Config | BetterAI Docs

Source: src/image_generation/utils/config.py. This page's original notebook was empty; the content was written from the code.

#Model catalog

IMAGE_MODELS_CATALOG describes what each Imagen model accepts. It is the source used by Params Validator.

Model	Max. images per prompt	Aspect ratios	Resolutions	MIME types
`imagen-4.0-generate-001`	4	`1:1`, `3:4`, `4:3`, `9:16`, `16:9`	`1K`, `2K`	`image/png`, `image/jpeg`
`imagen-4.0-fast-generate-001`	4	`1:1`, `3:4`, `4:3`, `9:16`, `16:9`	`1K`	`image/png`, `image/jpeg`
`imagen-4.0-ultra-generate-001`	4	`1:1`, `3:4`, `4:3`, `9:16`, `16:9`	`1K`, `2K`	`image/png`, `image/jpeg`

#Storage and database

Constant	Value	Use
`BUCKET_NAME`	`images`	Supabase Storage bucket.
`STORAGE_BASE_PATH`	`image_generations`	Base folder inside the bucket.
`ID_PREFIX`	`img_`	Prefix of each image identifier.
`BASE_URL`	Public URL of the bucket, ending in `/image_generations`	Base of the URLs returned to the client.
`DATABASE_NAME`	`GenAI`	MongoDB database.
`COLLECTION_NAME`	`ImageGeneration`	MongoDB collection.

#Default Gemini configuration

DEFAULT_CONTENT_CONFIG is the base that Image Generator Service merges with the configuration sent by the user.

Key	Default
`model`	`gemini-2.5-flash-image`
`temperature`	`0.75`
`top_p`	`0.85`
`max_output_tokens`	`1024`
`aspect_ratio`	`1:1`
`number_of_images`	`1`

Next to max_output_tokens, the code carries the annotation that gemini-2.5-flash-image uses 1024 or 2048 and gemini-3-pro-image-preview uses 4096 or 8192.

#Base prompt

BASE_PROMPT is the text, in English, that opens every prompt sent to Gemini. It defines the agent's role (visual creation and transformation) and the constraints: follow the user's request, apply the style of the reference images, preserve what was not asked to change, not add or remove elements without instruction, and prioritize consistency and quality.

#Helper types

Type	Description
`GeneratedImage`	`dataclass` with `image_bytes` and `mime_type`. It is the internal format of the generation output.
`Ratios`	Aspect ratios: `R1_1`, `R3_4`, `R4_3`, `R9_16`, `R16_9`.
`Resolutions`	Resolutions: `K1` (`1K`) and `K2` (`2K`).
`MimeTypes`	`PNG` (`image/png`) and `JPEG` (`image/jpeg`).
`ModelSpecs`	Groups `id`, `max_images`, `ratios`, `resolutions` and `mimes` of a model.
`Imagen`	Shortcuts for the models: `Imagen.GENERATE`, `Imagen.FAST` and `Imagen.ULTRA`.

from src.image_generation.utils.config import Imagen

# Access the identifier and the options of a model
model_id = Imagen.GENERATE.id
ratio = Imagen.GENERATE.ratios.R16_9
size = Imagen.GENERATE.resolutions.K2

Imagen.FAST defines Resolutions(K2=None), because this model only accepts 1K.

#Intro sentences

IMAGE_INTRO_SENTENCES is a list of 50 short sentences in Portuguese ("Aqui está a imagem conforme solicitado." and variations) to introduce a delivered image.