How to prompt Seedream

Studio product photo of a matte-black insulated steel bottle with a brushed bamboo cap, centered on a seamless #f5f5f5 background, soft top light with subtle bounce fill, true-to-life color, sharp edges, realistic reflections on the metal, product photography, 85mm lens.

The workhorse template: subject with real materials, hex-coded background, one light. The hex code is not pedantry — named colors drift between generations, hex values don't.

Overhead flat-lay of a skincare routine — amber glass serum bottle, white ceramic cream jar, jade roller — arranged on veined white marble, soft diffused daylight from the top-left, gentle shadows, editorial product photography, crisp texture detail, 4K.

Flat-lays reward explicit arrangement: name every object and the surface. One diffused source keeps the shadows coherent across all three products instead of fighting each other.

Bold gym poster, a chalk-dusted barbell in dramatic side lighting on a dark concrete background, the headline reads 'NO DAYS OFF' in heavy bold sans-serif at the top, high contrast, gritty editorial style, sharp detail.

Seedream renders short text reliably — keep it to a handful of words, wrap it in quotes and ask for "bold sans-serif" rather than a font name. Long phrases are where garbling starts.

Product photo of a clear glass bottle of citrus soda with visible meniscus, refraction through the liquid, liquid clarity, condensation beads on the glass, backlit on a seamless #ffffff background, single soft backlight, sharp caustic highlights on the table, commercial beverage photography.

Transparent materials need their physics named: meniscus, refraction, clarity, caustics. Without those words glass reads as plastic. Backlighting is the classic setup that makes liquid glow.

A woman in a linen shirt pouring coffee in a sunlit kitchen, posed action mid-pour, golden hour light through a window camera-left, warm tones, editorial lifestyle photography, rule of thirds with the subject on the right, 50mm lens, shallow depth of field.

"Posed action" instead of "motion blur" is the trick that keeps hands and details intact while still reading as a candid moment. One window, one direction — the light stays believable.

Streetwear lookbook shot, model in an oversized rust-orange puffer jacket against a pale blue wall, centered composition, hard midday sun with crisp shadows, fashion editorial style, 35mm lens.

Seedream 5.0 Lite at 1K with four images per run is the exploration lane: same prompt, four takes, pick a direction. Rerun the winner on 4.5 at high resolution and steps.

Replace the product in Figure 1 with the product in Figure 2. Keep Figure 1's scene exactly: same table, same background, same lighting direction and shadow softness. Match the new product's scale and perspective to the original placement.

The Figure-grammar is the native language of Seedream Edit: the prompt is an instruction about the references, not a description of them. Pinning lighting and scale is what makes the swap look shot, not pasted.

Place the woman from Figure 1 in a rainy neon-lit street at night, side profile, holding a transparent umbrella, cinematic teal and magenta palette, single strong key light from the neon sign camera-right. Keep her exact face, hairstyle and jacket from Figure 1.

Side profiles hold identity better than wide shots on this family. The keep-clause names the three anchors — face, hair, jacket — so the model knows what is identity and what is scene.

Change the sneaker in Figure 1 from white to deep forest green with gum soles. Keep the exact same angle, laces, stitching pattern, background and lighting. Only the colorway changes.

Lock the seed from your hero shot, then iterate one variable per run — colorway, background, prop. With the seed fixed, the composition stays put and you get a true A/B set instead of ten different photos.

Dress the model in Figure 1 with the jacket from Figure 2 and the sunglasses from Figure 3, place her against the textured plaster wall from Figure 4, fashion editorial style, soft overcast light, three-quarter framing. Keep the model's face and pose from Figure 1.

Up to 10 references, each with one job, each called by number. The more precisely every Figure is assigned, the less the model improvises — unassigned references get averaged into the result.

Cinematic still of a barista sliding a cappuccino across a walnut counter, warm tungsten practicals in the background bokeh, steam catching the light, moody café atmosphere, anamorphic framing, film grain, sharp focus on the cup, 4K detail.

Seedream shares an aesthetic lineage with Seedance, which makes it the natural choice for thumbnails, keyframes and campaign stills that must sit next to Seedance footage — reuse the same palette and lighting words across both prompts.