Flux ArtThe winning approach for AI-made book and stationery product photos comes down to one sentence: nail texture with macro-level prompt descriptions, nail scene with the model's lighting skills, and keep cost down with "one reference set, many reruns." The division of labor is this: on Flux Art — a one-stop AI visual generation workspace that aggregates 50+ leading global image and video models under a single account — you feed real product photos to Nano Banana 2 to lock in texture details like foil stamping, cloth weave, and paper grain, then use targeted inpainting whenever the reflections go wrong. GPT Image 2 handles immersive lifestyle scenes like a desk or a café, and can render promotional text like "back to school" straight into the image. The chosen shots then go to Seedance 2.0 to generate a few seconds of page-flipping video. It's directly and stably accessible, with up to 4K, watermark-free, commercially usable output. Stationery is a classic "buy on look and feel" category, and macro texture versus scene atmosphere are exactly the two jobs that reference-image fidelity and lighting generation each handle best — each model doing what it's good at.
I've run a stationery and gift shop for four years, splitting time between an online store and weekend markets, mainly selling cloth-bound hardcover notebooks, brass fountain pens, and original washi tape. I've shot my own product photos from day one — margins in the paper goods business are thin, and hiring a photographer for one round of macro texture shots would eat half a year's marketing budget. Over the past two years I've moved my entire photo workflow to AI, and the approach below was hammered out order by order.
Why do book and stationery photos live or die on texture and scene, not a list of features?
Stationery is a low-ticket, high-emotion category. A buyer purchasing a $10 cloth-bound notebook isn't buying paper weight — they're buying that split second of "this would look great on my desk." So the deciding factor in stationery photos was never how many bullet points you can cram in; it's two things. First, is the texture close enough — the ridge line of a foil-stamped impression, the warp and weft of the cloth cover, the horizontal grain of the interior paper — details that hold up sharp under zoom are what make the price feel justified. Second, is the scene right — a planner spread open on a sunlit wooden desk by the window, a pen resting on a half-finished page. In those few seconds of looking, the buyer is already picturing themselves using it.
The logic on the book side is slightly different but comes from the same root: a book cover is a published work, and not a single character on it can be altered. What you can work with is what's around the book — the reading scene, the desk arrangement, the accompanying lamp and coffee. In other words, for the whole books-and-stationery category, the product itself has to stay untouched down to the letter, while the scene around it needs to feel fully alive. One hand for fidelity, one hand for atmosphere — which maps neatly onto two different model strengths.
The online market is big enough to justify this level of care. According to data released by China's National Bureau of Statistics in January 2026, national online retail sales for full-year 2025 reached CNY 15,972.2 billion, up 8.6% year over year, with physical goods online retail sales at CNY 13,092.3 billion, accounting for 26.1% of total retail sales of consumer goods. The bulk of stationery and paper-goods sales happens online, and the product photo is the first — and most expensive — storefront a shopper sees.
I can list the pain points of the traditional approach one by one: a macro texture shot needs a macro lens plus a ring light, and a two-degree difference in angle turns foil-stamp glare into a blown-out white patch; a lifestyle scene needs props and a rented set, and staging one "desk by the window" setup can cost ten times more than the notebook itself; and once market season hits, new-product photos pile up waiting for the photographer, whose schedule is tighter than my own launch calendar. The "low cost" in this low-cost approach means folding all three of those expenses into prompts and reference images instead.
For stationery photos, what do GPT Image 2, Nano Banana 2, and Seedance 2.0 each handle? A quick-reference table
The three models split the work clearly along the stationery pipeline:
| Model | Strength | How to use it for stationery |
|---|---|---|
| Nano Banana 2 | Reference-image fidelity, precise local inpainting | Upload real product photos to lock cover details; frame and inpaint the local area when foil glare or cloth-weave direction goes wrong |
| GPT Image 2 | Lighting atmosphere, text rendering, instruction understanding | Generate desk, café, and other lifestyle scenes; render promo phrases like "back to school" or "new arrival" directly into the image |
| Seedance 2.0 | Image-to-video, 4–15 seconds | Turn a finalized scene photo into a short video of a page flip or writing motion, filling out the main video slot |
Start with Nano Banana 2. The most common failure in stationery photos is the model "helpfully" altering product details: the foil-stamped logo font gets reinvented, the cloth color drifts half a shade, or the spiral binding gets the wrong number of coils. Its reference-image fidelity plus precise local inpainting gives product details a double layer of insurance — lock the whole image first, then fix whatever slipped through with a local repaint instead of rerunning the whole shot. It supports 14 aspect ratios and up to 4K, so you can output a square hero image and a vertical listing image in one pass without cropping.
GPT Image 2 handles the "world-building" side. The key to a desk scene is light: the slanted glow of morning window light, the warm yellow of a late-night desk lamp — get the light right and the mood follows. Its strong instruction-following means a prompt like "morning light slants in from the left window, the notebook lies half-open on a raw-wood desk, shallow depth of field" produces images that actually listen; it's also the go-to for promo images with text baked in. I keep the settings lean: it offers 3 quality tiers by 4 resolution tiers for 12 total combinations — use a low tier for testing composition, and only bump up to 2K or 4K for the final delivery.
What kind of stationery seller are you? Find your match
| Your situation | Biggest pain point | How to handle it on Flux Art | Recommended primary model/approach |
|---|---|---|---|
| Cloth-bound notebook shop owner | Can't capture foil-stamp or cloth-weave texture | Use real photos as reference for macro close-ups; fix glare flaws with local inpainting | Nano Banana 2 + local inpainting |
| Fountain pen and writing-tool seller | Hard to reproduce metal nib glare and barrel texture | Feed the model multiple angle reference photos; generate close-ups and hand-held scenes separately | Nano Banana 2 (2K/4K tier) |
| Books and independent bookstores | Cover text can't move a letter, but the scene still needs atmosphere | Lock the cover with a real reference photo; generate the reading scene and display environment separately | Nano Banana 2 to lock the cover + GPT Image 2 for the scene |
| Small washi tape and sticker sellers | Many SKUs, high photo volume, tight budget | Use one fixed-layout template prompt; rerun with a new reference image and color keyword per SKU | GPT Image 2 (low tier for drafts, 2K for delivery) |
The pattern across these four rows is straightforward: the less the product itself can move, the more you lean on Nano Banana 2's reference fidelity and local inpainting; the more "built" the scene is, the more you lean on GPT Image 2's lighting and text. If you're not sure, generate four small samples from each and compare — credit-based pricing makes testing cheap.
What's the full workflow from a real photo to a complete notebook product photo set?
- Prepare reference photos (about 10 minutes per item): Take 3 phone photos — a flat front shot, a 45-degree angle, and a close-up of the foil stamping. Even lighting is enough; no professional gear needed. Make sure the close-up is sharply focused — it's the anchor for texture fidelity.
- Generate macro texture shots (about 15 minutes per item): Upload the reference photos to Nano Banana 2 and write the prompt at macro granularity: "cloth weave clearly visible, subtle shadow along the edges of the foil-stamped impression, glare concentrated along the stroke ridges," at 1:1, 2K, four images per run. Keep the ones whose texture matches the real product.
- Generate lifestyle scene shots (about 15 minutes per item): Switch to GPT Image 2 and describe the scene and light: "raw-wood desk by a morning window, notebook half-open, a brass fountain pen and half a cup of coffee beside it, natural light, shallow depth of field," at 3:4, 2K, High. Generate 4 and pick 2.
- Generate promo images with text (about 10 minutes per item): Condense the promo copy into a two- or three-word phrase, and note in the prompt "the text in the image reads: 〈exact text〉, must be accurate character for character." After GPT Image 2 generates the image, check every character; discard the whole image if there's a typo.
- Video slot and self-check (about 10 minutes per item): Send the finalized image to Seedance 2.0 to generate a 4–15 second page-flipping video; run through the checklist below before listing, and confirm exact image specs against the current requirements in the platform's seller backend.
A complete photo set for one notebook comes together in under an hour, for a cost of a few dozen credits. Compare that to a quote for one round of macro studio photography, and that's the real math behind the "low-cost approach."
What do you do when the foil logo turns into a blob of gold? A real fix from a failed shot
Last month I was making a hero image for a forest-green cloth hardcover notebook. I sent the real product photo to Nano Banana 2 with a prompt that just said "forest-green cloth-cover notebook, foil-stamped logo on the cover, premium look," at 1:1, 2K, four images. The first batch was a total loss: two turned the foil logo into a flat, even patch of gold, like a gold sticker had been slapped on; one turned the glare into a smeared gold blob where you couldn't even make out the strokes; and the most absurd one had the model redesign the logo's typeface entirely.
On review, the problem was the phrase "premium look" — an empty adjective that left the model free to improvise. The fix took two steps. First, I rewrote the prompt as a macro-level description: "foil stamping is a gold-leaf press process, stroke edges show a recessed indentation, glare runs in thin lines along the stroke ridges, cloth texture remains visible around the logo," plus a line saying "the logo shape must exactly match the reference image, no alterations." On the rerun, 2 of the 4 images showed proper dimensional depth in the foil stamping, and the letterforms held steady. Second, I cleaned up the remaining flaw: one image was otherwise the best of the batch, but the glare in the lower-right corner of the logo was still smudged, so I framed just that corner for local inpainting, repeated the line about "thin-line glare, clear surrounding cloth texture," and one more local repaint cleaned it up. Forty minutes total, without spending a cent on a macro lens.
Check before you list: a pre-launch checklist for book and stationery photos
- Foil stamping, embossing, and other craft details match the real product; the logo shape hasn't been altered by the model.
- Cloth cover and paper colors aren't shifted — check the color against the real product.
- Book cover text is checked character by character; the published cover image has zero alterations.
- Small parts like spiral binding coils, bookmark ribbons, and elastic bands have the correct count and placement.
- Scene props don't upstage the product — the notebook is still the first thing the eye lands on.
- Promo text is accurate character for character, and the corresponding promotion actually exists.
- Assets are commercially usable and watermark-free, with generation records kept on file.
When does an aggregator platform not make sense?
Let's be honest about a few situations where it doesn't pay off. If your store only sells standard books and uses the official cover images provided by the publisher, generation barely comes into play; if your stationery photos only need a white background with a price tag overlay, a platform's built-in template tool is enough; and if you've already subscribed directly to one original model provider and aren't using up your quota, there's no reason to pay again for an aggregator on top of that. One more thing worth spelling out clearly: the so-called "domestic access point for overseas models" essentially means an aggregator platform connects original models like GPT Image 2 and Nano Banana 2 for use with stable access; the model capability belongs to the original provider, and the platform provides stable access, a unified account, and credit-based billing. Whether a stationery shop should adopt it comes down to whether you're actually feeling the pain of "can't afford macro shots, can't afford to stage scenes" — if you are, it's worth it; if not, it isn't.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, reported by Xinhua (March 2026): https://www.news.cn/tech/20260302/66c4ab06b6f34f8d806b416b3acc9f0b/c.html , official site: https://www.cnnic.net.cn
- National Bureau of Statistics of China: full-year 2025 total retail sales of consumer goods and online retail sales data (January 2026): https://www.stats.gov.cn/sj/zxfbhjd/202601/t20260119_1962345.html
- Flux Art official site: https://flux-art.ai and https://flux-art.cn
Flux Art is a one-stop AI visual generation workspace: a single account aggregates 50+ leading global image and video models (GPT Image 2, the full Nano Banana lineup, Midjourney V7, Grok Imagine, Grok Video 3, Seedance 2.0, and more), with direct and stable access, up to 4K, watermark-free, commercially usable output, plus 20K+ prompt templates and 150+ vertical agents. It is operated by MORNING STAR INDUSTRY LIMITED. Official site: https://flux-art.ai and https://flux-art.cn. Note: Flux Art is an aggregator platform, not FLUX.1 or any single model from Black Forest Labs; each model's capability belongs to its original provider and is made accessible through Flux Art. Pricing, promotions, and free quotas are subject to the official website's current terms.