Resources // Voice & Multimodal AI
Black Forest Labs (FLUX)
- New
Offer FLUX image models, 1 credit = 1 cent
Suits for Non-VC-backedCreatorsOpen-source
Updated Jun 2026
The offer
- Open weights: FLUX.1 [schnell] is Apache-2 (free for commercial use) and the FLUX.1 / FLUX.2 [dev] weights are free to self-host (dev is non-commercial)
- First-party FLUX API, credit-based: 1 credit = $0.01, prepaid via Stripe; same price in the API and Playground
- FLUX.2 pricing (by megapixel): klein from ~$0.014/image, pro from ~$0.03, max from ~$0.07; FLUX.1 legacy: 1.1 Pro $0.04, Kontext max $0.08
- No automatic free credit: onboarding routes you to buy ~$10-20 of credits before you can generate
Who qualifies
- Anyone can register at the BFL dashboard (email confirmation); generating requires adding credits first
- Open-weight variants (schnell, dev) are free to self-host: schnell for commercial use, dev non-commercial
- The FLUX.2 API flags IP-adjacent prompts as "sensitive" even at the lowest setting
- No startup, accelerator, or grant program published
Community Insights
Black Forest Labs is widely respected for “doing open weights right.” It ships open weights plus training code with each commercial release and withholds only the top pro / max variants, which developers cite as the model others should follow. On FLUX.2 specifically, hands-on testers find better prompt adherence than FLUX 1.1 but mixed image quality versus rivals like Nano Banana, plus a notably stricter IP / safety filter. Most people reach FLUX through resellers (Replicate, fal) rather than BFL’s own API, partly because there’s no free trial on the first-party API.
Best Practices (from community tips)
- Prepay a small amount ($10-20) and scope experiments tightly. There’s no free credit.
- For FLUX.2, use JSON-structured prompts and hex color codes (officially encouraged) for better adherence.
- Expect the API to flag IP-adjacent prompts even at the lowest sensitivity; design prompts accordingly.
- If you don’t need pro / max quality, self-host the open weights (schnell is Apache-2; dev is non-commercial).
- Price-compare on Replicate / fal first. They sometimes hand out trial credits BFL doesn’t.
- Batch carefully: batch requests multiply the base cost by image count.