Imagen 4 follows a prompt more literally than almost any other image model, so the craft is in what you state, not in tricks. Google's own formula is the one that works: subject, then context, then style — "a sketch of a modern apartment building surrounded by skyscrapers" — layered with photography modifiers like lens type, camera position and lighting temperature. Text renders best at 25 characters or less. There is no editing mode and no negative prompt: you get exactly the picture you describe, which is the point.
The object, person, animal or scenery the image is of. Make it the grammatical core of the sentence — Imagen's prompt adherence means a precise subject comes back precisely.
Where the subject sits and what surrounds it: "on a rain-slicked rooftop at night", "surrounded by skyscrapers". Context words set scale, depth and story.
General ("photograph", "painting", "sketch") or very specific ("pastel painting", "charcoal drawing", "isometric 3D render"). The style word recasts the entire image, so choose it deliberately.
Photography modifiers Imagen responds to by name: proximity ("close-up", "zoomed out"), position ("aerial", "from below"), lens ("35mm", "50mm", "fisheye", "macro"), and looks like "motion blur", "soft focus", "bokeh", "portrait".
Temperature and drama: "natural light", "dramatic light", "warm", "cold", "golden hour", "studio lighting". Imagen renders lighting words faithfully rather than averaging them into a house look.
Quoted, 25 characters or fewer, at most two or three distinct phrases. Suggest a general font style and a size feel; expect to iterate a couple of times for perfect placement.
Template
A [style/medium] of [subject] [in context/background], [camera position and lens], [lighting], [optional quoted text under 25 characters]A photograph of a ceramicist shaping a bowl at her wheel in a sunlit studio, clay-dusted hands in sharp focus, shelves of glazed pottery softly blurred behind her, 35mm lens, natural window light, warm tones.
The full formula in one sentence: subject (ceramicist), context (sunlit studio, shelves), style (photograph, 35mm). Imagen's prompt adherence rewards this density — every stated element shows up.
A studio product photograph of a brushed-steel espresso machine on a seamless white background, slightly elevated three-quarter view, soft even studio lighting with a gentle reflection under the base, crisp detail, commercial photography.
For catalog-clean output, say "studio product photograph" and "seamless white" explicitly — Imagen will not improvise a lifestyle scene around a product unless you ask for one.
A minimalist concert poster with the text "MIDNIGHT RUN" in bold condensed sans-serif capitals, centered over a grainy duotone photograph of a runner silhouetted against city lights at night, deep navy and hot orange palette.
Twelve characters of quoted text is well inside the 25-character comfort zone, so the type comes out clean. Describing the font's character ("bold condensed sans-serif") steers letterforms without naming a specific typeface.
A macro lens photograph of a honeybee landing on a lavender sprig, extreme close-up on the wing texture and pollen grains, shallow depth of field with creamy bokeh, morning light, high detail.
"Macro lens" plus "extreme close-up" is a proximity instruction Imagen takes literally — you get true macro framing, not a cropped wide shot. Bokeh and depth-of-field words shape the background.
An aerial photograph taken from high above a winding turquoise river cutting through autumn forest, zoomed out to show the full oxbow curve, low warm evening sun casting long shadows across the treetops.
Camera position ("aerial", "from high above") and proximity ("zoomed out") are separate dials — set both. The lighting clause adds the long-shadow drama that makes aerials read as photography instead of maps.
A 35mm portrait of a detective in a rain-soaked trench coat under a streetlamp, film noir style, hard dramatic light from above, deep shadows across the face, cold tones, shallow depth of field.
"35mm portrait, film noir" is straight from Google's recommended modifier vocabulary. Style, lighting drama and temperature stack cleanly because each occupies its own slot instead of competing.
A charcoal drawing of an old fishing boat leaning on a misty shore at low tide, loose expressive strokes, smudged shading in the sky, white highlights scratched into the water.
Swapping the style word recasts everything: the same subject as "a pastel painting" or "an isometric 3D render" produces a completely different image. Naming medium-specific techniques (smudged shading, scratched highlights) sells the medium.
An isometric 3D render of a tiny cozy coffee shop interior, cutaway view showing the counter, espresso machine, two tables and a bookshelf, soft pastel palette, clean smooth surfaces, studio lighting.
"Isometric 3D" is one of the specific styles Imagen handles crisply — geometry stays consistent and edges clean. Cutaway framing plus an object list gives you an illustration ready for a landing page.
A photograph of a cyclist sprinting through a city intersection at dusk, panning shot with motion blur on the background and the wheels, the rider in sharp focus, neon storefronts streaking past, cool blue hour light.
"Motion blur" is a supported look — the trick is saying where it applies (background, wheels) and what stays sharp (the rider). Without that split the whole frame smears.
A photograph of a stack of blueberry pancakes on a rustic wooden table, morning light — then iterate: add "syrup mid-pour, steam rising, 50mm lens, shallow depth of field, warm golden tones" and rerun on Imagen 4 Ultra.
The intended workflow: start with subject and context on Imagen 4 Fast, then layer modifiers one pass at a time — each addition is honored, so the image converges instead of mutating. Promote the final prompt to Ultra unchanged.
Imagen 4 Fast to explore compositions cheaply, Imagen 4 as the everyday default, Imagen 4 Ultra for finals where detail and text quality matter. Prompts transfer across the three unchanged.
1:1, 3:4, 4:3, 9:16, 16:9. Pick it first and write framing to match — "aerial, zoomed out" wants 16:9, a portrait wants 3:4 or 9:16.
Up to 480 tokens, but you rarely need a tenth of that. One dense, well-ordered sentence usually beats a paragraph — add clauses only when each earns its place.
In Workflow Studio, Imagen 4 shines as the stills node in a chain: generate the frame here, then feed it downstream into video or editing nodes for animation and changes.
Imagen 4 was trained to respond to photographic vocabulary as control input. Treat these as four independent dials you can set in any combination — each one named explicitly in the prompt:
Imagen 4 produces clean typography inside the 25-character-per-phrase limit, with up to two or three distinct phrases in one image. Quote each phrase exactly, suggest a general font character ("bold condensed sans-serif", "elegant serif") rather than a typeface name, and hint at placement. Rendering position and size can vary between runs, so plan on a couple of iterations for a poster-grade result — regenerating is cheap on the Fast variant.
Because Imagen follows structure so faithfully, template prompts work: freeze the skeleton and swap the slots. "A [style] of [subject] in [context], [lens], [lighting]" with three slots filled per run yields a visually consistent series — same grammar, same modifier set, different content. This is the natural pattern inside Workflow Studio, where one image node with a templated prompt can serve an entire batch of variations.
The everyday default — full quality with strong prompt adherence and clean photographic detail.
Drafting and composition search: quicker, cheaper runs while you converge on the prompt, before promoting it unchanged.
Final assets — maximum detail and the best typography. Use it once the prompt is settled, not while exploring.
No — it is text-to-image only. Generate the still with Imagen 4, then do changes with an editing model on Clipwave (Nano Banana for conversational edits, Seedream Edit for reference-based swaps).
No. State what you want instead of what to exclude: "a deserted beach at dawn" rather than "beach, no people". With Imagen's literal prompt-following, positive description is enough.
Keep each phrase to 25 characters or fewer, and use at most two or three phrases per image. Quote the exact words and describe a general font style; expect to rerun once or twice to nail placement.
Fast while iterating — it is built for quick, cheap composition search. Ultra for the final render once the prompt is locked. The middle Imagen 4 covers everything in between. Prompts move across all three without changes.
In the Workflow Studio model picker, as three entries: Imagen 4, Imagen 4 Fast and Imagen 4 Ultra. Drop an image node into a flow, select the variant, and its output can feed any downstream node.
Five: 1:1 (the default), 3:4, 4:3, 9:16 and 16:9. Set the ratio to match your placement first, then write the composition language to fit that frame.