---
name: bildy-model-bake-off
description: Run one prompt through several image or video models at once and lay the results out side by side with what each one cost. Use when someone asks which model to use, whether a cheaper model is good enough for a job, or wants to see the difference between Nano Banana, Seedream, Flux and the rest on their own prompt.
---

# The same prompt on several models

Nobody can tell you which model suits a job, because it depends on the prompt.
Running the prompt on four models costs a few cents and settles it.

## Picking the line-up

Call `images_list_models` or `videos_list_models`. Each one tells you the
price per image or per second, so you can put a cheap model, a mid model and
an expensive one in the same run and see whether the expensive one earns it.

Four models is usually enough. For images that is four times the price of one
image, which on the cheaper models is under a cent each.

Some jobs have an obvious shortlist:

- Text inside the picture: the models that handle typography, which the list
  names in their descriptions.
- Editing a photo rather than making one: only the editing models, called
  through `images_edit`.
- A person who has to look the same across several pictures: the models built
  for character consistency.

## Steps

1. Get the prompt. Use theirs exactly, unchanged, on every model. Changing a
   word between runs makes the comparison worthless.
2. Fire one `images_generate` call per model, all of them, without waiting for
   each to finish.
3. Collect the job ids and poll `images_get_job` on each until they are all
   completed.
4. Show the results together, each one labelled with the model name and what
   that image cost.
5. Say which one you would pick and why, in terms of the prompt. "Seedream got
   the hands right and Flux did not" helps. "Seedream is better" does not.

## Reading the results

Judge against the prompt rather than against each other. A picture can be
prettier and still be the wrong answer, because it left out half of what was
asked for.

Run it twice on the two finalists before deciding. Models vary a lot between
runs on the same prompt, and one lucky image is not a verdict.

Weigh the cost in too. If the cheap model gets it right eight times out of ten
and the expensive one gets it right ten times out of ten at twelve times the
price, the cheap one wins for a big batch and loses for a single hero image.

## For video

The same idea works with `videos_generate`, but a clip costs far more than an
image and takes minutes. Run the shortest length each model allows, and only
compare two or three models.