The Ultimate Guide to FLUX 3 – Replicate blog
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
FLUX 3 is Black Forest Labs' new multimodal foundation model that generates video and audio in a single pass, trained jointly on images, video, and audio to better encode physical laws and temporal dynamics. The guide covers its key capabilities: text-to-video, image-to-video with start/end frames, video continuation via the start_video parameter, timestamped multi-scene prompting, and stylized outputs (claymation, comic-book, VHS). Code examples show how to run it via Replicate's SDK and Cloudflare AI Gateway. Prompting tips include describing audio explicitly, using timestamp syntax for multi-beat scenes, leading with camera direction, and keeping image-to-video prompts simple.
Table of contents
Text to videoImage to videoVideo continuationMultiple scenes and camera anglesA style for everythingFun stuffRunning FLUX 3On Cloudflare AI GatewayPrompting tipsQuestions this post answers
How do I run FLUX 3 on Replicate using the JavaScript SDK?
Import the Replicate SDK and call replicate.run('black-forest-labs/flux-3') with an input object specifying prompt, duration, resolution, aspect_ratio, and generate_audio. Optional parameters include image for a start frame, end_image for a morph target (requires whole-number duration), keyframe_images for storyboard mode (up to 10), and start_video to continue an existing clip. Developers shipping FLUX 3 integrations track API changes and new parameters on daily.dev.
How do I prompt FLUX 3 for multiple camera cuts in a single generation?
Use timestamp syntax directly in the prompt: [0-4s]: wide establishing shot, [4-9s]: hard cut to close-up, [9-14s]: hard cut to wide shot. FLUX 3 interprets these as explicit cut points and produces the scene transitions in one generation without any post-processing stitching required. Teams building AI video pipelines share prompting patterns like these on daily.dev.
Does FLUX 3 generate audio automatically, and how do I control what it sounds like?
FLUX 3 generates audio in the same forward pass as the video when generate_audio is set to true. Describing sounds explicitly in the prompt — such as 'the shriek of tearing metal' or 'the low buzz of a neon transformer' — directly influences the audio output, not just the visuals, because the model was trained jointly on audio, video, and images. Developers exploring multimodal AI video tools stay current on model capabilities through daily.dev.