Introduction

FLUX 3 Max just moved into your node graph. On October 21 (18:58 UTC, 02:58 MYT on October 22), ComfyUI announced that Black Forest Labs' FLUX 3 Max video model is live as Partner Nodes, with ready-made text-to-video and image-to-video workflows on Comfy Cloud. It's the same closed, hosted model fal Research and BFL launched on October 19, so there are no weights to download: you sign in, spend Comfy credits, and get a clip with synchronized audio back in your graph.

ComfyUI's own pitch is the headline number from launch week: a 10-second, 720/768p video in about 12 seconds. If you missed the model itself, our FLUX 3 Max launch writeup has the benchmarks and fal pricing.

What shipped

  • Model: FLUX 3 Max, the post-trained, latency-tuned branch of FLUX 3 Video
  • Vendors: Black Forest Labs with fal Research, wired into ComfyUI by Comfy-Org
  • Modality: video with native audio, from text or from 1 to 10 keyframe images
  • Open vs closed: closed and hosted; the nodes call the model through Comfy's API proxy
  • Where to run it today: Comfy Cloud templates, or a local ComfyUI updated to include PR #17202, merged October 20
  • Node names: FLUX 3 Max Text to Video and FLUX 3 Max Image to Video, under partner/video/BFL

What the nodes can do

Both nodes share the same controls:

  • Duration: any whole number from 5 to 20 seconds (default 5)
  • Resolution: 720p (default) or 1080p
  • Aspect ratio: auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4 or 9:16. Auto means 16:9 for text prompts and follows your first image otherwise
  • Reasoning level: low or high (default high). High spends longer thinking about and expanding your prompt, at the same price
  • Generate audio: on by default; switch it off for a silent clip
  • Prompt length: up to 50,000 characters, so long shot lists fit

The image-to-video node is the fun one. Feed it one image and it opens the clip. Feed it two and they become the first and last frame. Feed it three to ten and they're either spread evenly across the clip or pinned to the seconds you type, like 0, 2.5, 5. That's the keyframes mode from fal's API, folded into a single node.

What it costs in Comfy

The node's price badge bills per second of finished video, and there's a launch discount until November 7 (00:00 UTC, 08:00 MYT):

  • 720p: about $0.061 per second during the promo, so a 5-second clip is roughly $0.30 and a 20-second one about $1.22. The regular rate is about $0.243 per second
  • 1080p: about $0.104 per second during the promo, so 10 seconds comes to roughly $1.04. The regular rate is about $0.415 per second

That's higher than fal's direct list price of $0.17 per second at 720p, which is the trade for staying inside your graph with one login and one credit meter.

How to run it

The quickest path is Comfy Cloud. ComfyUI posted direct links to its FLUX 3 Max text-to-video template and image-to-video template.

Running locally? Comfy's docs changelog lists the FLUX 3 Max nodes under v0.42.2, while the latest tagged GitHub release is still v0.42.1, so update through ComfyUI Desktop or the Manager, or pull the latest code if you run from git:

cd ComfyUI
git pull
pip install -r requirements.txt

Then sign in with your Comfy account so Partner Nodes can bill credits, search the Node Library for FLUX 3 Max, and wire a prompt in. The text-to-video tooltip suggests describing ambient sound, music and speech separately for layered audio, so here's a starter that does exactly that:

A paper-lantern night market on a narrow rainy street, slow handheld push-in past steaming noodle stalls toward a street violinist under a lilac neon sign, warm gold light on wet cobblestones, shallow depth of field, cinematic. Ambient sound: soft rain, sizzling woks, distant chatter. Music: a gentle, wistful solo violin melody. Speech: a vendor calls out "fresh dumplings" once in the background.

Start at 720p with reasoning on high to lock the shot, then bump to 1080p for the keeper.

Also new in ComfyUI this week

The FLUX 3 Max drop rode in with a busy run of master commits:

  • FLUX 3 Video 2K and 4K: the standard FLUX 3 Video nodes picked up 1440p and 2160p output (PR #17200)
  • Native Fibo 2.0: local support for Bria's open-weights Fibo 2.0 image model (PR #17240), alongside a hosted FIBO 2.0 Image Generate Partner Node (PR #17229)
  • Krea 3 Ultra thinking: the Krea 3 Image Partner Node gained the Ultra thinking mode (PR #17260)
  • VEED nodes: Fabric 1.0, Lipsync 2.0, Background Removal and Clean Audio (PR #17242)
  • Mirelo sound effects: Video to SFX and Text to SFX nodes (PR #17265)

Original Source

ComfyUI's announcement:

The node code lives in comfy_api_nodes/nodes_bfl.py, and the Comfy docs changelog tracks the v0.42.2 Partner Node updates.

Conclusion

FLUX 3 Max was already one of the fastest frontier video models you could call. Now it's a node you can chain with upscalers, sound tools and everything else in your graph, and at roughly six cents a second until November 7, this is a lovely week to build a keyframe-driven short and see how far ten pinned images can take a story 🎬

—Aurelia ♡