Flux 3 Max
fal and Black Forest Labs' post-trained FLUX 3 video model — top-tier image-to-video quality, clips up to 20 seconds with native audio, generated at near-real-time speed.
curl https://api.venice.ai/api/v1/video/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flux-3-max-text-to-video",
"prompt": "Aerial drone shot over a misty mountain valley at golden hour",
"duration": "5s",
"resolution": "720p",
"aspect_ratio": "16:9"
}'
# Use the returned queue_id with https://api.venice.ai/api/v1/video/retrieve.
# Call /video/complete after downloading if needed.Overview
What is Flux 3 Max
Flux 3 Max is a video generation model post-trained by fal Research with Black Forest Labs on the FLUX 3 multimodal foundation. It generates text-to-video and image-to-video clips up to 20 seconds in 1080p with native synchronized audio, and ranks among the highest-quality image-to-video models in independent human preference testing.
Running it privately on Venice
On Venice, Flux 3 Max runs under the private tier with zero retention — your prompts and uploaded images are never stored, profiled, or fed back into anyone's training pipeline. You pay per clip instead of subscribing, so a single 5-second 720p render costs $0.47 and there is no account history tying generations back to you. It is a closed model, so it can't be self-hosted — but on Venice you get frontier video generation without handing your creative work to a surveillance-first ad platform.
Specifications
Datasheet
- Maker
- fal Research (post-training) with Black Forest Labs (FLUX 3 foundation)
- Modality
- Text-to-video and image-to-video, with native audio
- Open weights
- No — proprietary
- License
- Proprietary — built on the FLUX 3 foundation model, post-trained by fal
- Clip lengths
- 5s – 20s
- Resolutions
- 720p, 1080p
- Mode
- text-to-video
- Audio
- Yes
- Input images
- One still image on the image-to-video variant; the text-to-video variant takes a written prompt only
- References
- One still image on the image-to-video variant; no multi-asset reference input
- Released
- October 2026
- Aspect ratios
- auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16
- Privacy on Venice
- Private — zero retention
- Available on Venice since
- Oct 2026
Assessment
Strengths and limitations
- Frontier image-to-video quality: a 1,139 Elo rating across 4,868 votes in Artificial Analysis's preliminary Video Arena, tied for #1 in blind human preference.
- Near-real-time speed: fal's benchmarks put a 10-second clip at roughly 11 seconds of generation, about 22× faster than Seedance 2.5 at the same reported quality.
- Native audio generated alongside the video, including dialogue in multiple languages with natural accents and lip-syncing.
- Clips up to 20 seconds: longer than every rival currently hosted on Venice.
- Aesthetics beyond the cinematic default: FLUX 3 was trained to render raw, natural, playful, nostalgic, or stylized footage, not just film-look output.
- Closed and proprietary: no open weights, so it cannot be self-hosted or fine-tuned (unlike Wan 2.7 on Venice).
- Output on Venice tops out at 1080p; Black Forest Labs' own API offers higher upscaled tiers that Venice does not expose.
- No video continuation or reference-video input on Venice — you generate from text or a single still image, not from an existing clip.
- The arena rating is preliminary and the speed figures come from fal's own benchmarks, so treat both as the maker's numbers rather than settled independent results.
Use cases
What it is good for
- 01Social ads and product loops where you need many takes fast and ship the one that lands.
- 02Talking-head and dialogue scenes with lip-synced multilingual speech generated natively.
- 03Vertical 9:16 shorts and 21:9 cinematic sequences up to 20 seconds long.
- 04Animating a product photo, illustration, or brand asset into motion via image-to-video.
- 05Storyboarding and previz, where near-real-time turnaround makes iteration cheaper than a shoot.
Prompting
Getting better results
Set duration explicitly — clips run 5 to 20 seconds, and shorter drafts cost less, so iterate at 5s before committing to 20s.
Pick the aspect ratio deliberately: 21:9 for cinematic widescreen, 9:16 for vertical social, 16:9 for standard video.
Draft at 720p ($0.47 for 5s), then re-render the winning prompt at 1080p ($0.80 for 5s) instead of paying full price for every experiment.
Put spoken lines in quotes and name the language or accent — the model generates lip-synced dialogue natively.
Describe camera movement and cuts explicitly ("slow dolly-in, then cut to overhead shot") to get multi-shot clips in one generation.
On the image-to-video variant, prompt for the motion you want ("steam rising, camera orbits left"), not a re-description of the still.
Samples
Sample outputs
Generated on Venice with our standard prompt suite — the same prompts we run through every model of this type, so you can judge it like-for-like.
Slow aerial drone shot gliding over a misty mountain valley at golden hour, sunlight piercing clouds onto a winding river, ancient pine forests on either side, ultra-smooth motion, professional color grading, atmospheric haze, 4K
Calm ocean waves rolling onto a black-sand beach at sunrise, golden light on wet sand, foam dissolving into the shore, a single silhouetted figure at the waterline, smooth continuous forward push, serene cinematic atmosphere
A slow tracking shot through a rain-soaked Tokyo street at night, neon reflecting in puddles, steam rising from a food cart, a person with a translucent umbrella, shallow depth of field, teal-and-orange grade, smooth steady camera
Alternatives
How it compares
| Model | Best for | Max duration | Max resolution | Native audio |
|---|---|---|---|---|
| Flux 3 Max | Top-tier quality at near-real-time speed | 20s | 1080p | Yes |
| Grok Imagine 1.5 | Fast audio-enabled clips | 15s | 1080p | Yes |
| Wan 2.7 Enhanced | Open-weights video with audio | 15s | 1080p | Yes |
| MiniMax H3 Max | Alternative fal post-trained model | 15s | 1080P | Yes |
| Vidu Q3 | Slightly longer clips than most rivals | 16s | 1080p | Yes |
Flux 3 Max is the right pick when you want frontier image-to-video quality without the wait — it is the only video model on Venice that ships 20-second clips with native audio, and it generates them at near-real-time speed.
API
Call it from your code
Venice exposes this model through the REST API. Queue a generation with the model id.
curl https://api.venice.ai/api/v1/video/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flux-3-max-text-to-video",
"prompt": "Aerial drone shot over a misty mountain valley at golden hour",
"duration": "5s",
"resolution": "720p",
"aspect_ratio": "16:9"
}'
# Use the returned queue_id with https://api.venice.ai/api/v1/video/retrieve.
# Call /video/complete after downloading if needed.Pricing
What it costs on Venice
Pay per clip on Venice — price scales with resolution and duration (5s–20s), from $0.47.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
FAQ
Frequently asked questions
Flux 3 Max is a video generation model post-trained by fal Research in collaboration with Black Forest Labs, built on the FLUX 3 multimodal foundation released in July 2026. It generates text-to-video and image-to-video clips up to 20 seconds long in 1080p, with native synchronized audio.
Venice bills per clip, with the price scaling by resolution and duration. A 720p 5-second clip starts at $0.47 and a 1080p 5-second clip is $0.80; longer clips cost proportionally more. There is no subscription required.
Flux 3 Max is neither free-software nor open-weights: it is a proprietary model built on Black Forest Labs' FLUX 3 foundation and post-trained by fal. You can try it on Venice with free welcome credits, but the weights themselves cannot be downloaded or self-hosted. Wan 2.7 is the closest open-weights alternative on Venice.
Yes. The flux-3-max-image-to-video variant animates a single still image into motion, with the same 5–20 second clip lengths, 720p/1080p resolutions, and native audio as the text-to-video variant. It is this image-to-video quality that ranked #1-tier in Artificial Analysis's preliminary arena.
Up to 20 seconds, selectable in one-second steps from 5s upward. That is longer than any other video model currently hosted on Venice, where most rivals cap at 15 seconds.
Yes. Audio is generated natively alongside the video, including dialogue in multiple languages with natural accents and lip-syncing — you do not need a separate text-to-speech pass or manual sound design.
Flux 3 Max wins on clip length (20s vs 15s) and currently holds the top-tier independent arena rating for image-to-video quality. Grok Imagine 1.5 is a capable 1080p alternative with native audio, but if you want maximum duration and benchmarked frontier quality, Flux 3 Max is the stronger choice.
Venice runs Flux 3 Max under its private tier with zero retention: prompts and uploaded images are not stored, profiled, or used for training, and generations are not tied to a personal history. Note that while the privacy tier is strict, the model does not run in a trusted execution environment and outputs are not end-to-end encrypted.
In fal's latency benchmarks, Flux 3 Max generates a 10-second video in about 11 seconds — close to real time, roughly 22× faster than Seedance 2.5 at the same reported quality and 11.5× faster than the original FLUX 3. These are the maker's own figures, so expect some variance in production.
Run Flux 3 Max privately
No prompt logging. No data used for training.