AI product photography in 2026 — hero shots vs your real product

In 2026 the first question of an AI product shoot is not which model is best. It is whether the product in the frame has to be your product. If it can be invented, or is generic enough that nobody will check (a cup of coffee, a plain white sneaker), the text-to-image flagships on eroq now produce catalogue-grade hero shots from a sentence. If it has to be the exact bottle you ship, with its shape, cap and label, you need an engine that reads your photos. This guide splits the shoot along that line, with prompts and a priced launch set.

Two jobs, six engines

Job	Engine	Per take	Why this one
Concept or category hero	Seedream 5.0 Pro	36 credits / image	Materials and texture that survive a crop
Catalogue frame with label words	GPT Image 2.5 Sunburst	31 credits / image	Clean composition, the sharpest in-image text
The final hero that must be right	Nano Banana Pro	124 credits / image	Highest fidelity, accurate text
Lifestyle scene in daylight	Flux 3 Image	32 credits / image	Natural light, reliable typography
Your actual product, from photos	Eroq One	20 credits / image	Up to ten references, edits by instruction
Your product, fast and cheap	Krea2	10 credits / image	Up to four references, in-place edits

The first four are text-to-image only on eroq: drop a photo in and the studio tells you the engine reads no pictures. The last two are eroq's own reference engines.

When the product can be invented

This covers more work than it sounds: pre-launch concepts, pitch decks, packaging directions, category banners, ads where the product is a type rather than a SKU, and moodboards that brief a real photographer. Write the prompt from the object outwards:

1. The object and its finish. Hand-blown green glass, brushed aluminium, matte ceramic with a bamboo lid.
2. The surface. Travertine, walnut, wet slate, a seamless sage backdrop.
3. One light source. Raking morning sun, a soft top light, a single rim light.
4. A lens. 85mm for a hero, 100mm for a detail, 35mm for lifestyle.
5. Label words, in quotes, if the pack should carry any. Short.

Pick the engine by the job in the table. Seedream when the material is the story, Sunburst when the frame is a clean catalogue layout with words on the pack, Flux 3 when the scene is lived-in daylight, Nano Banana Pro for the one hero image that leads the page. For lens and light vocabulary, photoreal AI image prompts is the reference, and Seedream 5.0 Pro vs Nano Banana Pro settles the choice between the two detail flagships.

When it has to be your product

Eroq One takes up to ten reference pictures and builds the shot from all of them, which is how the shape, the cap and the label stay the same from one image to the next.

• Shoot good references. Front, three-quarter, a close-up of the label, the top. Sharp, evenly lit, on a plain background.
• Say what each picture is for. "The bottle from Picture 1, the label as in Picture 3." The studio numbers the pictures in its stack for you.
• Build the set first, if you like. Render the scene on a flagship (the travertine ledge, the morning light), then hand Eroq One both the set and the product photos: "place the bottle from Picture 1 on the ledge in Picture 2, match the light".
• Make variants by instruction. In Edit mode, write the change, "make the cap matte black" or "change the background to a beach at sunset", and the picture keeps its own frame.
• Go cheaper on Krea2 when four references are enough. Its in-place edits re-render the scene, so small details can drift; for exact instruction edits, stay on Eroq One.

Unreleased products have a privacy angle too. eroq runs Eroq One and Krea2 itself and never trains on your prompts, and with Private mode on the render stays in your browser. The text-to-image flagships run on outside providers, which may retain the prompt but never use it for training. Moderated vs private image models goes deeper.

Formats by channel

• 1:1 for marketplace grids and social feeds.
• 3:4 for tall product pages and pins.
• 4:3 for editorial and lookbook spreads.
• 16:9 for site banners and hero headers.
• 9:16 for stories and reel covers.

Every format costs the same. For listings that demand a plain white background, the image studio's Remove background mode cuts the product out of any render.

What a launch set costs

A realistic set for one product: four concept takes on Seedream to find the look, one Nano Banana Pro hero for the top of the page, and two batches of four Eroq One placements of the real product. That is 428 credits, about $4.28 at the entry pack. Failed renders refund themselves, and a prompt a provider refuses costs nothing.

Example prompts

1. Concept hero, Seedream, 4:3. "A hand-blown green glass carafe on a travertine ledge, raking morning light, caustic reflections dancing on the stone, beads of water on the glass, 100mm, quiet luxury catalogue photograph." Caustics and water beads are the material cues that make glass read as glass.
2. Catalogue frame, Sunburst, 1:1. "A clean catalogue photograph of a white ceramic skincare jar with a bamboo lid on a pale sage backdrop, soft top light, the label text "DAILY CALM" in small dark grey sans-serif, generous empty space on the right." The empty space on the right is where the price and the copy will go.
3. Final hero, Nano Banana Pro, 16:9. "Premium over-ear headphones in brushed aluminium and tan leather on a dark walnut desk, a single soft rim light tracing the earcups, fine stitching visible, deep shadows, 85mm, high-end product photograph." Stitching and brushed metal in shadow are the details that justify the top engine.
4. Lifestyle, Flux 3, 3:4. "A canvas tote bag hanging from a café chair by a sunny window, printed with "MARKET DAYS" in black block capitals, a coffee cup on the table, warm natural light, 35mm, relaxed lifestyle photograph." Daylight and a real setting sell use; the print is short enough to stay legible.
5. Your product, Eroq One, 1:1. "Place the perfume bottle from Picture 1 on the wet black stone in Picture 2, keep the bottle's shape, cap and label exactly as in Picture 1, a soft reflection on the stone, low-key light from the left." Each picture gets a job, and the instruction says what must not change.

FAQ

Can AI product photography use my actual product?

Yes, on Eroq One: attach up to ten photos of the product and say what each one is for, and it builds the shot around them. The text-to-image flagships on eroq cannot read pictures, so they suit concepts and generic products.

Which AI model is best for product photos?

It depends on the job. Seedream 5.0 Pro for texture, GPT Image 2.5 Sunburst for clean catalogue frames with label words, Nano Banana Pro for the final hero, Flux 3 Image for daylight lifestyle, and Eroq One for your real SKU.

Will the label text come out right?

Short label words in quotes render well on Sunburst, Nano Banana Pro and Flux 3. Your real label comes through Eroq One's references; for pixel-exact artwork, composite it in your editor. Legible text in AI images has the method.

How much does an AI product shoot cost?

At the entry pack, a take runs from about $0.10 on Krea2 and $0.20 on Eroq One to $1.24 on Nano Banana Pro, at any format. The launch set above comes to 428 credits.

Photograph the product once, then open the image studio.