Skip to content

Known Limitations

modelBridge Does Not Limit Model Capabilities

Section titled “modelBridge Does Not Limit Model Capabilities”

modelBridge is a bridge, not a filter. Every model is exposed with its full capability as published by fal.ai — no parameters are removed, no outputs are restricted, and no model behavior is modified. The dynamic UI engine renders every input field from the model’s OpenAPI schema exactly as the model author intended.

If a model has limitations — resolution constraints, duration limits, content restrictions — those are limitations of the model itself on fal.ai, not limitations imposed by modelBridge. modelBridge adds Premiere Pro integration, real-time cost estimates, timeline import, validation, and workflow tools on top of the model’s native capabilities.

For model-specific capability details, see the model’s page on fal.ai/models.

To run modelBridge.ai, you need:

  • macOS 12 (Monterey) or later — Windows and Linux are not supported
  • Adobe Premiere Pro 2023 (version 23.x) or later
  • Node.js 18 or later — required for the local backend server
  • Active internet connection — all AI generation happens on fal.ai’s cloud infrastructure
  • fal.ai account — free to create at fal.ai, required for API access and billing

modelBridge.ai runs exclusively on macOS. The backend server uses macOS-specific tools including lsof for port conflict detection and Homebrew-based Node.js paths. There is no Windows or Linux version at this time.

The extension is built for Premiere Pro’s CEP (Common Extensibility Platform). It does not work in After Effects, Media Encoder, or other Adobe applications.

All AI generation happens on fal.ai’s cloud infrastructure. An active internet connection is required for:

  • Generating content
  • Fetching model schemas
  • Uploading source media
  • Downloading results
  • License validation

There is no offline mode or local generation capability.

The local backend server runs on port 3000. If another application (another Express server, a development environment, etc.) is using this port, modelBridge.ai will report a port conflict error.

Pre-generation cost estimates are not guaranteed to match the actual charge from fal.ai. Some models have parameter-dependent pricing that is not fully captured in the estimate. The “From $X.XX” badge indicates a minimum starting price — actual costs may be higher.

When fal.ai does not expose pricing for a model, the cost badge shows “Price not available.” You can still generate, but you will not see a cost estimate beforehand. Check your fal.ai dashboard for the actual charge.

Some fal.ai models have constraints (minimum dimensions, maximum duration, file size limits) that are not declared in their OpenAPI schema. modelBridge.ai handles this through self-learning: when a generation fails due to an undeclared constraint, the plugin extracts the requirement from the error message and enforces it automatically on future attempts. The first attempt may fail, but the same error will not happen twice.

modelBridge.ai connects to existing models on fal.ai. You cannot train, fine-tune, or create custom models through the plugin. For custom model training, use fal.ai’s platform directly.

Models are hosted and maintained by fal.ai. A model may become unavailable, be renamed, or be removed without notice from modelBridge.ai. If a previously working model stops functioning, check fal.ai’s status page or search for the model by name to find its current endpoint.

You can generate one clip at a time per model (or two in Dual Mode). There is no queue system for scheduling multiple generations in sequence. However, the Active Generations Panel allows you to start a new generation while previous ones are still processing in the background.

When a generation is handed off to background processing, it continues polling for up to approximately 30 minutes. If the generation takes longer than that, it will time out. See Background Generations for details.

There is no streaming preview during generation. You see the result only after the full generation is complete and downloaded.

The plugin operates in Premiere Pro’s panel system, which constrains the available width to approximately 280px minimum. All UI elements are designed for this narrow layout, which limits the complexity of some interfaces (particularly mask drawing for inpainting).

The prompt field accepts very long text input without client-side truncation. However, the AI model may silently truncate or reject prompts that exceed its internal limit. Keep prompts concise and focused for best results.

Drag-and-drop from Finder is a macOS feature with no Windows equivalent at this time.