" --remote
browse wait load --remote
# Type the prompt into the "Describe the scene you imagine" textbox.
browse fill "textarea" "a serene mountain lake at sunrise, cinematic"
# Optional: change model (GPT Image 2 / Nano Banana Pro / Seedream…),
#   quality (High / 2K), and image count (1–4) via the settings row.
# Click the "Generate 8.5 6.5" button (8.5 = full credit cost, 6.5 = discounted).
browse click "@Generate"
browse wait timeout 8000 --remote   # generation is async; poll the result grid
browse screenshot --remote --path result-image.png

Select the model by setting the model= URL query param (e.g. ?model=gpt_image_2, ?model=nano_banana_pro) or clicking the model button in the settings row.

4. Generate — Video (/ai/video)

browse open "https://higgsfield.ai/ai/video" --remote
browse wait load --remote
# Optional: "Add references Image, Video or Audio" to condition the video.
browse fill "textarea" "slow dolly-in on a neon-lit Tokyo alley at night"
# Pick a model via the "Model" button (Genjutsu, Seedance, etc.);
#   Genjutsu is also reachable via /ai/video?model=genjutsu.
# Click "Generate 80 60" (video costs far more — ~80 credits full / 60 discounted).
browse click "@Generate"
browse wait timeout 15000 --remote   # video renders take longer than images

5. Generate — Audio (/audio)

browse open "https://higgsfield.ai/audio" --remote
browse wait load --remote
# Choose a mode tab: "Text to Speech" or "Voice Change".
# For TTS: type the script into "Write exactly what the voice will read out loud",
#   optionally describe the voice in the "Voice details" box, pick model (Seed Audio 1.0).
# For Voice Change: "Upload media" (up to 3 voices/audios or an image) first.
browse fill "textarea" "Welcome to Higgsfield. This is a generated voiceover."
browse click "@Generate"
browse wait timeout 10000 --remote

6. Other content types (same compose → Generate pattern)

7. Collect output and release

After each Generate, the finished asset appears in the result area / the account's project history. Read the CDN URL from the rendered <img>/<video> (cdn.higgsfield.ai/...) or download via the UI. Then:

browse cloud sessions update "$sid" --status REQUEST_RELEASE

Site-Specific Gotchas

• Account login is mandatory to see any output. Anonymous users CAN type a prompt and click Generate, but instead of an image they get a blurred "Generated preview" with an overlay: "Create an account to see generated image — Sign up and generate for free." The task cannot be completed without real account credentials. This was the terminal state in the run that built this skill (no credentials were provided).
• Generate spends credits — this is a write action. The button label doubles as the price tag: Image Generate 8.5 6.5, Video Generate 80 60 (first number = full cost, second = current-promo discounted cost). Do not loop generations blindly; each one bills the account. Video is ~10× the cost of an image.
• Anti-bot: Cloudflare + Cloudflare WAF + DataDome. --verified --proxies got through cleanly in testing (homepage, /ai/image, /ai/video, /audio all rendered). A bare session is not safe. If you hit a challenge, recreate the session rather than retrying in place.
• browse type broke on multi-word input in testing (Error: Unexpected arguments) — the space-delimited args were mis-parsed. Use browse fill "textarea" "<full prompt>" (quoted) instead; the page has a single primary textarea per generator so the textarea selector is reliable.
• browse fill did NOT auto-submit here (pressedEnter: false), which is what you want — pressing Enter mid-prompt is not required and the Generate button is the only submit path.
• Cookie consent dialog overlays the UI on first load. Click "Accept All Cookies" (or "Reject Advertising Cookies") before interacting, or clicks land on the banner.
• Model selection via URL query param. Many generators accept ?model=<id> (e.g. /ai/image?model=gpt_image_2, /ai/video?model=genjutsu) — faster and more deterministic than clicking the model dropdown.
• SPA rendering is progressive. After browse wait load, add browse wait timeout 2000-3000 before snapshotting — the compose form and settings row hydrate a beat after load.
• Generation is asynchronous. Generate does not block; the asset streams into the result grid seconds (image) to tens of seconds (video) later. Poll with wait timeout + re-snapshot rather than expecting an immediate result.
• Prefer the API/MCP for headless account generation. Higgsfield ships an official developer API (console.higgsfield.ai / open.higgsfield.ai) and an MCP server (/mcp). With an API key these are more robust than driving the DOM. Neither was verified in this run (no key available) — treat as unconfirmed leads, not guaranteed paths.
• Auth methods offered: Continue with Google, Apple, Microsoft, or Email. There is no visible username/password form until you pick "Continue with Email".

Expected Output

// Per-generation success shape (requires a logged-in account)
{
  "site": "higgsfield.ai",
  "logged_in": true,
  "content_type": "image",            // image | video | audio | effect | cinema | 3d | canvas | edit | marketing
  "url": "https://higgsfield.ai/ai/image?model=gpt_image_2",
  "model": "gpt_image_2",
  "prompt": "a serene mountain lake at sunrise, cinematic",
  "credits_charged": 8.5,
  "status": "completed",
  "assets": [
    "https://cdn.higgsfield.ai/card/<uuid>.webp"
  ]
}

// Terminal state actually observed in this run (no account credentials available)
{
  "site": "higgsfield.ai",
  "logged_in": false,
  "content_types_available": [
    { "type": "image", "url": "/ai/image", "generate_credits": "8.5 / 6.5", "auth_required": true },
    { "type": "video", "url": "/ai/video", "generate_credits": "80 / 60", "auth_required": true },
    { "type": "audio", "url": "/audio", "modes": ["text_to_speech", "voice_change"], "auth_required": true },
    { "type": "effects", "url": "/effects/use", "auth_required": true },
    { "type": "cinema_studio", "url": "/generate", "auth_required": true },
    { "type": "3d_jutsu", "url": "/3d-jutsu", "auth_required": true },
    { "type": "canvas", "url": "/canvas", "auth_required": true },
    { "type": "edit_layers", "url": "/layers", "auth_required": true },
    { "type": "marketing_studio", "url": "/marketing-studio", "auth_required": true }
  ],
  "auth_gate": "Clicking Generate while anonymous shows a blurred preview with 'Create an account to see generated image / Sign up and generate for free'.",
  "auth_methods": ["google", "apple", "microsoft", "email"],
  "status": "blocked_needs_account",
  "note": "Compose + Generate UI reachable anonymously; output and download require login + credits. No credentials were supplied to this run."
}