October 13, 2026Pixyn Team
FLUX: A Review of the AI Image Generator (2026)
A breakdown of FLUX from Black Forest Labs: Pro 1.1, 1.1 Ultra, Kontext, FLUX.2 and dev, what the model is good at, a ready prompt and how to use it in Pixyn.
TL;DR. FLUX is a family of image models from Black Forest Labs (Germany). Its strong sides: photorealism, precise following of a long prompt and text in the frame, where FLUX consistently outperforms most generators. In Pixyn the catalog has FLUX Pro 1.1, 1.1 Ultra, FLUX Kontext, FLUX.2 (with Flex and LoRA) and FLUX.1 dev.
What FLUX is
FLUX is a lineup of image generation neural networks released by Black Forest Labs, a team of former Stable Diffusion authors. The first generation came out in 2024, and on November 25, 2025 the company presented FLUX.2, a 32-billion-parameter flow-matching model that combines text-to-image, reference-based editing and multi-image composition in a single checkpoint.
The key profile of FLUX is not "a pretty picture at any cost" but predictability: the model does what is written in the prompt, including small scene details, the relative position of objects and lettering. This puts it closer to production (advertising, product cards, mockups) than to art experiments.
Unlike Midjourney, which lives in Discord and its own web interface, FLUX is first of all a model with an API. In Pixyn you work with FLUX from the browser or Telegram.
Technical capabilities
Resolution and speed
FLUX Pro 1.1 Ultra generates images of up to 4 megapixels (about 2048×2048) and delivers such a frame in roughly 10 seconds. This is not a speed record among "fast" models, but for native 4MP it is a decent pace. FLUX.2 works with the same 4 MP bar for both generation and editing.
Text and typography
Readable text in the frame is historically a strong side of FLUX. In independent tests the model renders words without distortion in multi-word lettering noticeably more often than competitors. This makes it a working tool for banners, covers and interface mockups, where the letters need to add up to real words and not to pseudo-Latin.
References and editing
FLUX.2 accepts up to 10 reference images at once, to keep one character, product or brand style across the frames of a series. The edit mode changes an existing image: background replacement, local edits, object removal, style transfer. For instructions of the form "change X, keep Y" FLUX has a separate Kontext line, with its own breakdown (link below).
Versions in the Pixyn catalog
- FLUX Pro 1.1: the basic production version, a balance of quality and speed.
- FLUX Pro 1.1 Ultra: native 4 MP, maximum detail of photorealistic scenes and portraits.
- FLUX Kontext (Pro / Max / Multi): editing by instruction, "change X, keep Y", including from several references.
- FLUX.2 and FLUX.2 Flex: the new generation for generation and editing from references; Flex opens control over the number of steps and guidance, to change the "speed vs detail" balance. FLUX.2 LoRA applies a LoRA for your style.
- FLUX.1 dev: the open-weights version, in Pixyn for image-to-image.
The exact cost of each version in tokens is visible right in the Studio: the price is loaded from the database before the run, and plans are described on /pricing.
| Version | Resolution / speed | Strong side | Good for |
|---|---|---|---|
| FLUX Pro 1.1 | A balance of quality and speed | A production all-rounder | Everyday generation |
| FLUX Pro 1.1 Ultra | Native 4 MP, ~10 s | Maximum detail | Marketplace cards, portraits |
| FLUX Kontext | Edits by instruction | Changes only what you name | Editing a finished image |
| FLUX.2 / Flex / LoRA | 4 MP, flexible steps | Generation + edit + up to 10 references, LoRA | Series, reference-based editing, your style |
| FLUX.1 dev | Open weights | Image-to-image | Variants of a picture |
The table's takeaway: Ultra is taken for native 4 MP and detail, FLUX.2 for editing and references, Kontext for instruction edits, and FLUX.2 LoRA for your own style.
Where FLUX is good
Marketplace product cards. Marketplaces need a clean background, a centered product and accurate rendering of the shape. FLUX Pro 1.1 Ultra confidently holds perspective, lighting and material detail: the photo does not "float" on the texture of fabric or packaging.
Ad creatives with text. A banner with a real slogan, a price tag or a brand name is a scenario where FLUX outperforms most generators. The model puts readable lettering right into the frame rather than leaving room for manual layout.
Photorealistic portraits and product shots. Skin, reflections, metal, glass: FLUX is careful with textures and does not slide into the "plastic" AI look as often as more artistic models.
Complex multi-element scenes. When the prompt has five objects with a specified arrangement, FLUX handles more instructions in one pass: this is its architectural strength, prompt adherence.
Series with one character or product. Up to 10 references in FLUX.2 let you build a line of frames where the same product or hero looks the same.
How to run it in the Studio
- Sign up: email, Telegram or Google, about 30 seconds. At the start you get demo access to test the model without paying.
- Open the Studio, the "Images" section, and choose the FLUX version you need. The quick way is the direct link: Open in the Studio →
- Describe the scene with a prompt. FLUX likes specifics: the object, the material, the lighting, the angle, the background and, if needed, text in quotes.
- Choose the format and resolution (for a final marketplace image, Ultra and the maximum resolution make sense).
- Press "Generate": the cost in tokens is shown before the run.
A working prompt for a product shot with text:
A photorealistic shot: a white ceramic mug on a light concrete table,
soft diffused studio light from the left, a light shadow on the right, shallow
depth of field. On the mug, a neat black inscription "MORNING".
A neutral light gray background, minimalism, commercial product photography,
high detail of the ceramic texture.
What works here: the material and the light are specified (FLUX is precise with textures), a short text is given in quotes (a strong side of the model), and the background and composition are described for marketplace requirements.
What it costs in tokens
FLUX: from 2 tokens per image. Within the family the price depends on the version and the resolution.
The studio shows the exact charge before you run it, so check against it rather than the article: the catalog gets updated. Pack contents are on the pricing page.
FAQ
How is FLUX different from Midjourney? FLUX follows the prompt more precisely, renders text in the frame better and is more stable in photorealism and product photography. Midjourney is more expressive in artistic styles, concept art and a "painterly" look.
How is FLUX.2 different from FLUX Pro 1.1? FLUX.2 is the new generation (November 2025) that combines generation, editing and work with several references (up to 10) in one model. Pro 1.1 and 1.1 Ultra are the previous generation; Ultra is still strong in native 4 MP and detail.
What is FLUX dev and what is it for? FLUX dev is the open-weights version; in Pixyn, FLUX.1 dev works in image-to-image mode. To apply a LoRA for your style, take FLUX.2 LoRA. For most commercial tasks the Pro versions are enough.
Is FLUX suitable for marketplace cards? Yes. FLUX Pro 1.1 Ultra holds perspective, lighting and texture and can also put readable lettering into the frame, which covers the main marketplace requirements for a product photo.
How much does a FLUX generation cost? The cost in tokens is shown in the Studio before the run and is loaded from the database, so it is always current. Plans and subscription options are on /pricing.
Want to test FLUX on your own task? Start for free → Sign-up takes about 30 seconds via email, Telegram or Google.
More on the topic:
- AI for design 2026: where FLUX stands among a designer's tools
- Image generation in the Studio: try FLUX and other models right now
Sources
Try FLUX Pro v1.1 right now
The model opens in the studio with settings ready.
Read next
Models from this article
Try Pixyn for free
Free start and a 3-day Premium trial, no card required.