Skip to content
trending

News

FLUX 3 Image: Native 4K, Bounding Boxes & Multi-Edit

Black Forest Labs released FLUX 3 Image on October 1, 2026: native 4K up to 5456×3072px, bounding box composition, and multi-area editing without touching the rest.

By 3 min read
FLUX 3 Image: Native 4K, Bounding Boxes & Multi-Edit

👁 124 reads

Black Forest Labs released FLUX 3 Image on October 1, 2026 - the image generation component of the FLUX 3 family announced in July. Three features stand out: native 4K up to 5456 × 3072 px, scene composition using bounding boxes (each object gets coordinates), and editing multiple areas at once without touching the rest of the image. Through the fal.ai API it costs $0.024 per megapixel until October 8 (1K image ~$0.024, 4K ~$0.40), then double that. We tested it the same day.

The FLUX 3 family has now seen three releases. In July, Black Forest Labs showed a single model trained simultaneously on images, video, and audio. In August they released FLUX 3 Video, and on September 23 FLUX 3 Action for robot control. Images came last, after ten weeks of waiting.

FLUX Hub - all Black Forest Labs models, our tests and prompts: FLUX 3 Image and FLUX.2.

What’s New in FLUX 3 Image

  • Native 4K. Five resolution levels: 512sq, 768sq, 1k, 2k, and 4k. In 4k mode I got a 5456 × 3072 px image (16.8 Mpx) without separate upscaling.
  • Bounding box composition. The canvas is a grid from 0 to 1000 regardless of aspect ratio. Each object gets an identifier, description, and bounding box in [y_min, x_min, y_max, x_max] format.
  • Multiple edits at once. According to the maker, pixels outside the edited boxes remain bit-for-bit identical.
  • Up to 10 reference images. In the prompt you reference them as ref_image_0 through ref_image_9. Through fal.ai each input image can be up to 4 Mpx.
  • Better text rendering. The maker recommends the model for posters, ads, and infographics.
  • Weights. Commercial license for weights is available immediately, and an open version is set to appear “within weeks.”

How a Bounding Box Prompt Looks

Instead of describing “lamp in the upper left corner,” you provide coordinates. Here’s my first test prompt, entered in the prompt field on fal.ai:

{"caption": "Editorial photo of a minimalist desk in a bright apartment, morning light, 35mm.",
 "elements": [
  {"id": "lamp_1", "bbox": [80, 60, 520, 300], "desc": "a brass desk lamp"},
  {"id": "plant_1", "bbox": [300, 700, 900, 960], "desc": "a tall monstera plant in a white pot"},
  {"id": "laptop_1", "bbox": [450, 320, 850, 680], "desc": "an open silver laptop showing a spreadsheet"},
  {"id": "mug_1", "bbox": [700, 80, 900, 260], "desc": "a red coffee mug with text "[BRAND]""}]}
Desk with lamp, laptop, monstera plant, and [BRAND] mug arranged according to bounding boxes in FLUX 3 Image
Lamp, monstera, laptop, and mug landed exactly where I placed them. Text on the mug rendered without error.

Pricing and Access

Where Price Notes
fal.ai (text-to-image and editing) $0.024 per megapixel until Oct 8, then $0.048 1K ~$0.024, 4K ~$0.40
Black Forest Labs API 50% off until October 8 playground at dashboard.bfl.ai
Weights for commercial use license from maker open version announced

At 1K resolution, FLUX 3 Image with the promotional pricing is several times cheaper than comparable models at highest quality. At 4K the difference disappears.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

First Impressions from Testing

We ran three tests, each comparing FLUX 3 with top models: 4K under the magnifying glass, bounding box composition, and five edits in one prompt. Results and all images are in separate articles, with a summary in the FLUX hub.

Kinetyra

// read next

More on this from QUANTSLANT

Piotr Olszewski

Editor and publisher · Quantslant

I edit and publish Quantslant. AI tools help with research, drafts and translation; choosing topics, running the tests, checking facts and the decision to publish are mine, and so is the responsibility for what appears here.

read nextMagic Patterns in Practice: Prototype Instead of Docsai toolsMagic Patterns in Practice: Prototype Instead of DocsMagic Patterns works best in product teams where an idea passes through product managers, designers, and developers. Here's where it saves the most time.

Most read

  1. 1Best AI Video Generator Apps for Your Phone (2026)
  2. 2Claude Opus 5.5 Motion Graphics: 60s Showreel from One Prompt
  3. 3Wan 2.2 on RTX 4090: timings, VRAM and four test clips
// site map

This page in other languages1

Quantslant in numbers 1,423posts in archive22news from the last 7 days18video models in the observatory
🎬 AI Video Benchmarks: Higgsfield DoP vs OpenAI Sora • Higgsfield vs Runway Gen-3 • AI Video Credit Calculator
× ‹ enlarge ›