FLUX — Black Forest Labs' open-weight image model family known for realistic text and hands
What it is
FLUX is the image-model family from Black Forest Labs, built by the original Stable Diffusion team. The open-weight FLUX.1 variants (schnell under Apache 2.0, dev non-commercial) run locally in ComfyUI or diffusers and are known for readable text and believable hands; the current FLUX 3 flagship is multimodal, covering image, video and audio with native speech and up to 20-second clips. Larger checkpoints want serious VRAM, and the fine-tuning ecosystem is still younger than Stable Diffusion's.
Editor's review
Long-form introduction by the BetterPicker editors · checked against the official site · Oct 18, 2026
FLUX is the image-model family from Black Forest Labs, founded by the team behind the original Stable Diffusion, and it made its name on the two things image models usually fumble: readable text and believable hands. The open-weight FLUX.1 lineage still anchors the local-generation scene - schnell runs in one to four steps under an Apache 2.0 license you can ship commercially, dev trades speed for the highest open quality under a non-commercial license, and both run in ComfyUI, diffusers or one-click installers on an RTX GPU. Above them, the current flagship FLUX 3 is a multimodal model spanning image, video and audio, with native speech, twenty-second clips and action prediction aimed at robotics. The family is powerful, young and moving fast, and the reviewer walkthrough gets a first-run image with fully accurate lettering from a plain desktop install.
What it does well
Text and hands that survive zooming: prompt following and photoreal lighting match closed alternatives according to the model card, and rendered signage, labels and fingers hold up where older open models dissolved into mush - the reason it became the default for mockups, thumbnails and any image that must contain words.
Open weights that run anywhere: schnell and dev download to your own hardware through ComfyUI or diffusers, one-click installers bundle Python, Git and the rest of the dependency chain, and API partners such as Replicate and fal.ai rent the same models by the second when no GPU is handy.
A live roadmap: FLUX 3 unifies image, video and audio generation with multilingual speech and ambience in one take, starts from text, an image or keyframes, and the model card stays candid about limits - it is not a factual-knowledge source and can amplify biases.
Who it's for
Local-generation hobbyists with RTX-class GPUs, ComfyUI workflow builders who want a model they can fine-tune, chain and ship inside their own pipelines, developers integrating image generation through APIs without maintaining hardware, and product teams that need legible text inside generated images - mockups, thumbnails, packaging concepts - where older open models dissolved words into noise.
Where it falls short
Heavyweight hardware: the quality tiers want serious VRAM, and consumer GPUs feel the weight on the larger checkpoints - the reviewer recommends RTX-class cards outright for a reason.
A younger ecosystem: fine-tunes, LoRAs and community tooling trail Stable Diffusion's mature library, and dev's non-commercial license narrows shipping work to schnell or paid APIs.
Prompt-style dependence: output quality swings with how you phrase prompts, and the model can simply fail a request - iteration is part of the budget, not a sign of a broken setup, so budget a few reruns per usable frame.
Specs at a glance
Facts from the official site · not editorial opinion
| What it is | Image-generation model family from Black Forest Labs (founded by the original Stable Diffusion team), spanning open-weight FLUX.1 variants and the multimodal FLUX 3 flagship |
|---|---|
| Open weights | FLUX.1 schnell: 12B-parameter rectified flow transformer, 1-4 step generation, Apache 2.0 license for personal, scientific and commercial use |
| Quality tier | FLUX.1 dev: higher open-weight quality under a non-commercial license; pro and newer models served via API (bfl endpoints, Replicate, fal.ai and others) |
| Local run | Runs in ComfyUI node workflows and the diffusers library; one-click installers bundle Python, Git and dependencies; NVIDIA RTX GPUs recommended |
| FLUX 3 | Multimodal model covering image, video and audio with native multilingual speech, up to 20-second clips in one generation, and action prediction for robotics |
| Strengths | Cutting-edge output quality with competitive prompt following; readable text rendering and believable hands |
| Limitations | Not a factual-knowledge source; may amplify societal biases; may fail to match prompts; prompt following depends heavily on prompting style |
Frequently asked questions
▸Can I use FLUX commercially?
FLUX.1 schnell is Apache 2.0, so yes for personal, scientific and commercial use. FLUX.1 dev is non-commercial; commercial work at that quality tier goes through paid API endpoints.
▸Do I need a GPU to run FLUX?
For local runs an NVIDIA RTX card is recommended and the larger checkpoints want serious VRAM. Without a GPU, the same models are available per-second through API partners such as Replicate and fal.ai.
▸How fast is FLUX.1 schnell?
Distillation training lets it produce high-quality images in one to four steps, far fewer passes than standard diffusion models need, at some cost to peak quality.
▸What is FLUX 3?
The current flagship: one multimodal model for image, video and audio, with native speech and ambience, up to 20-second clips in a single generation, and action prediction aimed at robotics.
Video reviews
4 review videos aggregated · praise and criticism included alike · click through to the original video
Channels that covered it
Channels are aggregated as sources only — we don’t rate creators
Related tools
Where to go next
External links open in a new tab; external content is independent of this site.
Link down? Every object page is re-checked monthly.