Skip to main content

Content Engine

Influgen's content engine is the production layer behind every image, caption, reel, and calendar slot. Its job is not only to generate assets, but to generate assets that still feel like the same character when they reach the queue.

What goes into a generation​

Every content request combines three inputs:

  1. Character context: niche, appearance, identity, personality, style, settings
  2. Request context: platform, content type, prompt, planned date, language, quality preferences
  3. Consistency context: face lock, reference images, and optional LoRA

Influgen stores this as a content item plus a config snapshot. That gives you a traceable record of what the system actually used, instead of a fragile prompt hidden in UI state.

Supported content types​

The engine currently supports:

  • single
  • carousel
  • story
  • video
  • reel
  • short

Platform validation happens before generation starts. For example:

  • Instagram supports all content types
  • TikTok and YouTube require video-based types
  • X and Pinterest accept single-image or carousel items

The generation lifecycle​

Content items move through a clear state model:

  • queued
  • generating
  • ready
  • approved
  • rejected
  • published
  • failed

That state model is what makes Influgen operationally useful. You can batch work, inspect output, and publish later without losing the link between generation and execution.

[screenshot: Queue view showing multiple content items moving from queued to ready, with face-check and caption details in the item drawer]

Prompt assembly​

The engine does not send your raw prompt directly to the model. It enriches the prompt with character data and platform-specific context. The result is closer to:

  • who this character is
  • how this platform behaves
  • what style system the character should stay inside
  • how strict identity preservation should be

This is why a good character profile makes every later generation better.

Face-lock and quality gates​

By default, generation includes identity-preservation settings such as:

  • preserve_identity: true
  • face-lock strength from character settings, default 0.9
  • quality threshold from character settings, default 0.85
  • reference-image count for the adapter layer

In practice, Influgen uses those values in two places:

  • while generating, to steer the output toward the character
  • after generating, to help decide whether the result should be considered good enough

Manual mode vs. autopilot mode​

The same generated asset lands differently depending on the character's approval mode:

  • Manual mode: new content finishes as ready
  • Autopilot mode: new content can finish directly as approved

Autopilot is powerful, but only once the character has proven stable.

What makes a strong request​

The best prompts are concrete and visual:

  • environment
  • lighting
  • wardrobe
  • camera framing
  • emotional tone

Weak prompt:

Make a cool post for my influencer.

Strong prompt:

Golden-hour rooftop portrait in tailored cream athleisure, editorial framing, warm sun flare, calm luxury wellness mood, confident direct gaze.

Review controls​

Influgen supports review at two levels:

  • whole content item approval or rejection
  • per-media approval or rejection inside multi-image results

That matters for carousels. You can reject one weak frame without throwing away the whole batch.

Batch generation vs. one-off generation​

Use one-off generation when you are exploring the character. Use weekly generation when you already know the tone and want throughput. The underlying engine is the same, but batch generation adds planning and queue orchestration on top.