FAL Vidu Q3 Reference To Video Mix (Soze)
Throw several references at it and let one coherent clip come back
- image_1
- image_2
- image_3
- image_4
- image_batch
- video
- video_url
- status
Where Vidu Q1 is a focused single-purpose node, Q3 "Mix" is the opposite: it exists to take a small pile of references and merge them into one coherent scene. The "mix" in the name is the point - you hand it several images and it figures out how they belong together, which makes it the node you reach for when you have a character here, a setting there, an object somewhere else, and you want a single video that respects all of them.
The inputs
Four image slots is the cap here - image_1 through image_4 (or image_batch, capped at 4) - cited in the prompt as @Image1, @Image2, etc. That's fewer references than the other FAL nodes in the pack, and it's deliberate: Mix is about combining a handful of distinct elements, not about piling on variations of one subject. Prompt cap is 2000 characters.
The controls:
duration- 1 to 16 seconds as an integer. The longest range in the FAL family here.aspect_ratio- five options:16:9,9:16,4:3,3:4,1:1.resolution- 360p, 540p, 720p, or 1080p. Same cheap-iteration ladder as the PixVerse node.audio- default on, which is unusual. Flip it off if you'd rather add sound in post.seed- 0 = random; set it to keep a take reproducible.
How it works
Same FAL machinery: reference images upload to FAL's CDN, the node POSTs to fal-ai/vidu/q3/reference-to-video/mix, waits synchronously, then downloads the mp4 into your output folder as a VIDEO object. Outputs are video, video_url, and status.
The mix behavior means the prompt matters in a specific way: say how the references relate. "Character in @Image1 sits at the desk from @Image2 while the skybox from @Image3 rolls past the window" - that's the kind of prompt Mix rewards. Just listing the references and letting the model guess the relationship is how you get a Frankenstein clip.
Install and key
From the Soze pack ("Quality of Life Nodes for ComfyUI"). ComfyUI Manager → search "Soze", or:
cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt
Restart, then set FAL_KEY as an environment variable or in ComfyUI_Soze/config.ini:
[API]
FAL_KEY=your_key_here
Key at fal.ai/dashboard/keys; keep config.ini out of git.
Common issues
- The references don't blend - they fight. The model needs the relationship spelled out in the prompt; describe how the elements interact rather than just listing them.
- Audio comes back unwanted.
audiodefaults to on. If you plan to score it yourself, turn it off before rendering. - Skips with no output. Needs at least one reference image connected.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Max 2000 characters. | |
| duration | INT | 51–16 | Seconds (1-16). |
| aspect_ratio | COMBO | 16:9 | 5 options: 16:9, 9:16, 4:3, 3:4, 1:1 |
| resolution | COMBO | 720p | 4 options: 360p, 540p, 720p, 1080p |
| audio | BOOLEAN | true | Generate audio with the video. |
| seedopt | INT | 00–18446744073709550000 | 0 = random. |
| image_1opt | IMAGE | Reference image 1 (cite as @Image1 in the prompt). | |
| image_2opt | IMAGE | Reference image 2 (cite as @Image2 in the prompt). | |
| image_3opt | IMAGE | Reference image 3 (cite as @Image3 in the prompt). | |
| image_4opt | IMAGE | Reference image 4 (cite as @Image4 in the prompt). | |
| image_batchopt | IMAGE | Optional IMAGE batch (capped at 4). Ignored if any image_N slot is connected. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| video_url | STRING | — |
| status | STRING | — |