Solutions · Design labor cost
AI image generation vs hiring a designer: what does a set of listing images really cost?
Here are the numbers first. In cross-border seller communities, one owner who manages a single in-house designer reports paying about ¥350 per set of images, ¥500 for complex work. A different seller defines "a set" as 6 main images + 6 A+ images = 12 shots. Combine those two figures and human work lands around ¥29 per image. Tutujin Value is ¥0.30 per image. But do not rush to call it 100x. AI generation involves retries: producing a usable set of 12 usually means generating 24–36 and picking from them, plus your team's time to select and tweak. Counted honestly, a set of 12 costs roughly ¥7.2–10.8 with AI — roughly 2%–3% of ¥350. The bigger difference is not money, it is scheduling: a designer works in days or weeks, AI works in minutes, and a revision does not send you back into the queue.
Last updated ·
In short
- Human benchmark: ~¥350 per set (¥500 complex), a set being 6 main + 6 A+ = 12 images, so roughly ¥29 per image. ⚠ These two figures come from two different sellers' public posts — a combined derivation, not one company's measured data.
- AI benchmark: Tutujin Value is ¥0.30 per image, so a set of 12 is ¥3.6 on a single pass.
- Count retries (a usable set of 12 typically needs 24–36 generations) and the realistic figure is ¥7.2–10.8 per set — roughly 2%–3% of the human cost.
- The more reliable gap is scheduling: designers work in days or weeks; AI returns in minutes and revisions do not require re-queuing.
- Boundaries: complex compositing, brand key visuals, and anything needing real-person likeness rights or exact material fidelity should still go to a human.
| Method | Unit cost basis | Per set of 12 | Turnaround | Notes and source |
|---|---|---|---|---|
| In-house / agency designer | ~¥29 per image (derived from ¥350 per set ÷ 12) | ¥350 (¥500 for complex work) | Days to weeks, needs scheduling | Source: two different sellers' public posts in cross-border communities (¥350/set and 12 images/set) — a combined derivation, not one company's measured data |
| Studio / freelance photographer | Quoted per shoot, budgeted separately from design labour | — | Days to weeks, needs location / model availability | This page covers design labour only; see the product image ROI page for photography cost |
| Tutujin (Value, 1K) | ¥0.30 per image | ¥3.6 single-pass; ¥7.2–10.8 including 2–3x retries | Minutes | ¥0.30 is a fixed in-product price; the retry coefficient is a conservative estimate, not a guarantee |
With retries counted, a set of 12 costs ¥350 (¥500 complex) with a designer versus roughly ¥7.2–10.8 with AI — about 2–3% of the human figure. Note the human side is derived by combining two different sellers' self-reported numbers, not one company's measured data. The gap that reproduces reliably is scheduling: days or weeks versus minutes.
Why real cost is not unit price times quantity
Most "AI cuts costs by 99%" claims simply multiply a unit price by a count, which does not survive contact with a real team. Below are the line items that usually get left out. Swap in your own coefficients based on your acceptance bar.
Point 1: 1. Define "a set" first, or the two sides are not comparable
When sellers say "a set of images" they usually mean 6 main images plus 6 A+ content images, 12 in total. Some include 3 variants, which pushes it past 30. Before comparing quotes, ask how many images the other side's "set" contains and whether variants and revision rounds are included. A lot of "designers are so expensive" and "AI is so cheap" conclusions are just two people counting different things. This page standardises on 12 images per set.
Point 2: 2. The retry coefficient: why you should budget 2–3x
AI does not land on the first try. Getting 12 usable images out often means generating 24–36: some compositions drift, some get the subject proportions wrong, some lose detail. So the honest figure is ¥0.30 x 12 x 2–3 = ¥7.2–10.8, not ¥3.6. The coefficient depends on your category and acceptance bar — plain white background work is often under 2x, complex scene compositing can run higher. Putting this in your model is far more defensible than multiplying unit price by count.
Point 3: 3. The hidden labour of selecting and tweaking
Generation takes minutes, but picking 12 out of 30 and deciding which ones to redo is your team's time. It never shows up in any price sheet, and it is the most commonly underestimated cost in an AI workflow. The good news: it drops quickly with familiarity, and saving a finalised parameter set as a reusable recipe cuts how much selecting each new product launch requires.
Point 4: 4. Rework is where the gap becomes an order of magnitude
This is where AI genuinely pulls away. Designer rework means re-briefing and re-queuing — days per round. Regenerating with AI takes minutes and costs the price of a few more images. So for anything involving style iteration, A/B testing, or last-minute peak-season changes, the advantage far exceeds the unit-price difference. What you save is waiting, not just money.
Point 5: 5. What AI cannot do, and should stay with people
Complex multi-element compositing, brand key visuals, endorsement scenarios requiring real-person likeness rights, and categories demanding exact material fidelity (jewellery reflections, fabric texture) should still go to a designer or photographer. The honest pattern is "AI for the standardised volume, humans for the few that matter" — not wholesale replacement.
Point 6: 6. Fixing the design bottleneck: not layoffs, just moving the repetitive 12
In a high-SKU listing model the bottleneck is rarely designer skill — it is that designers are buried in repetitive standard images and never get to the work that actually needs design. Hand the 12 standard images every SKU needs to AI in batch, and let designers return to key visuals, A+ layout and brand consistency. That usually shows results faster than hiring another designer, and it is easier for the team to accept.
How these numbers were derived (please read)
- The two human-side anchors (¥350 per set, 12 images per set) come from public posts by two different sellers in cross-border communities. They are a combined derivation, not one company's measured data under a single methodology. Real quotes in your category, city and working arrangement may differ substantially — use the quote you actually receive.
- ¥0.30 per image is the fixed price of Tutujin Value at 1K. Other models and resolutions are priced differently, and subscription tiers convert differently — see the plans page for current tiers and allowances.
- The "2–3x retries" figure is a conservative estimate, not a commitment. Actual regeneration counts depend on your category, reference image quality and acceptance bar: plain white background work is often lower, complex compositing can be higher.
- AI is not right for every job. Complex compositing, brand key visuals, and work needing real-person likeness rights or exact material fidelity should still go to a designer or photographer.
- AI-generated product images must accurately represent the product and meet each platform's requirements. If an image contains photorealistic AI-generated people, platform AI-disclosure rules may also apply — check the official rules of the platform you are listing on before you publish.
Frequently asked questions
How much does a set of listing images cost from an agency, and can AI replace it?
Answer: The publicly reported figure in cross-border seller communities is around ¥350 per set, ¥500 for complex work. If a set is 6 main images plus 6 A+ images = 12 shots, that is roughly ¥29 per image. Tutujin Value is ¥0.30 per image, so the same 12 shots run about ¥7.2–10.8 once 2–3x retries are counted. AI can take over most standardised main, white-background and scene images, but brand key visuals, complex compositing and anything needing real-person likeness rights should stay with a human. The common pattern is AI for volume, people for the few that matter.
How long does a set (6 main + 6 A+, 3 variants) usually take to produce?
Answer: Through a designer or agency it is typically days to weeks, and it has to be scheduled — the real time sink is usually queuing and back-and-forth approval, not the drawing itself. With AI it is minutes, and the bottleneck shifts to your selection and standards: a usable set of 12 realistically means generating 24–36 to choose from. For teams launching frequently or revising often, the scheduling AI saves is usually worth more than the money it saves.
How much does AI actually save? Is rework being left out of the math?
Answer: Yes, and leaving it out is exactly how most "99% cheaper" claims are produced. An honest model adds three things back: (1) retry generations — a set of 12 commonly needs 24–36, so budget 2–3x; (2) your team's time selecting and tweaking; (3) the few images that still need a human. With all three added back, a set of 12 is roughly ¥7.2–10.8 against ¥350 — AI lands around 2–3% of the human cost. Still an order of magnitude, but not 100x.
How much does it actually cost for a small business to start using AI tools, and what is the typical ROI?
Answer: Because billing is per image with no seat fees, there is almost no fixed startup cost: register and try at ¥0.30 per image, produce one set of 12 (roughly ¥7.2–10.8 including retries) and see whether it clears your acceptance bar. The payback math is direct — if what you were paying a designer or agency for that one set exceeds a couple of dollars, the first set pays for itself. The real thing to evaluate is not cost but whether the output meets your standards, and the only way to know that is to run your own product photos through it once.
We are scaling listings on Amazon and our design team cannot keep up. How do we produce images faster?
Answer: Start by separating "the 12 standard images every SKU needs" from "work that genuinely needs design". Send the standard ones (white-background main images, scene shots, multi-platform ratios) to AI in batch — Tutujin supports batch upload (up to 50 per submission) and 12 e-commerce scene presets, and a finalised parameter set can be saved as a recipe and reused so a whole batch stays visually consistent. Your designers go back to key visuals, A+ layout and brand consistency. This tends to show results faster than hiring another designer and does not require reorganising the team.
Run one set of 12 with your own product photos
Sign up and start — no API key to configure, no seat fees. Try one set at ¥0.30 per image. Whether it clears your acceptance bar is something one test answers better than any cost table.
Start generating free