Model Details
Imagen 4 turns a text prompt into a high-quality, photorealistic image. Its strengths are strong prompt adherence (it follows detailed scene descriptions closely), markedly improved in-image text rendering (legible words, signage, and short labels), and crisp detail at up to 2K resolution. It is a general-purpose text-to-image model well suited to product shots, marketing visuals, concept art, and any scene where realism and faithful prompt-following matter. This is the Standard tier — the balanced default for quality and cost.
The only required input is `prompt`. Use `aspect_ratio` to set the frame (`1:1`, `16:9`, `9:16`, `4:3`, `3:4`) and `resolution` to choose `1K` (default) or `2K` output. Output is one generated image returned as a URL.
## Best for - Photorealistic product, lifestyle, and marketing images from a written brief - Scenes that need legible in-image text — short signage, labels, posters, packaging mockups - Detailed prompts where faithful adherence to the described composition matters - High-resolution stills up to 2K for print or large-format use
## Choose another model when - You want to edit, restyle, or extend an existing image rather than generate from text — use an image-to-image or image-editing model - You need motion or animation — use a text-to-video or image-to-video model - You want long passages of perfectly spelled text rendered in the image — even with improved text rendering, dense paragraphs can still degrade - You need many distinct variations in one call — this endpoint returns a single image per request
## Tips - Describe the subject, setting, lighting, lens/feel, and any short text you want rendered explicitly in the prompt - Use `resolution: "2K"` for print or large displays (same price as 1K); keep `1K` for faster, web-sized output - Set framing with `aspect_ratio` (e.g. `16:9` for banners, `9:16` for mobile/social, `1:1` for thumbnails)
## Safety & Content Moderation - `safety_tolerance` (advanced) controls how strict content filtering is, as a string from `"1"` (strictest) to `"6"` (most permissive); the default is `"4"`. Lower it for tighter moderation, raise it only when stricter filtering is rejecting acceptable prompts.
To run via the ModelRunner JavaScript client: ```js import { modelrunner } from "@modelrunner/client";
const result = await modelrunner.subscribe("google/imagen4", { input: { prompt: "a glass coffee mug on a marble counter, soft morning light, the word 'FRESH' on the mug", aspect_ratio: "1:1", resolution: "2K", }, }); ```