Batch generation in Workflow
Run one line with many inputs using Batch, Dice and Stack, and see how each one affects the price.
What this video covers
In Workflow, there are three ways to run one line with many inputs: Batch, Dice and Stack.
Drop a few images into a Batch node, and each one runs the rest of the line once, in turn. A Batch holds up to twenty items.
Wire the Batch into the Generator's image input. Here, Nano Banana 2 Edit reimagines each photo.
The price is multiplied by the number of items: three images, three generations. The tilde in front shows it's an estimate.
A Dice picks one of its connected lanes at random and passes it on. There's a Dice under Media for images and video, and one under Operator for prompts.
We wire three style prompts into the Dice. This run, it picked the rainy neon night, so all three photos came out in that style. The next run rolls the Dice again.
If Generators feed a Dice, be careful: the lanes it doesn't pick still run, and you're charged for them.
When you press Run, the items go through one at a time: the current one glows, and finished ones fade. If an item fails, it's marked red and the line moves on.
The Media node at the end collects all three results, and they're saved to your library as well.
To give one generation several references, there's the Stack. Pick a multi-reference model on an empty Generator, and LUVI offers to wire in a Prompt, a Stack sized to the model, and a Media node for you.
A Stack bundles its inputs into one cable, and the jack number sets the reference order. It's a single generation, so the price isn't multiplied.
In short: a Batch reruns the line for every item, a Dice picks one lane each time, and a Stack feeds them all into one generation.
In the next part, we'll look at operators.
More from Going further
All tutorials
1:43Tutorial 14Editing your images
Edit an image with an editing model: put it in the reference slot, say what should change and what should stay, keep the aspect ratio and combine several images.
1:31Tutorial 15aYour first workflow
Build your first chain in Workflow: add nodes, pick models, wire an image into a video, check the price and run it.
1:49Tutorial 15cWorkflow operators
Shape the text before it reaches the Generator with the Merge, Caption, Style and Pose operators.
1:49Tutorial 15dManaging workflow runs
The Preflight Check, running part of a graph, groups, cancelling, saving workflows and the panels on the right of the canvas.
1:24Tutorial 16aAI voice-over
Turn text into speech in LUVI: pick a voice-over model and a voice, set the stability and language, and generate.
1:26Tutorial 16bAI music
Make background music, or a full song with your own lyrics, in LUVI with MiniMax Music 2.6.
1:06Tutorial 16cTranscribing audio
Turn a recording into text with xAI STT v1, with speaker labels and filler words when you need them.
2:06Tutorial 17Talking avatars and lip sync
Make a portrait talk from an image and an audio file, or give an existing video a new voice with lip-sync models.
1:23Tutorial 18aGenerating 3D models
Turn an image or a description into a 3D model with Hunyuan 3D or HI3D, and see how extra views and parameters shape the result.
1:17Tutorial 18bThe 3D viewer and downloads
Inspect your 3D model in the viewer with lenses, lighting, wireframe and texture views, then download it as a GLB.