Industry News

FLUX 3 Drops: Black Forest Labs Merges Image, Video and Audio Into One Model

Black Forest Labs releases FLUX 3, the first multimodal model generating images, 20-second video with native audio, and robot actions from one architecture.

4 min readBy The Arena Editor
#FLUX 3#Black Forest Labs#multimodal AI#AI video#AI image generation
Abstract visualization of FLUX 3 unified AI model for image video and audio generation

Some links are partner links: if you subscribe through them, we may earn a commission, at no extra cost to you. The crowd verdicts stay independent.

FLUX 3 marks Black Forest Labs' first unified multimodal model. Released on July 23, 2026, it generates images, 20-second video with native audio, and robot action predictions from a single set of weights. This is not three models under one brand: it is one architecture trained jointly on visual, auditory, and physical data.

Key takeaway

FLUX 3 is the first model to ship image, video (up to 20 seconds) and synchronized audio generation in one architecture. Early access is now open for video; image access follows in August 2026.

What FLUX 3 actually delivers

Black Forest Labs built FLUX 3 on Self-Flow, their March 2026 method that aligns generation and understanding within the same model. The claim: training jointly on images, video, and audio improves each modality rather than diluting them.

The video component generates up to 20 seconds at 720p with native audio. Inputs include text, image, video, or keyframe-based prompts. Multilingual dialogue, multi-shot sequencing, and keyframe-to-video transitions ship at launch. Image generation, editing, and readable text rendering come next.

The robotics angle is real. FLUX-mimic, a video-action model developed with mimic robotics, is already running in Audi production facilities for dexterous manipulation tasks that conventional robots cannot handle cost-effectively.

How FLUX 3 compares to the current FLUX.2 lineup

FLUX.2 remains the current public product. In the GLAD-AI-TOR arena, FLUX.2 holds a 71% crowd recommendation rate based on community votes, with API pricing from $0.014 per image (klein 4B) to $0.03 (pro) and $0.07 (max).

ModelGenerationCrowd scorePriceStatus
FLUX.2 [pro]Image only71% recommend$0.03/imagePublic API
FLUX.2 [dev]Image only71% recommendFree (self-host)Open weights
FLUX 3 VideoVideo + audioTBDTBDEarly access
FLUX 3 ImageImageTBDTBDComing August 2026
FLUX 3 DevAll modalitiesTBDFree (self-host)Later 2026

FLUX 3 Dev, the open-weight release, is scheduled for later 2026. If it ships, it would be the first publicly available open-weight model that jointly generates video, audio, and images.

What this means for creative workflows

Unified multimodal generation collapses a pipeline that currently requires separate tools for each format. Consider a typical video ad workflow: image generation in FLUX or Midjourney, video conversion in Runway or Luma, audio scoring or voiceover added separately.

FLUX 3 theoretically handles all three in one pass. That saves cost, reduces handoff friction, and keeps style consistent across formats. The practical question is quality: FLUX.2 [pro] at $0.03 per image already undercuts Midjourney ($10/month minimum, 57% crowd score) on price while leading open-weight benchmarks. If FLUX 3 maintains that edge across video and audio, the cost math changes.

FLUX

The open-weight champion: 32B FLUX.2 wins 66.6% of head-to-heads vs open rivals, API from $0.014 per image

$3/mo71%(3)

Partner link. The crowd verdicts stay independent.

Early access: who can use FLUX 3 now

FLUX 3 Video is gated. Access requires an application at bfl.ai/models/flux-3, with API and private weights available to approved partners. Image generation rolls out "in the following weeks" according to the launch announcement. Open-weight FLUX 3 Dev follows sometime in late 2026.

For production use today, FLUX.2 remains the active option. The FLUX tool page lists current API pricing and capabilities.

How it stacks against the competition

The direct comparison depends on which modality matters most.

For images: Midjourney V8.1 still leads on aesthetics according to most 2026 reviews, but requires a $10+/month subscription with no API. Recraft V4 tops the Hugging Face arena for human preference and generates native SVGs. FLUX.2 [pro] at $0.03 per image remains the cheapest frontier-quality API.

For video: Runway Gen-4.5 bundles Veo 3.1 and Kling 3.0 Pro in one $12/month subscription, with native audio since mid-2026. Luma Ray3.14 offers 1080p in 2 seconds with 16-bit HDR but no audio. FLUX 3 Video, once public, enters at 720p with native audio.

ToolFocusCrowd scoreEntry price
FLUX.2 [pro]Image API71%$0.03/image
MidjourneyAesthetic images57%$10/mo
Recraft V4Design + SVG71%$12/mo
Runway Gen-4.5Video suite57%$12/mo
Luma Ray3.14Cinematic video57%$10/mo

FLUX

The open-weight champion: 32B FLUX.2 wins 66.6% of head-to-heads vs open rivals, API from $0.014 per image

$3/mo71%(3)

Partner link. The crowd verdicts stay independent.

The verdict

FLUX 3 is a statement of direction. Black Forest Labs is moving from best-in-class image generation to unified multimodal output. The architecture, the Audi robotics deployment, and the planned open-weight release signal serious intent.

For now, FLUX.2 remains the actionable product. The arena data shows a 71% recommendation rate, API pricing from $0.014 per image, and open weights that run on consumer GPUs. That positions FLUX as the developer's pick in image generation.

When FLUX 3 Image and Dev ship, the comparison gets more interesting. A single model that handles image, video, and audio with FLUX.2-level quality and open weights would reshape the category.

  • Who should apply for early access: Teams already using FLUX.2 who need video output without switching tools.
  • Who should wait: Anyone not currently blocked by the lack of video. FLUX.2 [pro] remains the best value in image APIs.
  • What to watch: FLUX 3 Dev release timing and licensing terms.

Explore the full comparison: FLUX vs Midjourney | Best AI image generators

Keep exploring

Every claim above is backed by the arena's live data: crowd votes, verified pricing, honest pros & cons.

More from the arena journal