Explore FLUX 3 Dev, the open-weights multimodal foundation model by Black Forest Labs. FLUX 3 Dev combines image synthesis, video creation with synced audio, and robotic action prediction in one unified Self-Flow architecture.
Result will appear here
Unified multimodal generation
Released in July 2026, FLUX 3 Dev represents a major leap from Black Forest Labs. Where FLUX.2 Dev focused solely on still images, FLUX 3 Dev merges image creation, video production with synchronized sound, and physical action prediction into one cohesive system. Powered by the Self-Flow framework, FLUX 3 Dev learns a shared world model across visual, auditory, and temporal modalities. As the open-weights release, FLUX 3 Dev gives developers unrestricted access to fine-tune, adapt, and deploy this multimodal engine for creative and industrial applications.
Up to 1024px on the longest edge
Text 2 credits · Edit 2 credits
Up to 1280px on the longest edge
Text 3 credits · Edit 3 credits
Up to 1536px on the longest edge
Text 4 credits · Edit 4 credits
Real model outputs
Both examples were generated through the exact WaveSpeed model used by the tool above and are served from this site.
Text to Image
Created by FLUX 3 Dev, this image showcases the model's ability to render convincing materials, reflections, and studio lighting within its multimodal framework.
Prompt
Premium studio product photograph of a sculptural amber perfume bottle on dark travertine, controlled rim lighting, realistic glass refraction, restrained luxury campaign styling, crisp commercial detail
Image Edit
FLUX 3 Dev retains the original product while completely replacing the studio backdrop, proving its precision editing within the unified multimodal system.
Prompt
Keep the perfume bottle unchanged and replace the dark studio with a warm ivory stone set, soft morning sunlight, and delicate botanical shadows; premium commercial photography
With FLUX 3 Dev, you can produce still images, generate 20-second video clips with matching audio, and edit existing visuals — all from a single model. FLUX 3 Dev eliminates the need to juggle separate tools for different media types.
FLUX 3 Dev creates motion content from text, animates photos, interpolates between keyframes, and renders multilingual speech with synchronized lip movement. Styles range from handheld camcorder looks to polished cinematic sequences, all generated natively by FLUX 3 Dev.
FLUX 3 Dev exposes the full multimodal backbone as open weights. Customize FLUX 3 Dev with your own data, build LoRA adapters for brand-specific styles, create video production pipelines, or leverage FLUX 3 Dev action prediction for robotics and automation projects.
How it works
Pick FLUX 3 Dev Text to Image for new stills, switch to video mode for motion content with built-in audio, or use Image Edit to rework existing visuals with reference images.
For still images, specify materials, lighting, and camera angles. For video, outline the scene, motion, dialogue, and sound cues. FLUX 3 Dev responds well to detailed, structured instructions across all modalities.
Start with Standard quality for rapid prototyping, then scale up to High or Ultra for final deliverables. FLUX 3 Dev handles multiple aspect ratios and output formats for both image and video pipelines.
Compare models
FLUX 3 Dev is the open-weights release of FLUX 3, the multimodal foundation model from Black Forest Labs. It brings together image creation, video generation with synchronized audio, and action prediction for robotics — all powered by the Self-Flow architecture. FLUX 3 Dev gives developers full access to this unified backbone.
Absolutely. FLUX 3 Dev generates video clips up to 20 seconds long with native audio, including speech that matches lip movements and sound effects tied to on-screen actions. It supports text-to-video, image animation, keyframe interpolation, and video-to-video transformation.
FLUX.2 Dev handled only still images. FLUX 3 Dev expands into video, audio, and robotics through the Self-Flow architecture. By training jointly on visual, auditory, and temporal data, FLUX 3 Dev develops a richer understanding of the physical world than its predecessor.
FLUX 3 Dev generates audio natively alongside video — speech tracks sync with lip movements, and sound effects align with physical events on screen. Audio represents less than 0.5% of the computational tokens in a FLUX 3 Dev video, making it an efficient addition to the visual pipeline.
Self-Flow is the core framework behind FLUX 3 Dev — a modality-agnostic, self-supervised approach to flow matching. It blends generation and representation learning across images, video, audio, and actions. Video accounts for over 95% of FLUX 3 Dev compute, with other modalities layered on efficiently.
Yes. FLUX 3 Dev has built-in action prediction. The FLUX-mimic system, developed on the FLUX 3 Dev backbone with mimic robotics, runs in Audi's production lines for tasks like part kitting and cable handling. It delivers up to 10x better sample efficiency than traditional vision-language-action approaches.
Yes. FLUX 3 Dev releases the multimodal backbone as open weights for both content creation and action prediction. You can fine-tune FLUX 3 Dev on custom datasets, train adapters, and deploy it for specialized video, image, or robotics workflows.
Go with FLUX 3 Dev when you need open access for customization, fine-tuning, or robotics integration. Choose FLUX 3 Pro when you want premium commercial output via API without managing your own infrastructure. FLUX 3 Dev prioritizes flexibility; FLUX 3 Pro prioritizes polish.
Start from a prompt or upload references, choose the output quality, and generate directly on this page.
