What’s new
- The quality tier of the Imagen 4 family: launched 2025-08-15 alongside the general availability of the whole Imagen 4 family, positioned as the tier with the highest detail and strictest prompt adherence, built for demanding use cases.
- Resolution and text rendering both improved over Imagen 3: supports up to 2K resolution, with meaningfully better text rendering per Google.
- Google specifically calls it out for advanced use cases: Vertex AI's docs recommend starting with Gemini for image generation by default, and reaching for Imagen 4 Ultra when image quality, texture, or branding/logo/layout work demands it.
- Covers the common social/ad/TV aspect ratios: 1:1, 16:9, 9:16, 3:4 and 4:3, defaulting to 1:1.
What it’s good for
- Work where image quality, fine detail and strict prompt adherence matter most — Google explicitly recommends it for these "advanced use cases."
- Static, precisely composed brand assets — logos and layout work that can't tolerate flaws.
- Work that needs accurate on-image typography (like poster copy), as long as the text stays short.
Prompting tips
- Structure prompts in three parts — subject, context, style: say what's in the shot, where it's placed, then what artistic approach you want.
- Lean into quality modifiers: since you've already chosen Ultra, terms like "high-quality," "4K," "HDR" or "professional photography" help it show its strengths.
- Use photography language directly: shot size, lighting direction, lens type (macro/wide-angle/35mm) and bokeh backgrounds can all be specified directly.
- State exclusions plainly: write "exclude: walls, frame" rather than "no walls."
- Keep on-image text to 25 characters or fewer: longer text tends to render poorly.
- Name the art style specifically: naming a movement (impressionist, renaissance, pop art) gives more consistent results than a vague adjective.
Product photography of a mechanical watch, high-quality, 4K, HDR, professional studio lighting,
cool dramatic lighting, macro lens, emphasizing the dial and gear details, bokeh background.
1:1. Exclude: people, hands, text.
Known limits
- Pure text-to-image — no editing support: there's no reference/input image field in the API, so it can't edit or work from an existing image.
- On-image text is recommended to stay under 25 characters: Google's guide suggests this limit — longer text tends to render poorly.
- Speed isn't the point here: Google positions it as the quality-first tier — for generation speed or high-volume work, the Fast tier is the better fit.
Using it in Nomi
On the generation canvas, add a image card, pick “Imagen 4 Ultra” in the model picker and choose a mode (Text to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
Sources
- Announcing Imagen 4 Fast and the general availability of the Imagen 4 family (Google Developers Blog, 2025-08-15)
- Imagen 4 overview (Google Cloud Vertex AI official docs)
- Generate images with Imagen (Google Cloud Vertex AI official docs)
- Imagen prompt and image attribute guide (Google Cloud Vertex AI official docs)
- Imagen 4 Ultra API reference (Kie.ai)
