Flux 3 Extend
Black Forest Labs' video-continuation model — feed it an existing clip and it generates 5–20 new seconds from the final frame, with synchronized native audio.
# video_url: the clip to restyle (MP4, MOV or WebM)
# duration: the source clip's length in seconds
curl https://api.venice.ai/api/v1/video/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flux-3-extend-video",
"prompt": "Restyle this footage as a rainy autumn evening, keeping the composition and camera movement",
"video_url": "https://example.com/source.mp4",
"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 Extend
Flux 3 Extend is Black Forest Labs' video-continuation model. It takes an existing clip and generates new footage continuing from its final frames, with synchronized native audio. Each generation adds 5–20 seconds at up to 2160p on Venice, letting you lengthen a keeper take instead of re-rolling it from scratch.
Using it anonymously on Venice
On Venice, Flux 3 Extend runs under the anonymized privacy tier: your source clips and prompts are not stored, profiled, or used for training. You upload footage and extend it permissionlessly — pay per clip in credits, no subscription, no personal generation history attached to your work. The surveillance-ad-supported video tools keep every upload you give them; Venice keeps nothing.
Specifications
Datasheet
- Maker
- Black Forest Labs
- Modality
- Video continuation (video-to-video) with native audio
- Open weights
- No — proprietary
- License
- Proprietary
- Clip lengths
- 5s – 20s in 5s steps
- Resolutions
- 720p, 1080p, 1440p, 2160p
- Mode
- video
- Audio
- Yes
- References
- One source video clip (MP4, under 50 MB) — no image, audio, or document references
- Released
- August 4, 2026
- Source clip requirements
- MP4, under 50 MB, under 15 seconds
- Aspect ratios
- auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, 9:21
- Privacy on Venice
- Anonymized — prompts not stored
- Available on Venice since
- Aug 2026
Assessment
Strengths and limitations
- Purpose-built continuation: it extends a clip you already like, preserving its motion, timing, and framing instead of gambling on a fresh generation.
- Native synchronized audio carries through the extension — picture, speech, effects, and ambience come from one multimodal system, with no separate sound pass to stitch.
- Up to 20 seconds per generation: the longest single output among Venice's hosted video models — and up to 2160p (4K) resolution on Venice.
- Text-prompt steering: you direct how the scene continues from the final frame rather than accepting a blind auto-extension.
- Eight aspect ratios including 9:16 vertical and 21:9 cinema, so one continuation can serve trailers, social cutdowns, and widescreen edits.
- Closed and proprietary: no open weights, so it cannot be self-hosted or fine-tuned (Wan 2.7 and Wan 2.7 Enhanced are the open-weight options on Venice).
- Continuation only: it cannot invent a clip from a text prompt or image — that is FLUX.3 Text to Video and Image to Video's job.
- The source clip must be MP4, under 50 MB and under 15 seconds, so long masters need to be trimmed before extension.
- No image, audio, or document references: the only input is the video itself plus a text prompt.
Use cases
What it is good for
- 01Lengthening a keeper take — a trailer shot, a product spin, a performance take — without re-generating and losing what worked.
- 02Building multi-beat sequences by chaining extensions off the previous clip's final frame.
- 03Extending talking-head or dialogue clips with speech and ambience continuing natively.
- 04Producing vertical 9:16 social extensions of widescreen masters for UA campaigns.
- 05Iterating on cinematic game trailers and ad variants while preserving temporal consistency with the approved source.
Prompting
Getting better results
Keep the source clip under 15 seconds and under 50 MB in MP4 — longer or heavier files won't be accepted as input.
Describe what happens next from the final frame ("she turns and walks toward the door"), not a recap of the footage you're extending.
Pick the hop length deliberately: 5s for a single beat, 20s for a full scene — price scales with duration and resolution.
Test the continuation at 720p first, then re-run the winning take at 1080p or 2160p for delivery.
Set the aspect ratio explicitly — 9:16 for social cutdowns, 21:9 for cinematic trailers — instead of leaving it on auto.
Write dialogue and sound cues into the prompt: native audio follows the picture, so say what should be heard as well as seen.
Alternatives
How it compares
| Model | Best for | Max duration | Max resolution | Native audio |
|---|---|---|---|---|
| Flux 3 Extend | Extending existing clips | 20s | 2160p | Yes |
| Wan 2.7 Enhanced | Open-weight generation with audio | 15s | 1080p | Yes |
| Grok Imagine 1.5 | Fast text-to-video with sound | 15s | 1080p | Yes |
| Vidu Q3 | Slightly longer single takes | 16s | 1080p | Yes |
| Kling O3 Pro | Cinematic motion quality | 15s | — | Yes |
Flux 3 Extend is the right pick when you already have a clip worth keeping and need to lengthen it with sound intact — every other Venice video model starts from a prompt, while this one starts from your footage.
API
Call it from your code
Venice exposes this model through the REST API. Queue a generation with the model id and the inputs it needs.
# video_url: the clip to restyle (MP4, MOV or WebM)
# duration: the source clip's length in seconds
curl https://api.venice.ai/api/v1/video/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "flux-3-extend-video",
"prompt": "Restyle this footage as a rainy autumn evening, keeping the composition and camera movement",
"video_url": "https://example.com/source.mp4",
"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 $2.26.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
FAQ
Frequently asked questions
Flux 3 Extend is Black Forest Labs' video-continuation model, released August 4, 2026. It takes an existing MP4 clip and generates new footage continuing from its final frames, with synchronized native audio. Each generation adds 5–20 seconds, and on Venice it can output at up to 2160p.
On Venice you pay per clip, with the price scaling by resolution and duration (5s–20s): $2.26 for a 720p 5-second clip, $2.92 at 1080p, and $3.33 at 1440p. There is no subscription requirement — you buy credits and spend them per generation.
Neither. Flux 3 Extend is proprietary to Black Forest Labs with no open weights, so it cannot be self-hosted or fine-tuned. On Venice it is billed per clip starting at $2.26. If you need open weights, Wan 2.7 and Wan 2.7 Enhanced are the open alternatives hosted on Venice.
A single source clip in MP4 format, under 50 MB and under 15 seconds. The model continues from that clip's final frames; it does not accept image, audio, or document references. Longer masters need to be trimmed to a 15-second segment before extension.
Yes. Like the rest of the FLUX.3 video family, it produces synchronized native audio — speech, effects, and ambience generated alongside the picture rather than bolted on afterward. You can steer the soundtrack by describing dialogue and sound cues in the prompt.
Each generation adds 5–20 seconds, selectable in 5-second steps, and the source clip it extends must itself be under 15 seconds. By chaining extensions — extending the output of a previous run — you can build longer sequences beat by beat.
They solve different problems. Flux 3 Extend continues footage you already have, preserving its motion and framing, and reaches 2160p with 20-second generations. Wan 2.7 Enhanced generates new clips from prompts with open weights you can self-host, at up to 15s and 1080p. Choose Flux 3 Extend to lengthen a keeper; choose Wan 2.7 Enhanced for open, prompt-driven generation.
Venice runs Flux 3 Extend under its anonymized privacy tier: your prompts and source clips are not stored, profiled, or used for training. You can use it in the Venice app or via the Venice API without building a personal generation history — no account-level trail tying your footage to your identity.
Black Forest Labs shipped the FLUX.3 video family, including Extend, on August 4, 2026, and Venice listed it the same day. It has been live and available on Venice since August 2026.
Use Flux 3 Extend anonymously
Venice does not store your prompts. Chat history stays in your browser.