Photoreal

FLUX Schnell

Black Forest Labs' fastest open-weight FLUX.1 model

FLUX Schnell (FLUX.1 [schnell]) is a 12-billion-parameter text-to-image model from Black Forest Labs. It makes an image in 1 to 4 steps, and its weights are Apache 2.0, free for personal and commercial use. On FluxNote it costs 1 credit per image.

Fast draftsPrompt testingHigh-volume social images

Free to start. No credit card required.

See pricing

Cost per image

1 credit

Free renders

~100 on free plan

Watermark

None on paid plans; free downloads carry a small FluxNote.io mark

Animate to video

1-click, 5-10s

What is FLUX Schnell?

FLUX Schnell, officially FLUX.1 [schnell], is a text-to-image model that Black Forest Labs released on 1 August 2024. It is a 12-billion-parameter rectified flow transformer from the same FLUX.1 family as FLUX.1 [dev] and FLUX.1 [pro], trained with latent adversarial diffusion distillation so that it can produce an image in 1 to 4 sampling steps. Schnell is German for fast, and BFL describes it as its fastest model, tailored for local development and personal use.

Its weights are published on Hugging Face under the Apache 2.0 license, which allows personal, scientific and commercial use. That is the biggest practical difference from FLUX.1 [dev], whose weights come under the FLUX.1 [dev] Non-Commercial License for non-commercial, non-production use (Dev's outputs can still be used commercially under that license's terms). Downloading Schnell costs nothing, although Hugging Face asks you to share contact details first. Running it is another matter: the main checkpoint is 23.78 GB in BF16 and the T5 text encoder adds about 9.5 GB, which is why community quantized versions and hosted APIs are popular.

Schnell also behaves differently from Dev. It is timestep-distilled, so the Hugging Face Diffusers documentation runs it at 4 steps with the guidance scale set to 0, while guidance-distilled Dev needs about 50 steps. With guidance fixed at 0 there is no classifier-free guidance strength or negative prompt to adjust. And Schnell reads at most 256 text tokens, while Dev has no such cap, so long paragraph prompts lose their ending. Like every FLUX.1 variant, it supports a range of aspect ratios between 0.1 and 2.0 megapixels.

Is it good? At launch, BFL's own benchmark put Schnell ahead of other few-step models and of Midjourney v6.0 and DALL·E 3 (HD). That claim comes from the model's maker in August 2024, and BFL has since released its FLUX.2 family. The model card itself warns that Schnell may fail to match a prompt and that prompt following depends heavily on prompting style. In practice it is strong for drafts, iteration and volume, while Dev or a FLUX.2 model is the better choice when detail and long, specific prompts matter.

You can run Schnell locally in ComfyUI or Diffusers, through hosted APIs such as fal (listed at $0.003 per megapixel when checked in October 2026), Replicate or Cloudflare Workers AI (default 4 steps, maximum 8), or inside apps that wrap it. On FluxNote it is one of the 1-credit image models: the Free plan includes 100 Image Studio credits with no card, enough for about 100 Schnell images, and paid plans add commercial usage rights for social posts, ads and client work.

Spec sheet

Hard numbers, what FLUX Schnell accepts and what it produces.

Resolution

0.1 to 2.0 MP; FluxNote samples are 1024×1024 and 736×1280

Generation time

1 to 4 sampling steps; seconds per image

Aspect ratios

1:1 and 9:16 in FluxNote samples; the model supports other ratios

Inputs

Text prompt, up to 256 tokens

Real outputs from FLUX Schnell

12 samples generated by FluxNote users. Hover any image to read the prompt.

make a grandpa dunk on lebron james 16:9

Quando você tenta começar a vida fitness na segunda-feira

DURAÇÃO:

3040 segundos (perfeito pra viralizar)

ROTEIRO + CENAS
CENA 1 (05s)

Pessoa olhando no espelho, motivada
Fala:
Agora vai segunda-feira eu começo de verdade!

Prompt (IA - inglês):
A determined young person looking in the mi — Quando você tenta começar a vida fitness na segunda-feira DURAÇÃO: 3040 segundos (perfeito pra viralizar) ROTEIRO + CENAS CENA 1 (05s) Pessoa olhando no espelho, motivada Fala: Agora vai segunda-feira eu começo de verdade! Prompt (IA - inglês): A determined young person looking in the mi

Slight cinematic push-in, sculpting tool exits as a curing light moves into position, emitting a soft blue glow that intensifies, the composite surface subtly shifts from soft to hardened with a polished sheen forming

Trying to get pregnant but its not happening? Heres why

"Your body needs the right conditionshealthy & balanced hormones, regular ovulation, and low stress. But today, things like poor diet, stress, irregular cycles, and even sleep problems can silently affect fertility!

"And Most women dont eve — Trying to get pregnant but its not happening? Heres why "Your body needs the right conditionshealthy & balanced hormones, regular ovulation, and low stress. But today, things like poor diet, stress, irregular cycles, and even sleep problems can silently affect fertility! "And Most women dont eve

Ancient Arabian era, cinematic style, warm lighting, highly detailed. Men wear traditional galabiyas, cloaks, and turbans. The woman is very beautiful, wearing a long modest red dress and a white hijab, elegant and covered, emotional cinematic atmosphere. المشهد 1
A beautiful young woman enters a ga — Ancient Arabian era, cinematic style, warm lighting, highly detailed. Men wear traditional galabiyas, cloaks, and turbans. The woman is very beautiful, wearing a long modest red dress and a white hijab, elegant and covered, emotional cinematic atmosphere. المشهد 1 A beautiful young woman enters a ga

Top-down view, figures peel off from a central point moving in different directions simultaneously like a slow explosion of intent, each group moves with purpose in their assigned direction, the city map below them suggests the grid of Mumbai streets, camera zooms very slowly out to show the full ex

HOOK 0:000:03
If you hear knocking at your window tonight dont look.

STORY 0:030:15
Last night at 2:17 AM, I heard itthree slow knocks.
My room is on the second floor. Theres no balcony.
I froze then my phone buzzed.
It was a text from an unknown number:
Stop pretending you dont see m — HOOK 0:000:03 If you hear knocking at your window tonight dont look. STORY 0:030:15 Last night at 2:17 AM, I heard itthree slow knocks. My room is on the second floor. Theres no balcony. I froze then my phone buzzed. It was a text from an unknown number: Stop pretending you dont see m

Background: summer gradient with light rays

حبيبتي أنا مستني اليوم ده من زمان يا عمري.

FLUX Schnell: strengths & limitations

An honest picture of what FLUX Schnell does well, and where it doesn't. Use this to decide whether to pick FLUX Schnell or one of FluxNote's 18 other image models for a given job.

Strengths

Few-step generation: trained with latent adversarial diffusion distillation to produce an image in 1 to 4 steps, where FLUX.1 [dev] takes about 50.
Apache 2.0 weights: the model can be used for personal, scientific and commercial purposes, unlike FLUX.1 [dev], whose weights are licensed for non-commercial use.
Full FLUX.1 architecture: the same 12B rectified flow transformer family as Dev and Pro, with outputs from 0.1 to 2.0 megapixels across aspect ratios.
Lowest-cost tier on FluxNote at 1 credit per image, so the Free plan's 100 Image Studio credits cover about 100 drafts.

Limitations

Prompts are capped at 256 text tokens (Dev has no such cap), so long, paragraph-style prompts lose their final details.
Guidance is fixed at 0: there is no guidance scale or negative prompt to tune, so you steer it only through the wording of the prompt.
Black Forest Labs positions FLUX.1 [pro] and [dev] as its higher-quality variants; for detailed hero images, FLUX Dev or FLUX 2 Pro is the better pick.
The model card warns it may fail to match a prompt and that prompt following depends heavily on prompting style, so expect to re-roll some images.

Photoreal

Try FLUX Schnell yourself

FLUX Schnell costs 1 credit per image. Free downloads carry a small FluxNote.io mark; paid plans download without it.

3 ways creators use FLUX Schnell

Real-world workflows we see most often in the FluxNote dashboard.

Find the prompt before a premium render

Scenario:You need one polished product image from FLUX 2 Pro (4 credits per image) but have not settled the angle, background or lighting.

Walkthrough:Run eight short variations on FLUX Schnell for 8 credits in total, changing one element at a time. Keep the composition that works, then send the same prompt to FLUX 2 Pro for the final render.

Vertical frames for a faceless video

Scenario:You are storyboarding a 9:16 short and want one still per scene before spending credits on animation.

Walkthrough:Generate one 9:16 frame per scene with FLUX Schnell (FluxNote's vertical samples come out at 720×1280 or 736×1280), drop the weak frames, then animate only the keepers with Animate to Video, which is a paid, credit-based feature.

A month of square social images

Scenario:A small brand wants one fresh image a day for its feed without paying premium-model prices for every post.

Walkthrough:Thirty 1:1 images at 1 credit each use 30 credits, a small share of the 2,500 monthly credits on the Rise plan. Re-run the best few on FLUX Dev or FLUX 2 Pro if they need more detail.

Prompt examples that work with FLUX Schnell

Copy-paste any of these prompts into the FLUX Schnell model to get a result close to the example.

Product shot

Matte black ceramic coffee mug on a light oak desk, soft morning window light from the left, gentle shadow, shallow depth of field, plain white wall behind

Vertical scene (9:16)

Lone hiker standing on a rocky ridge at sunrise, low fog filling the valley below, warm orange sky, wide shot from behind, cinematic

Short on-image text

Minimal concert poster with the headline "NIGHT SHIFT" in bold white sans-serif at the top, neon pink and teal gradient background, silhouette of a saxophone player

How to get the best results from FLUX Schnell

Prompt strategies that consistently improve output quality, based on how FLUX Schnell was trained and what it responds to.

Stay under 256 tokens

Schnell's text input is capped at 256 tokens, so the end of a long prompt is dropped. Put the subject first, then setting, lighting and style, and cut repeated adjectives: "red vintage bicycle against a yellow wall, late afternoon sun, 35mm photo" keeps every detail inside the limit.

Describe what you want, not what to avoid

Schnell runs with guidance fixed at 0, so there is no negative prompt field or guidance slider. Write "plain white background" rather than listing the things you do not want in the frame.

Keep the default 4 steps

If your tool exposes a steps setting, 4 is the default in BFL's Diffusers example and on fal and Cloudflare Workers AI. Cloudflare allows up to 8 and notes that more steps can improve quality but take longer.

Change one thing at a time

When a result is close, keep the rest of the prompt identical and change a single phrase, such as the lighting or the camera angle. If your tool exposes a seed, lock it so the comparison is fair.

Quote short on-image text

For words in the image, put the exact text in quotes and keep it to a few words, then check the spelling before publishing; the model card notes that outputs may not match every prompt.

Draft here, finish on a slower model

Use Schnell at 1 credit to settle composition and wording, then re-run the same prompt on FLUX Dev (2 credits) or FLUX 2 Pro (4 credits) for the final image.

FLUX Schnell vs other FluxNote models

Quick reference for picking the right model, every alternative below ships in the same FluxNote subscription.

vs FLUX Dev

2 credits. Guidance-distilled sibling that takes about 50 steps, accepts longer prompts and adds detail; its open weights are licensed for non-commercial use only.

vs FLUX 2 Pro

4 credits. FluxNote's higher-detail FLUX option for final renders; draft on Schnell first, then re-run the winning prompt here.

vs FLUX 2 Max

8 credits. FluxNote's top FLUX tier for complex scenes; worth it only for the image you will actually publish.

How to generate with FLUX Schnell

Four steps from prompt to publish-ready output. Total time: under 30s.

STEP 01

Open Image Studio

Create a free FluxNote account (no card needed) and open Image Studio. The Free plan includes 100 credits for trying Image Studio.

STEP 02

Choose FLUX Schnell

Pick FLUX Schnell from the image model list. It costs 1 credit per image, so you can try several prompt variations cheaply.

STEP 03

Write a short, positive prompt

Lead with the subject, then setting, light and style, and stay under the 256-token cap. Describe what you want rather than what to avoid, because Schnell has no negative prompt.

STEP 04

Keep, refine or animate

Download the keeper, edit it with FLUX Kontext Edit, upscale it with Nucleus Image, re-run the prompt on FLUX Dev or FLUX 2 Pro for more detail, or turn it into a clip with Animate to Video (paid credits).

Why creators pick FluxNote for FLUX Schnell

FLUX Schnell is one of 19 image models on FluxNote. One subscription unlocks them all.

1 credit per FLUX Schnell image, drawn from one shared monthly credit pool that also covers video, UGC, audio and assistant work.

The Free plan includes 100 Image Studio credits a month with no card, enough for about 100 Schnell drafts.

Switch models without new accounts: FLUX Dev, FLUX 2 Pro, FLUX 2 Max, Imagen 4 Fast, Seedream and the FLUX Kontext editors sit in the same Image Studio.

Paid plans include commercial usage rights for social posts, paid ads and client work. Rise costs $20 a month billed monthly, or $14 a month billed quarterly.

Stills can move straight into FluxNote's video tools, including Animate to Video, the timeline editor and scheduled social publishing, without exporting and re-uploading.

FLUX Schnell FAQ

The 10 questions creators ask most often before switching to FLUX Schnell.

Is FLUX Schnell free?

The model is free: FLUX.1 [schnell] weights are released under Apache 2.0 and can be downloaded from Hugging Face after you share your contact details. Running it is not automatically free. Locally you need a GPU that can handle a 23.78 GB checkpoint (or a quantized version), and hosted APIs charge per image; fal, for example, lists $0.003 per megapixel. On FluxNote it costs 1 credit per image, and the Free plan's 100 monthly Image Studio credits cover about 100 images.

What does "flux schnell" mean?

Schnell is German for fast or quick. Black Forest Labs, which has labs in Freiburg and San Francisco, uses it for the fastest FLUX.1 variant. The official name is FLUX.1 [schnell]; its siblings are FLUX.1 [dev] and FLUX.1 [pro].

Is FLUX.1 Schnell good?

For speed, yes. It makes usable images in 1 to 4 steps, and at launch in August 2024 BFL's own benchmark placed it ahead of other few-step models and of Midjourney v6.0 and DALL·E 3 (HD). Its limits are real: prompts are capped at 256 tokens, there is no guidance or negative prompt control, and the model card notes it may not match every prompt. Use it to explore and iterate, then switch to FLUX.1 [dev] or a FLUX.2 model for detailed final images.

What is the difference between FLUX Dev and Schnell?

Both are 12B FLUX.1 models from Black Forest Labs. Schnell is timestep-distilled for 1 to 4 steps, runs with a guidance scale of 0, accepts up to 256 prompt tokens and is licensed Apache 2.0. Dev is guidance-distilled, takes about 50 steps, supports adjustable guidance (3.5 in the reference example) and longer prompts, and gives more detail; its weights are under a non-commercial license, although its outputs may be used commercially under that license's terms. On FluxNote, Schnell costs 1 credit and Dev 2.

Can I use FLUX Schnell images commercially?

The Apache 2.0 license lets you use the model itself for commercial purposes, so a self-hosted Schnell can power commercial work. If you generate through a service, that service's terms also apply. On FluxNote, paid plans include commercial usage rights for social posts, paid ads and client work, while the Free plan is for personal and evaluation use.

What resolution does FLUX Schnell support?

Black Forest Labs says all FLUX.1 variants support aspect ratios and resolutions between 0.1 and 2.0 megapixels; 1024×1024 is about 1 megapixel. The sample outputs on this page, generated in FluxNote, are 1024×1024 for square images and 720×1280 or 736×1280 for vertical 9:16 images.

How many steps should I use with FLUX Schnell?

Four is the default in the Hugging Face model card example, on fal and on Cloudflare Workers AI, and the model is trained to work in 1 to 4 steps. Cloudflare allows up to 8 and notes that higher values can improve quality but take longer.

Can I run FLUX Schnell locally?

Yes. Black Forest Labs publishes reference inference code on GitHub, and Schnell works in ComfyUI and Hugging Face Diffusers. The full-precision files are large (23.78 GB for the main checkpoint plus about 9.5 GB for the T5 text encoder), so the Diffusers example turns on CPU offloading to save VRAM, and community quantized versions exist for smaller GPUs.

Does FluxNote add a watermark to FLUX Schnell images?

No. FluxNote's image models export without a watermark, including on the free tier.

Can I turn a FLUX Schnell image into a video?

Yes, with FluxNote's Animate to Video, which uses your still as the starting frame and a motion prompt to direct the movement. It is a paid, credit-based feature, and the current model cost is shown before you render. Model labels in the product include Seedance 1.5, Runway Gen-4, Hailuo 02 and Kling 2.6, subject to availability.