1 Prompt File Basics
AGITube accepts raw text input directly through the extension textarea or via uploaded .txt or .json prompt files.
Plain Text Format
In standard mode, simply place each prompt on a separate line. The automator processes each line sequentially from top to bottom:
Cinematic drone shot of futuristic cyberpunk Tokyo at night, neon reflections, 8k
Portrait of an ancient explorer holding a brass compass inside a dusty library
Minimalist architectural photography of a concrete villa in Scandinavia, soft morning lightBlank lines or lines beginning with // or # are automatically treated as comments and skipped by the queue runner.
Structured JSON Format
For advanced workflows requiring per-prompt aspect ratio or model overrides, use standard JSON array formatting:
[
{
"prompt": "Cyberpunk street vendor cooking noodles in rain",
"aspectRatio": "16:9",
"count": 2
},
{
"prompt": "Macro shot of mechanical watch gears rotating",
"aspectRatio": "1:1",
"count": 1
}
]2 Reference Image Anchors
Google Flow provides specialized reference slots to condition visual outputs. AGITube allows you to upload reference assets once and propagate them across your entire batch queue:
| Reference Slot | Purpose | Best For |
|---|---|---|
| Subject Reference | Locks identity, face, character, or product appearance | Portraits, consistent characters, mascot branding |
| Scene Reference | Preserves background geometry, architectural context, and spatial layout | Interior design, recurring environments, world-building |
| Style Reference | Directs artistic medium, color palette, lighting, and rendering brushwork | Oil painting, anime, vintage film grain, 3D claymation |
3 Tagging & Dynamic Variables
When running complex multi-character or episodic productions, you can assign tags to prompt inputs using the #name syntax and invoke them using @name.
Naming Generated Outputs (#name)
Appending #tag to any prompt instructs AGITube to remember the resulting generated asset in memory for subsequent steps in the same batch:
Portrait of Captain Elena in a futuristic naval uniform #elena
Close-up of a holographic navigational map #mapInjecting References Dynamically (@tag)
Reference previously captured outputs or uploaded library assets directly inside later prompts:
@elena looking intently at @map inside the starship bridge
Cinematic profile of @elena commanding her crew during an asteroid storm4 Procedural Sequence Chaining
Chaining allows you to create continuous storyboards or video sequences where each frame builds upon the previous generation:
// Step 1: Establish the character
Portrait of a lone astronaut exploring an alien crystal cave #astro
// Step 2: Progress the action using the established character
@astro discovering a glowing luminescent flower in the cavern walls #flower
// Step 3: Render cinematic video from the result
Animate @astro gently reaching out to touch @flower, camera slowly pushes in5 Video Generation & Frame Controls
AGITube natively drives Google Flow video generation modes. You can schedule pure text-to-video or precise image-to-video runs.
First & Last Frame Controls
For smooth morphs and transitions, specify start and end frames inside your video prompt line:
@start=scene_a.png @end=scene_b.png Time-lapse transition from sunset into starry night
@start=#astro Camera orbits 360 degrees around the astronaut standing on the cliff6 Pro Tips & Performance
To maximize speed and avoid Google Flow browser rate limits:
- Keep the Tab Active: Browsers throttle background tabs. Keep your Google Flow tab visible in a split window or dedicated screen for maximum processing throughput.
- Batch Sizing: We recommend running prompts in batches of 50 to 150 items for optimal memory management.
- Clean Prompts: Avoid special characters that could break URL encoding in Google Flow sessions.
- Download Folder Permissions: Ensure your browser has permission to save multiple files without prompting for each individual download.