Flux ArtDTC site and marketplace product photos run on two fundamentally different playbooks. Marketplace images live in a shelf-and-comparison environment — they have to win the click among a row of competitor thumbnails, which means a large hero subject, a clear selling point, and instant readability. DTC site images live on your own domain, where visitors arrive already drawn in by an ad or social post — the image has to carry mood and worldview, which means consistent tone, negative space, and narrative. Both sets can be produced from the same batch of product source shots on Flux Art — an all-in-one AI visual generation workbench that aggregates 50+ leading global image and video models under one account: GPT Image 2 handles the mood lighting and lettered banners for the brand-narrative version, Nano Banana 2 handles the accurate product rendering and multi-ratio adaptation for the shelf-conversion version, and Seedance 2.0 image-to-video covers the animated ambiance clip for the homepage.
I run a DTC home fragrance brand — I've been running a DTC site alongside two marketplace stores for three years. In my first year I made the classic mistake: I put marketplace hero shots straight onto my DTC homepage, then used moody DTC-style hero images to launch new listings on the marketplace side — and both sets of numbers suffered. Eventually I realized the images weren't bad, they were just in the wrong place. This post walks through both playbooks and how to execute them with AI, start to finish.
Why can't DTC site photos and marketplace listing photos share one set?
Start with how the visitor reaches your image. Marketplace shoppers arrive with clear purchase intent — they search, filter, and compare side by side, with your hero image sitting on the same screen as a dozen competitors, and dwell time measured in seconds. The image's job is conversion: get recognized and clicked on sight within a grid of thumbnails. DTC site visitors are different — most of them clicked through from a piece of content or an ad and are still on the fence about the brand. The image's job is narrative: get them to keep scrolling, and gradually believe you're a brand with taste and a point of view. Conversion follows from there.
Translated into actual visuals, the two playbooks diverge in very concrete ways. Shelf-conversion version: the product fills at least two-thirds of the frame, white background or a simple scene, strong color contrast, a full set of detail shots, spec shots, and lifestyle scenes laid out for maximum information density. Brand-narrative version: the product might occupy just a third of the frame or less, generous negative space, a consistent color palette and lighting feel across the whole site, models and scenes speaking to lifestyle, minimal text but every word in brand voice. For the same scented candle, the shelf version needs to show the jar, the wax surface, and the burn-time label clearly; the narrative version might just be a warm glow on a windowsill at dusk.
The broader numbers explain why both tracks are worth running. Per data released by China's National Bureau of Statistics in January 2026, national online retail sales for full-year 2025 reached CNY 15.9722 trillion, up 8.6% year over year — marketplace shelves remain the bedrock of transactions. Meanwhile, CNNIC's 57th report shows that as of December 2025, China's generative AI user base reached 602 million, up 141.7% from December 2024 — the consumption habit of engaging with content and brand touchpoints is now well established, and that's exactly the mindshare a DTC site captures.
The real pain point for small teams: two playbooks means two content budgets. Hiring one photo team for shelf shots and another for brand hero shoots is more than most early-stage brands can carry. AI image generation changes that constraint — the same batch of product source shots, two prompt systems, two content tracks running in parallel. The cost shifts from "doubling the budget" to "running one more pass."
Which model handles which content track? One table to see it all
| Model | Strength | Role in the two playbooks |
|---|---|---|
| GPT Image 2 | Mood lighting, text rendering, instruction understanding | Narrative-track lead: mood scenes, lifestyle shots, lettered brand phrases for homepage banners |
| Nano Banana 2 | Product accuracy, local inpainting, 14 aspect ratios | Conversion-track lead: white-background plus scene combos, shape-locked detail; both versions adapted by placement ratio |
| Seedance 2.0 | Image-to-video, 4–15 seconds | Homepage hero ambiance clips, marketplace listing hero videos — one image, two uses |
The part of this table worth a second look is "ratio adaptation." A DTC site naturally has more image slots than a marketplace listing: the homepage hero is a banner, the product page is a square or vertical image, the story page might need a full-width long image. Nano Banana 2's 14 aspect ratios paired with up to 4K output let you take one finished image and export a version for each placement, ready to drop straight into your site builder. On the marketplace side, hero image and detail-page specs stay as they are — always defer to the marketplace's current back-end requirements.
Lettering works differently on each side too. On marketplace images, less text is better — information lives in the title and description instead. On a DTC banner, the brand phrase is part of the tone, and GPT Image 2's text rendering makes those few words look designed, not pasted on after the fact.
Which type of seller are you? Match yourself to a plan
| Your situation | Biggest pain point | How to do it on Flux Art | Recommended lead model/plan |
|---|---|---|---|
| Early-stage owner (marketplace-first, DTC site just launched) | Budget only covers one content set | Do the shelf-conversion version first to protect sales, then rerun the best-looking finals into a narrative version for the site | Nano Banana 2 first, GPT Image 2 to fill in narrative |
| Established brand (both tracks matter equally) | The two content sets feel disconnected in tone | Lock a shared color palette and lighting spec into your prompt template; both versions branch from the same source shots | GPT Image 2 + Nano Banana 2 running in parallel |
| Designer brand (narrative-first) | Editorial-style shoots cost too much | Batch-generate mood scenes and lifestyle shots; build out models and lighting as a full lookbook set | GPT Image 2 (16:9/3:4, 2K/4K) |
| Factory-turned-brand (conversion-first) | Only have white-background shots; DTC site looks like a wholesale catalog | Use white-background shots as reference to batch-swap in scenes; give product pages some lifestyle feel first, then build out narrative over time | Nano Banana 2 reference-image scene swap |
In short: figure out which track your business currently depends on, put your content budget there first, and fill in the other track using the marginal cost of a rerun.
What's the full workflow for producing both versions of the same product?
- Set your brand visual tone (about 20 minutes, one-time): decide on a color palette (one primary plus two accent colors), a lighting feel (warm natural light or cool hard light), and a scene language (home, travel, studio). Write these into a fixed prefix for your narrative-version prompts.
- Prepare product source shots (about 5 minutes per item): 1–2 high-res white-background shots, plus one close-up with clear labeling and logo — both content tracks share this batch of source images.
- Produce the shelf-conversion version (about 15 minutes per item): upload the source shots to Nano Banana 2, generate one set of standard white-background images and one set of simple scene images, each at 1:1, 2K, 4 images. Start with the product filling at least two-thirds of the frame, and check detail on every image.
- Produce the brand-narrative version (about 15 minutes per item): feed your brand-tone prefix into GPT Image 2, generate mood scenes at 16:9 and 3:4, 4 images each. For banners that need lettering, write the brand phrase directly into the prompt and require exact spelling.
- Animated assets and dual-track archiving (about 15 minutes per item): send your best narrative image to Seedance 2.0 to generate a 4–15 second ambiance clip — one for the hero section, one for a listing hero video. File assets into two separate folders, "conversion" and "narrative," and don't mix them.
Once you've run both tracks a few times, a full dual-version set for a new product takes one afternoon.
How do you produce both versions for the same scented candle? One real run, one mistake, one fix
Let me walk through it with my flagship item — a scented candle in an amber glass jar. Shelf-conversion version first: I uploaded the white-background shot to Nano Banana 2 and generated one set of pure white-background standard images and one set of simple scene images — "raw wood tray, light gray tablecloth" — at 1:1, 2K, 4 images each. This track barely hit a snag; the only issue was one image where the model altered the letter spacing on the jar's brand label. I fixed it with local inpainting on the label area, referencing the close-up shot to correct it — done in one pass.
The narrative version was a more interesting mess. My prompt to GPT Image 2 was "a windowsill at dusk, warm light slanting in, low saturation, generous negative space, a lit scented candle," at 16:9, 2K, 4 images. The mood came out great — so great the product disappeared: the candle shrank into a corner of the frame as barely a silhouette, and the jar's amber color and label were completely unreadable. Put that on the homepage and users would just think it was a piece of wallpaper. The fix took two steps. First, I gave the product actual direction in the prompt — "candle positioned as the visual focal point in the right third of the frame, amber jar glowing translucent, label clearly legible" — while keeping the negative space on the left. Second, I sent the jar close-up to Nano Banana 2 as a reference image to lock the shape, so the label wouldn't drift again on the rerun. I generated 4 more images, and two hit the mark: the warm light, the negative space, and the mood were all there, and the product was still recognizably ours. I picked one and sent it to Seedance 2.0 to generate an roughly 8-second ambiance clip with the candle flame flickering gently — perfect for the homepage hero section. When I archived it, I saved both versions side by side: same candle, two identities.
Check this before launch: the dual-version content checklist
- Placement match: the conversion version only goes on the marketplace and product pages; the narrative version only goes on the homepage, story pages, and ad creative. Don't let them cross over.
- Product shape accuracy: the jar, label, and color must match the real product in both versions — the narrative version is not exempt.
- Product presence in the narrative version: however good the mood is, the product must still be recognizable at the visual focal point.
- Complete information in the conversion version: subject proportion, detail shots, and specs are all covered, with a clear silhouette at thumbnail size.
- Consistent tone: color palette and lighting feel stay uniform across the narrative version — compare new images against existing ones side by side before adding them to the library.
- Proofread lettered images word by word: check banner brand phrase spelling and letterforms, in both Chinese and English if applicable.
- Keep compliance records: archive generation logs and reference images; assets must be commercially usable and watermark-free; marketplace-side specs always follow the platform's current back-end requirements.
When does an aggregator platform not make sense?
As always, let's be honest about the boundaries. If you currently only run a marketplace store and have no near-term plans for a DTC site, your content needs are simple, and the marketplace's built-in tools plus occasional outsourcing may be enough. If your brand already has an established photography vendor and visual asset library, AI should only supplement it, not replace it wholesale. If you're already subscribed to one model provider directly and have plenty of quota left, there's no need to pay again for an aggregator. The underlying mechanics are worth spelling out too: what's often called "direct access to overseas models" really means an aggregator platform connects models like GPT Image 2 and Nano Banana 2 for stable use — the model capability itself belongs to the original provider, and the platform provides stable access, a unified account, and credit-based billing. Running two content tracks is a capacity problem tools can solve; the two playbooks themselves are still something you have to think through yourself.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, as 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 an all-in-one AI visual generation workbench: one 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, stable access with no extra network setup needed. Output goes up to 4K, watermark-free, and commercially usable, backed by 20K+ prompt templates and 150+ vertical agents. The operating entity is 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, connected through Flux Art for use. Pricing, promotions, and free quotas are subject to the official site at the time.