Skip to content

Image Generation

Image Generation

Generates images from a text prompt, with support for style instructions, model settings and reference images.

On this page

#Image Generation - `/davinci/image-generation`

POST/davinci/image-generation

Generates images from a text prompt, with support for style instructions, model settings and reference images. Use it when you need to create or transform images programmatically via a multimodal LLM.

#Parameters

Parameter	Type	Description	Example
`user_input`	`string`	Main prompt describing the image to be generated.	`"Uma pizzaria napolitana à noite"`
`instructions`	`string` — optional	Additional style instructions or creative guidelines for the generation.	`"Estilo de animação 3D como as da Disney"`
`config`	`string` (JSON) — optional	Model settings: `model`, `temperature`, `top_p`, `max_output_tokens`, `aspect_ratio`, `number_of_images`.	`{"model": "gemini-2.5-flash-image", "aspect_ratio": "9:16"}`
`files`	`file[]` — optional	Reference images to guide the generation. Accepts multiple uploads.	`reference.png`

#Config (explanation of each parameter)

The config field is optional. When not sent, the API uses the generation module's internal DEFAULT_CONTENT_CONFIG. The fields sent override only the corresponding values.

Field	Type	What it does	Default	When to adjust
`model`	`string`	Multimodal model used for generation (`gemini-2.5-flash-image`, `gemini-3-pro-image-preview`).	`"gemini-2.5-flash-image"`	Switch to the `pro` model when you need higher quality or more output tokens.
`temperature`	`float` (0.0–2.0)	Controls the creativity of the generation. Low values = more predictable results.	`0.75`	Lower it for greater fidelity to the prompt, raise it for more creative variation.
`top_p`	`float` (0.0–1.0)	Nucleus sampling: limits the set of tokens considered.	`0.85`	Lower it for more conservative and consistent outputs.
`max_output_tokens`	`integer`	Output token limit. `gemini-2.5-flash-image`: `1024` or `2048`. `gemini-3-pro-image-preview`: `4096` or `8192`.	`1024`	Raise it when using `pro` models or when the text response is truncated.
`aspect_ratio`	`string`	Aspect ratio of the generated image. See accepted values below.	`"1:1"`	Adjust it according to the image's destination (feed, stories, banner).
`number_of_images`	`integer` (≥ 1)	Number of images generated. Each image is an independent call to the model.	`1`	Raise it to get variations, remembering that cost and latency grow proportionally.

#Accepted values for `aspect_ratio`

Value	Format	Typical use
`1:1`	Square	Feed post, avatar, thumbnail
`3:4`	Portrait	Portraits, product catalog
`4:3`	Landscape	Classic images, presentations
`9:16`	Full portrait	Stories, Reels, TikTok
`16:9`	Widescreen	Banners, covers, video/hero

#Request

curl --location 'http://localhost:8000/davinci/image-generation' \
--form 'user_input="Crie a imagem de uma pizzaria napolitana"' \
--form 'instructions="O estilo deve ser uma animação 3D como as da Disney"' \
--form 'config="{
  \"model\": \"gemini-2.5-flash-image\",
  \"temperature\": 0.75,
  \"top_p\": 0.85,
  \"max_output_tokens\": 1024,
  \"aspect_ratio\": \"9:16\",
  \"number_of_images\": 2
}"' \
--form 'files=@"/path/to/file"'

#Response

{
    "job_id": "job_1788272755499472800EkNu",
    "status": "success",
    "status_code": 200,
    "result": {
        "text_response": "Com certeza! Aqui está uma pizzaria napolitana em estilo de animação 3D da Disney:\n\n \n",
        "images": [
            "https://hsenyunovbrmjejxqvjn.supabase.co/storage/v1/object/public/images/image_generations/img_1788272774157403200z8UW",
            "https://hsenyunovbrmjejxqvjn.supabase.co/storage/v1/object/public/images/image_generations/img_17882727741574032007MBY"
        ]
    },
    "time": {
        "start": "2026-09-01 11:25:55",
        "end": "2026-09-01 11:26:19",
        "duration_seconds": 24
    }
}

Source: src/web_services_network/routes/davinci.py