Skip to content
All field notes

Models

Choose the model for the shot, not the whole project

A grounded approach to selecting image, video, audio, and 3D models inside a mixed-media workflow while keeping cost and iteration visible.

Several distinct creative materials aligned for comparison.

“Which model is best?” is usually the wrong level of question. A campaign can need crisp product photography, expressive motion, synchronized sound, and a rotatable object. Those are different jobs with different constraints.

Storyloom keeps a curated model catalog behind a shared provider contract. The catalog currently spans image, video, audio, and 3D workflows, and the public Models page reads directly from that source. When the catalog changes, the marketing count changes with it.

Define the shot before the model

Write down the output you need in observable terms. Is the frame expected to preserve a product exactly? Does the clip need native audio? Is typography inside the image important? How long can the team wait for a result, and how many iterations are likely?

Those questions are more useful than a generic quality ranking. A slower model can be a good choice for a final hero and a poor choice for a broad composition search. A fast open-weight image model can be ideal for the first thirty minutes of a direction, even if another model handles the finishing pass.

Use the canvas as the comparison surface

Run a controlled comparison with the same prompt and references. Place the results next to each other and label the model and intent. Compare what matters for that shot: subject consistency, composition, motion coherence, texture, or timing.

This is different from browsing a provider gallery. You are testing models against your material and your constraints. Keeping every result on the board also makes the trade-off discussable. “This one is better” becomes “this one holds the package shape and gives us enough negative space for the headline.”

Separate exploration from finishing

A useful workflow has at least two model roles:

  1. An exploration model that is fast or economical enough for broad iteration.
  2. A finishing model chosen for the specific quality the selected direction needs.

Video and audio may add more roles. You might use one model to create motion, another to generate a sound bed, and a local FFmpeg step to assemble the sequence. Storyloom's typed action layer lets those steps share a board without pretending they came from one universal model.

Keep price information close to the action

The studio's catalog stores a unit price and unit type for each listed model. The copilot can show an estimate before a paid action runs. Per-second and per-character models need output assumptions, so estimates are useful decision inputs rather than promises about the final provider invoice.

Storyloom is bring-your-own-key. Provider charges go to the account connected by the creator or the deployment. That makes cost visibility especially important: the interface should help you understand the action before it uses your provider budget.

Browse the source-backed model catalog, then test the models that fit your shot inside the studio.

Continue on the canvas

Turn the method into a piece of visible work.

Open Storyloom without an account or a provider key. Add your own models when the direction is ready for production.