FLUX image detector
FLUX models from Black Forest Labs are usually run through ComfyUI or a similar front-end, and those tools commonly embed the full generation workflow into the output PNG. When that block survives, identification is immediate and certain.
FLUX is also a good illustration of why classifier-only detection ages badly: it is newer than the training data behind most public detectors, so purely statistical detection performs noticeably worse on it than on older diffusion output. The metadata layer matters more here, not less.
Drag a file here, or
JPEG, PNG, WebP, AVIF, PDF or .txt · up to 10 MB · you can also paste a screenshot
By uploading you agree to our Terms and Privacy Policy. Your file, IP address, approximate location and device details are stored.
What this checks
- ComfyUI workflow JSON embedded in PNG text chunks
- Metadata naming FLUX or Black Forest Labs
- FLUX's typical output dimensions
- Image classifiers, with the caveat that FLUX postdates their training data
Frequently asked questions
Why is FLUX harder to detect than older models?
Public detectors were largely trained on earlier diffusion output. FLUX's artefact profile differs enough that classifier accuracy drops, which is exactly why embedded metadata carries more of the weight.
Does FLUX add a watermark?
The open-weight releases do not embed an invisible watermark of the kind Google applies. Detection depends on metadata and statistics.
What does the ComfyUI workflow show?
When present, the embedded JSON records the full node graph — model, prompts, samplers and seeds — which is conclusive about how the image was produced.