Self-Learning Validation

Self-learning validation is one of eight intelligent systems that make modelBridge an adaptive, self-improving system. While the dynamic UI renders itself from schemas and cost estimates update in real time, the validation system takes it a step further — it learns from failures and improves over time.

The self-learning pattern now extends beyond validation — modelBridge also learns costs from your billing history (see modelBridge Cost Intelligence for how the same observe-learn-improve cycle applies to cost estimation) and quietly adapts to model catalog changes on fal.ai’s side (see Endpoint and Schema Adaptation below).

modelBridge learns from every generation. When something goes wrong — a file that’s too large, an aspect ratio the model doesn’t support, a clip that’s too long — modelBridge remembers the exact requirements for that model.

The next time you work with the same model, those requirements are checked automatically before anything is sent. Repeat mistakes on the same model are caught before they cost money.

This happens silently in the background. You don’t manage it, configure it, or think about it. It just gets smarter.

How It Works

1. A generation fails

You generate with a model that has undeclared requirements. fal.ai returns an error — for example, “Image dimensions are too small. Minimum dimensions are 300x300 pixels.”

2. modelBridge remembers

The plugin reads the error, identifies the exact requirement, and saves it permanently for that model. This data survives cache clears, Premiere Pro updates, and plugin reinstalls.

3. Future attempts are checked automatically

On every subsequent attempt, the learned requirement is checked before anything is sent. If your media doesn’t meet the requirement, you see:

• Red border on the media card
• Specific error message: “Image too small (128x128, min 300x300)”
• Generate button disabled

No API call. No charge. No waiting.

What It Catches

The system covers common media constraints: image dimensions (too small or too large), file size limits, aspect ratio requirements, and video duration limits (too long or too short). New constraint types are handled automatically as they appear — the system is designed to grow.

The Acknowledgment

When a constraint is learned for the first time, the error banner shows:

“Requirement saved — this will be caught before generation next time.”

This only appears once per model per constraint type. If the same error fires again, the acknowledgment is suppressed — the preflight system should have caught it.

Constraint Priority

Learned constraints never override schema-declared values. The merge strategy is “fill gaps only” — if the model’s schema already declares a maximum width, a learned constraint will not overwrite it. Schema is the source of truth. Learned constraints cover what the schema did not declare.

Bootstrap from Generation History

On first load — or after a reinstall — the self-learning system bootstraps from your existing generation history. Any constraints previously learned are restored from the disk backup file automatically, so you do not start from zero even if localStorage was cleared.

Staleness

Each constraint is timestamped when learned. Older constraints are still enforced but may not reflect recent model changes. Re-learning (hitting the same error again) refreshes the timestamp.

Persistence Through Updates

Learned constraints survive:

• Plugin updates (data stored outside the extension folder)
• Premiere Pro updates
• CEP cache clears (disk fallback restores localStorage automatically)

You do not lose learned data when updating modelBridge.

Endpoint and Schema Adaptation

AI model providers regularly update, move, or deprecate their API endpoints. modelBridge handles all of this automatically. You never deal with broken models, stale interfaces, or cryptic API errors from endpoint changes.

Silent endpoint migration

If a model’s API endpoint changes — renamed, versioned, or reorganized by the provider — modelBridge detects the move and silently updates the connection. Your installed model continues working without any action on your part.

You won’t see a notification, a loading screen, or a “we fixed this for you” message. The model just works.

Schema drift detection

When a model’s input parameters change — fields added, removed, renamed, or constraints modified — modelBridge detects the difference and silently updates the interface. No stale controls, no missing inputs, no outdated validation rules.

This matters because AI model providers frequently iterate on their parameters. A model might add a new quality setting, change a slider’s range, or rename a field between versions. Without drift detection, you’d see broken forms or miss new options entirely.

Graceful deprecation

If a model is permanently removed by the provider with no replacement, modelBridge clearly communicates this on the model card with an option to remove it — instead of showing cryptic API errors when you try to generate.

A model is only marked as removed after multiple consecutive failed checks, preventing false alarms from temporary provider outages. Health checks run in the background during normal plugin use — throttled, non-blocking, no loading screens.