Data & Localization Nodes

Work with data feeds and create localized content variants.

Data Feed

Import structured data into workflows

Provides a structured data source for batch processing. Loads data items and applies field mappings to make them available to downstream nodes through the batch loop system.

Property	Description
Inputs	trigger (any, optional)
Outputs	items (any - Items Array), count (text - Item Count), schema (any - Data Schema)

Configuration:

Setting	Description
Items	Array of data objects (entered via JSON or table editor)
Field Mappings	Alias/override mappings from source fields to target fields

Data Structure:

Each item is a JSON object with any fields you need:

[
  {"name": "Widget A", "price": 29.99, "image_url": "https://..."},
  {"name": "Widget B", "price": 39.99, "image_url": "https://..."}
]

Field Mappings:

Field mappings create aliases for item properties. For example, mapping prompt to name means downstream nodes can access $item.prompt which resolves to the item's name field.

How it works:

1. Items are loaded from node config
2. Field mappings are applied as aliases/overrides
3. Each mapped item preserves all original fields at the top level
4. An _index field is added for tracking
5. Items, count, and schema are exposed as outputs
6. The dataFeed is set in context for downstream nodes

Use with Batch Processing:

Data Feed --> Batch Processor --> Text to Image --> Batch End
    |
[array of items]

Context variables available during batch iteration:

• $item - Current item object
• $item.name - Access any item property
• $item.image_url - Access image URLs for Remote Media
• $index - Current iteration index (0-based)
• $total - Total number of items

Use cases:

• Drive batch image generation from product catalogs
• Provide structured data to workflows
• Feed multiple items through processing pipelines

Credit cost: Free (no credits)

---

Variant Generator

AI-powered content variant generation with localization

Creates multiple image variants from a source image using a two-phase pipeline: first generates variation prompts via AI, then generates each variant image via WaveSpeed.

Property	Description
Inputs	content_in (image - Content), variant_count (text - Variant Count)
Outputs	variants (image - Variants Array), metadata (any - Variant Metadata)

Configuration:

Setting	Default	Description
Mode	auto	Generation mode: auto or guided
Variant Count	3	Number of variants to generate (1-50)
Dimensions	['color']	Variation dimensions (color, layout, composition, etc.)
Guidance Prompt	(empty)	Theme description for guided mode
Custom Instructions	(empty)	Additional generation instructions
Naming Pattern	{original}_{locale}_{variant}	Output file naming pattern
Seed	(empty)	Reproducibility seed
Include Original	true	Include the source image in output
Model	bytedance/seedream-v5.0-lite/edit	Image generation model

Supported Models:

Model	Description
bytedance/seedream-v5.0-lite/edit	Default - SeedDream v5 Lite edit model
google/nano-banana-2/edit	Nano Banana 2 edit model

Generation Modes:

Auto Mode

AI generates unique variations automatically based on the selected dimensions without specific constraints.

Guided Mode

AI follows your guidance prompt theme:

• Provide a theme description (e.g., "cyberpunk style, different lighting")
• AI generates variations following the theme
• Ideal for creative exploration with direction

Two-Phase Pipeline:

1. Phase 1 - Prompt Generation: Calls the generate-variants edge function with promptOnly=true to get AI-generated variation prompts based on the source image, dimensions, and mode
2. Phase 2 - Image Generation: For each prompt, calls generate-image via WaveSpeed and polls for completion. Each variant is generated independently.

Locale-Aware Generation:

When locale data is connected (via context or upstream nodes), the Variant Generator adapts variants for different markets:

Locale Code	Cultural Adaptation
en-US	Bold, direct, aspirational imagery
en-GB	Understated elegance, subtle sophistication
ja-JP	Clean minimalism, harmony and balance
ko-KR	Modern aesthetics, soft gradients, youthful energy
zh-CN	Vibrant reds and golds, modern luxury
de-DE	Precision, quality-focused, clean functional design
fr-FR	Chic, sophisticated, artistic flair
es-ES	Warm colors, passionate energy
pt-BR	Vibrant tropical colors, energetic mood

With locales: total variants = variant count x number of locales.

Naming Pattern Variables:

Variable	Description
{original}	"original" for source, "variant" for generated
{variant}	"original" or "v1", "v2", etc.
{index}	Variant index number
{locale}	Locale code (removed if no locales)

Output:

• variants: Array of image artifacts (includes original if includeOriginal is true)
• metadata: Object with count, dimensions, naming pattern, and model info

Use cases:

• A/B testing content variations
• Multi-market campaign imagery
• Color and composition variants
• Locale-adapted marketing content

Credit cost: Credits per variant based on the image generation model used

---

Workflow Examples

Data-Driven Batch Generation

Generate images from a product catalog:

Data Feed (products) --> Batch Processor --> Text to Image ($item.name) --> Batch End --> Preview Media

Variant Generation Pipeline

Create multiple variations of an image:

Upload Media --> Variant Generator (3 variants, color) --> Preview Media

Localized Campaign

Generate locale-adapted variants:

Upload Media --> Variant Generator --> Preview Media
                    |
             [locales: en-US, ja-JP, fr-FR]
             [3 variants x 3 locales = 9 outputs]

Data Feed with Remote Media

Process external images from data:

Data Feed --> Batch Processor --> Remote Media ($item.image_url) --> Image Upscale --> Batch End

---

Tips

Data Feed

1. Field mappings - Use mappings to normalize data from different sources
2. JSON format - Ensure items array contains valid JSON objects
3. Batch pairing - Always pair Data Feed with Batch Processor for iteration

Variant Generator

1. Start small - Test with 2-3 variants before scaling up
2. Guided mode - Use guidance prompts for more controlled variations
3. Include original - Keep original in output for easy comparison
4. Cultural context - Locale-aware generation adapts imagery to market preferences automatically