flux-1-schnell

FLUX.1-schnell (KNOT)

Apache-2.0 text-to-image model from Black Forest Labs, mirrored in the sovereign KNOT format for the Aether edge-inference runtime on Edgework.ai.

Model Details

This repository mirrors the FLUX.1-schnell weights as a single .knot container (plus *.knot.manifest.json and *.knot.admission.json sidecars). It is a format mirror of the upstream model under its original Apache-2.0 license — not a new or fine-tuned model.

Components

The KNOT carries the FLUX component tensors used by the runtime: the flow transformer (transformer), the autoencoder (vae), and the text encoders (text_encoder, text_encoder_2). The manifest sidecar records the tensor inventory; the admission sidecar records the artifact's quality-gate status.

Status

The artifact is shape-admitted (weights load and produce finite, correctly shaped tensors). Full prompt-faithful, visible-image generation through the runtime is tracked separately in the project's admission ledger — treat this repo as the canonical weight mirror, not a turnkey image endpoint.

About

Published by forkjoin-ai · Managed by @buley. We mirror production weights in the KNOT format for distributed edge inference via the Aether runtime.

• All models · GitHub · Edgework.ai

Also available: .knot (sovereign format)

This repo ships flux-1-schnell.knot — the model weights in the KNOT container that the Aether distributed-inference runtime loads natively (the GGUF, when present, sits right beside it). A KNOT is a single self-describing file with a JSON table-of-contents, so any single tensor is one HTTP Range request — ideal for streaming weights to edge nodes.

huggingface-cli download forkjoin-ai/flux-1-schnell flux-1-schnell.knot --local-dir ./knots

Full format spec: KNOT_FORMAT.md. Inspect the header with bun run open-source/bitwise/scripts/dump-knot.ts flux-1-schnell.knot.