FLUX.1-schnell
Black Forest Labs · image gen · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
Few-step by design and Apache 2.0 from day one — the open sibling of dev; our 3060 box runs it quantized, but no official VRAM figure exists yet.
#8 of 20 in Image Gen
ToolScore breakdown — why 78/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸Few-step by design — the speed profile is the architecture
- ▸Apache 2.0 from day one
- ▸The open sibling of FLUX.1-dev — same family, none of the NC strings
- ▸Our 3060 box runs FLUX quantized — production-proven here
Where it costs you
- ▸No official VRAM figure exists yet — verify pending
- ▸Previous generation now that FLUX.2 is out
- ▸No AA Elo listing in this snapshot
Hidden traps — the stuff the pricing page doesn't say
The VRAM cell is “—” — verify pending. Our 3060 runs FLUX quantized, but that is a community-quant path; plan hardware on community numbers until an official figure lands.
Know the limits before you pay: read the traps above, then decide with eyes open.
Affiliate link where available — never affects the score.
- internal ToolScore: The License Traps Hiding in “Open” Image Models — nine verified traps, one US ban verified
- internal ToolScore: One GPU Beats Ten Subscriptions — the 3060 stack math verified
- internal ToolScore: FLUX vs Qwen-Image on a 3060 — our production notes from the 12GB box verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
The only API here that outputs real vectors ($0.08/img) — the logo/icon niche — with V4.1 Flash rast
The Apache half of the FLUX.2 klein pair — 4B params and one of the few models that publishes a real
The quality peak: AA Elo 1262, the top of the arena — at $0.21/img @1024 (AA-listed), roughly 10× Se
FLUX.1-schnell review
In-depth analysis · last updated 2026-11-12
Apache from day one
FLUX.1-schnell is the open sibling of dev, and the contrast is the point: schnell is Apache 2.0 from day one, while dev carries the non-commercial model license. Few-step by design — the speed profile is the architecture, not a benchmark fluke. Our 3060 box runs FLUX quantized, so this is a production path we use, not a download suggestion.
The economics
Watts only: ≈$0.03/hour on a 170 W card, re-rolls free. Few-step design compounds the local advantage — fewer steps per image means less electricity per keeper.
The open question
The one dark cell is hardware: no official VRAM figure exists yet (verify pending). We publish hardware requirements only when a source confirms them, so the cell stays “—” until then.
How the pillars land
Price/Value 10: watts only, few-step by design. Quotas & rate limits 10: none locally. Generation speed 9: few-step is the design — the model is the speed play. Training use & license 9: Apache from day one. API & weights 8: open weights, community quants, no official VRAM figure. Feature set 6: the open sibling of dev — lineage without proof; no AA Elo listing in this snapshot. Track record 7: the FLUX.1 line, previous generation but running in production here. Ease of use 7: works, but you are on community numbers for hardware. Pricing transparency 6: the missing official VRAM figure is the one thing it does not publish. Support & docs 6: model card and ecosystem — the neutral-ish end.
Verdict — 78/100
The license-clean way into the FLUX.1 line: fast by design, Apache forever, and proven on our own 12 GB box. It would score higher the day BFL publishes an official hardware figure.