fo-model-picker: model defaults

Plugin types: Model Pickers.

fo-model-picker v1 supplies a fixed list of default models and routes for each render kind and stage: text, still images, video and speech. It needs API 1, revision 1. It works locally without a key or network setup; it does not render. FilmOpen checks whether the selected model and route are available.

Choose a stage

Install and allow fo-model-picker in Settings → Plugins. On an editable character’s Media card, choose Prototype, Draft or Final. The stage applies to that version’s renders. On narrow cards the control shows initials; its tooltips give the full meaning:

• “Prototype: a fast, inexpensive model, to brainstorm concepts that are still fluid.”
• “Draft: the same model as Final; where the technology allows it, it first renders a low-resolution version, which you can render again at final resolution.”
• “Final: the best model, at full resolution.”

Open Render from a media place, such as Reference images, Sheets, Motion clips or Voice samples. The popup confirms the stage with Renders at: Prototype, Renders at: Draft or Renders at: Final. Inspect Model and Runs on before continuing; a default is not proof that its provider is ready.

The fixed image defaults

Image task	Prototype	Draft and Final
Generate from a prompt	Z-Image Turbo on fal; 1024×1024, prompt expansion off	GPT Image 2.5 Sunburst on fal; 1024×1024, high quality
Edit from a reference	FLUX.2 klein 4B on fal	GPT Image 2.5 editing, Sunburst on fal; high quality

For still images, Draft and Final use the same model and settings; Draft is not automatically smaller or cheaper. The package supplies no alternative suggestions.

Text, video and speech

• Text (llm): Claude Sonnet 5 through OpenRouter at every stage.

• Video, from text (t2v) or from an image (i2v), on fal, each 5 seconds:

  Prototype	Draft	Final
  MiniMax H3 Max, 480P	Seedance 2.5, 480p	Seedance 2.5, 1080p

  For video, Draft renders at 480p and Final at 1080p.

• Speech (tts): Eleven v3 on fal at every stage.

The other kinds have no default, each with a reason: flf2v has no last-frame control, ia2v has no lip-sync route, and flfa2v, ttv and ttm have no fal model.

The image, video and speech routes use fal, which must be enabled with a usable fal key. A character sheet needs a reference image first. The prompt writer can supply the words; you can inspect the model, prompt and price before rendering.