arkennemasis fal · Flux 3 Extend Video · $0.41-0.53/s
FLUX 3 extend video
- video
- video
- video_path
- seed
- info
Here's a surprise worth knowing before you wire this up. FLUX 3's continue-a-clip endpoint is the most expensive node in the Flux 3 family: $0.41 per second at 720p, $0.53 at 1080p. Text-to-video is $0.17/s at 720p. So extending a clip costs roughly two and a half times what generating one from scratch does - which is the opposite of the intuition that "more of the same" should be cheap.
Why? Continuation is a harder ask. The model has to honour the source's final frames, its motion, its lighting and its audio, from a starting point it didn't choose. You're paying for constraint, not for novelty.
ArkFal_blackforestlabs_flux_3_extend_video is FLUX 3 Video's continuation endpoint: a source clip plus a prompt describing what happens next, out comes footage that picks up where the original stopped.
What you set
video- required. The source clip. MP4, under 50MB, under 15 seconds. There's no client-side length check here, so an over-long clip is caught by fal after the upload rather than before - keep your sources short and trimmed.prompt- how it continues. The tooltip says it plainly: describe what happens next from the source's final frames.resolution-720p(default) or1080p. At these rates the difference is $0.12 per second, so nearly a dollar on an 8-second extension.duration-auto, or pinned. Pin it when you're assembling a timeline;autolets the model decide how much it needs.generate_audio- on by default, and at these prices you should at least be getting a soundtrack for it.aspect_ratio-automakes sense here, since the source has a shape already; force it and you're asking for a reframe.safety_tolerance- 0 to 4, default 2.
Optional and shared: max_cost_usd (default $20 - set it to $5 on this node, honestly), reuse_identical_run, extra_json, max_concurrent.
Outputs: video, video_path, seed, info.
Where extend actually earns its price
Not for "make my clip longer". For shot ends that need a landing: the character reaches the door and you need three more seconds for the dialogue to finish. Or the reverse - a clip whose final moment is perfect but whose setup drags, which is what the edit endpoint is for. Extension is a repair tool as much as a creative one, and thinking of it that way keeps the bill sane.
For iterating, the draft version of this node is $0.12/s and returns a draft cache you can finish with draft_enhance at $0.29/s. On a 10-second continuation that's $1.20 to find out whether the join works, versus $4.10 to find out the hard way.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis
Restart, or let ComfyUI Manager handle it - search arkennemasis. Then:
FAL_KEY=your-key-from-fal.ai/dashboard/keys
in the .env beside run_nvidia_gpu.bat (portable build) or in the ComfyUI folder. The node has no key field by design, so a key can never be saved into a workflow you share. Nothing is downloaded; FLUX 3 is fal-only at this point.
Gotchas
Your source clip is uploaded to fal's CDN. That's free and it happens before the paid call - but it does mean your footage leaves the machine, same as every other API node in this category.
Seconds come from the clip you connect or from the duration box, and the browser-side badge can't measure a clip, so it shows a rate while the real figure appears at run time. Since this is the priciest node in the family, that's the one place I'd actually set max_cost_usd low and let it argue with you.
Files are saved to output/fal/blackforestlabs_flux-3_extend-video/, every submit is logged in output/fal/_requests.jsonl, and a ComfyUI restart mid-render doesn't lose the job - fal Recover Result collects it by request id, free.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | The text prompt describing how the video continues from the source clip's final frames. | |
| aspect_ratio | COMBO | auto | Aspect ratio of the generated video. `auto` lets the model choose. |
| resolution | COMBO | 720p | Resolution of the generated video. |
| duration | COMBO | auto | Duration of the generated video in seconds. `auto` lets the model choose. |
| generate_audio | BOOLEAN | true | Whether to generate audio for the video. |
| safety_tolerance | INT | 20–4 | The safety tolerance level for the generated video. 0 is the strictest and 4 is the most permissive. |
| video | VIDEO | URL of the input video. MP4, under 50 MB and under 15 seconds. | |
| max_cost_usdopt | FLOAT | 20.00–100000 | Safety cap. If the estimated cost of this run is above this many US dollars the node stops BEFORE sending anything to fal. 0 = no cap. |
| reuse_identical_runopt | BOOLEAN | true | If these exact inputs (the settings AND the same pictures/clips/audio) were already paid for and the files are still in output/fal, return them instead of paying again - also after a ComfyUI restart. Turn off for a fresh take with the same settings. |
| extra_jsonopt | STRING | Advanced: a JSON object merged into the request last, for any fal field this node has no box for. Example: {"seed": 7}. Leave empty normally. | |
| max_concurrentopt | INT | 11–32 | How many paid fal calls may run at the same time in one run, across every fal node. 1 (default) = one after another. This node starts its paid part only while fewer than this many fal calls are running. Once any fal node fails, the ones still waiting are not started, so nothing more is billed. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| video_path | STRING | — |
| seed | INT | — |
| info | STRING | JSON: request id, cost estimate, saved files and fal's full answer. |