fluidghost

Official Python client for the FluidGhost API.

FluidGhost turns one photo or video into a set of distinct variants — each a fresh, standalone capture with its own metadata, its own pixels, and a device-authentic filename. This SDK spoofs media with your API key, over the submit → poll → download loop (a single photo, a video, or a whole carousel), and inspects metadata (see Inspect metadata).

• Full API docs: https://ghost.fluidvip.com/docs/
• Python 3.8+ (depends on requests).

Install

pip install fluidghost

Authentication

Mint an API key in the dashboard (Settings → API keys at https://ghost.fluidvip.com). Keep it server-side — never ship it in client code. A secret API key can submit jobs, read and download their results, and inspect metadata (management keys only provision).

Spoof a photo or video

If your API key has a saved preset bound to it (pick it when you create the key in the dashboard), the key already carries the settings — the whole call is just the image:

job = fg.run(image="photo.jpg")

Otherwise name the settings on the call, as below. Note that recipe_id, recipe_config and preset_id override the key's bound preset.

run() submits a job and waits for it to finish. Results are one-time links: download_variant() streams the bytes and the object is deleted on a successful download (a second call raises FluidGhostError with status 410). Photos and videos both go through the same call — the type is detected from the file.

import os
from fluidghost import FluidGhost

fg = FluidGhost(api_key=os.environ["FLUIDGHOST_KEY"])

job = fg.run(
    image="photo.jpg",                 # or a video: "clip.mp4"
    recipe_id="full-refresh",
    copies=3,
    options={"noise": 35, "ultraSafe": True},
    on_poll=lambda j: print(j["status"], j.get("progress")),
)

for v in job["variants"]:
    if v["status"] != "completed":
        continue
    data = fg.download_variant(job["jobId"], v["index"])
    with open(v.get("fileName") or f"variant-{v['index']}.jpg", "wb") as f:
        f.write(data)
    print(v.get("fileName"), "risk", v.get("detectionRisk"), v.get("riskBand"))

You can pass in-memory bytes instead of a path with a (filename, bytes) tuple:

fg.create_job(image=("photo.jpg", my_bytes), recipe_id="full-refresh")

Spoof a carousel (matched set)

A carousel is a set of photos posted together. spoof_carousel() submits one linked job per file — sharing a session and a running filename sequence so the outputs read as one cohesive post — and waits for all of them.

jobs = fg.spoof_carousel(
    ["slide1.jpg", "slide2.jpg", "slide3.jpg"],
    recipe_id="full-refresh",
)

for job in jobs:
    v = job["variants"][0]
    data = fg.download_variant(job["jobId"], v["index"])
    with open(v["fileName"], "wb") as f:
        f.write(data)

For a fully cohesive set (the same device across every slide), pass a recipe_config or preset_id that pins a device — otherwise each slide picks its own device family.

Inspect metadata

Four read-only calls read metadata without creating a job. They need the inspect:read scope (secret API keys have it by default) and share one rate limit of 60 requests per minute per owner (a 429 raises FluidGhostError with retry_after). Every file argument takes a path or a (filename, bytes) tuple, like run().

# Metadata of an uploaded photo or video, grouped by container.  POST /api/inspect
meta = fg.inspect("photo.jpg")

# Deep forensic verdict (metadata, container, C2PA).  POST /api/inspect/forensics
# Fail-open: a missing engine returns {"status": "unavailable"} or {"status": "error"}.
report = fg.inspect_forensics("photo.jpg")

# Perceptual (Meta-PDQ) compare of an original and a variant.  POST /api/inspect/compare
cmp = fg.inspect_compare("original.jpg", "variant.jpg")
print(cmp["detectionRisk"], cmp["band"], cmp["matchThreshold"])

# Metadata of a finished variant, read server-side (does not consume the one-time
# download link).  GET /api/inspect/result/{jobId}/{index}
variant_meta = fg.inspect_result(job["jobId"], 0)

Settings

Provide exactly one recipe source (or bind a default preset to your key and omit all three; any of the three you do send overrides the key's preset):

Errors

Every non-2xx response raises FluidGhostError, mirroring the API's {error, message, details}:

from fluidghost import FluidGhost, FluidGhostError

try:
    fg.run(image="photo.jpg", recipe_id="full-refresh", copies=3)
except FluidGhostError as e:
    print(e.status, e.code, e.details)
    if e.code == "rate_limited":
        time.sleep(e.retry_after or 5)
    if e.code == "insufficient_balance":
        ...  # top up in the dashboard

Configuration

FluidGhost(
    api_key="fg_sk_live_…",
    base_url="https://api-ghost.fluidvip.com",  # default; override for a proxy
    timeout=60.0,
)

License

MIT