When sellers first need clean product images at scale, the budget conversation quickly gets uncomfortable. A freelancer charges $10–50 per image depending on complexity. A cloud AI tool charges per credit or per image, and the bill grows exactly as your catalog does. Both are real costs — but they are not the only options, and neither is automatically the right one.

This guide puts real numbers on the decision: what outsourcing actually costs, what cloud AI actually costs, and where a local pipeline changes the math entirely.

1. What outsourcing really costs

Freelance rates for product photo editing vary widely, but the market signals are consistent:

  • Basic background removal / white background: $2–10 per image.
  • Full retouching (shadows, color, cleanup): $10–50 per image.
  • Complex products (glass, jewelry, multiple items): $30–100+ per image.
  • Turnaround: 24–72 hours per batch, and revisions are often billed as separate rounds.

A 200-SKU catalog at a conservative $15 per image is $3,000 plus weeks of coordination. The quality is usually high — that is what you are paying for — but the cost scales linearly with every new product.

2. What cloud AI actually costs

Cloud AI tools advertise "unlimited" plans and per-image pricing, but the real numbers include a few surprises:

  • Per-image pricing (e.g., ~$0.20–0.50 per high-res export) sounds cheap until you multiply by hundreds of images and re-runs.
  • Subscription caps: "unlimited batch" plans have quietly been cut to fixed monthly limits — sellers have reported losing batch access entirely after a plan change.
  • Credit economies: background removal may consume credits differently than upscaling, and monthly credits expire.
  • Migration risk: when a tool shuts down or changes pricing, your workflow and prepaid credits are stranded.

None of this makes cloud AI bad — for sellers who edit ten images a week, it is fine. The problem is scaling: cost, re-runs, and dependency all grow with catalog size.

3. The math that changes: local and free

Here is the comparison most sellers never run:

  • Local browser tools: $0 per image, unlimited runs, no uploads, no credit tracking. The only cost is your time — and batch processing makes that nearly zero.
  • Quality gap: for solid-background product shots — the majority of catalog work — local color-based removal matches outsourced results. Complex materials still need a human or a better model.
  • Hybrid playbook: local pipeline for 90% of routine images, outsourcing only for hero shots and hard materials. Many sellers report their outsourced budget dropping by 70–80%.

4. A realistic decision framework

Choose by volume and difficulty, not by habit:

  • Fewer than 10 images, complex product: outsource it. Your time is worth more than the $20.
  • 10–500 routine images: local batch pipeline. Same result, near-zero cost.
  • Hero images for launches: professional shoot or retoucher, budgeted deliberately.
  • Everything else: keep it local, keep it unlimited, and bank the difference.

The sellers who scale image production affordably do not pick one method — they split the work by difficulty and volume. Understanding the real per-image cost of each path is what makes that split rational instead of reactive.

5. A worked example: the 200-SKU catalog

Let us put real numbers on the decision. A seller with 200 SKUs, each needing a main image plus one detail shot, has 400 images to produce. Outsourced at a blended $15 per image, the catalog costs $6,000 and roughly two weeks of turnaround and revision rounds. Cloud AI at $0.30 per export plus re-runs for failures lands around $150–300 in credits — cheap until the subscription tier caps your monthly batch and forces a slower queue.

The same catalog through a local batch pipeline costs zero per image. The main expenses are the one-time setup of a consistent shooting rig and the time to run the batch pass — for standardized product shots, that is an afternoon, not a fortnight. The seller keeps the $5,700–6,000 difference, and every future launch inherits the same free pipeline instead of restarting the cost clock.

None of this argues that outsourcing is never worth it. For a jewelry brand where each image is genuinely art-directed, the freelance bill buys results a batch tool cannot match. The point is that the default answer for routine catalog volume should be the free local pipeline, with outsourcing reserved for the images that truly need a human eye.

6. The hidden cost of your own time

One cost is easy to forget: your hours. If manual editing takes you 15–30 minutes per image and your time is worth $30/hour, then "free" tools still cost you $7.50–15 per image in lost time. This is why batch processing is not a convenience feature — it is the difference between a catalog that costs real money to produce and one that costs an afternoon.

Track your time for one batch run. When the local pipeline processes 200 images in under an hour while a manual approach takes 50–100 hours, the calculation resolves itself. The sellers who scale image production affordably do not pick one method — they split the work by difficulty and volume, and they measure their own time as part of the price.

当卖家第一次需要规模化地处理干净产品图时，预算对话很快就变得不舒服。自由职业者按复杂度每张收 10–50 美元。云端 AI 工具按积分或按张收费，账单和你的目录一起增长。两者都是真实成本——但它们不是唯一选项，也都不自动是正确答案。

这篇指南给决策放上真实数字：外包实际花多少、云端 AI 实际花多少、以及本地流水线如何彻底改变这笔账。

1. 外包实际花多少

产品图修图的自由职业费率差异很大，但市场信号是一致的：

  • 基础抠图/白底：每张 2–10 美元。
  • 完整精修（阴影、校色、清理）：每张 10–50 美元。
  • 复杂产品（玻璃、珠宝、多物品）：每张 30–100+ 美元。
  • 周期：每批 24–72 小时，改稿经常按独立轮次另行计费。

一个 200 SKU 的目录，按保守的每张 15 美元算，是 3000 美元加上数周的协调。质量通常很高——那是你付的钱买的东西——但成本随每一个新产品线性增长。

2. 云端 AI 实际花多少

云端 AI 工具宣传"无限"套餐和按张定价，但真实数字里藏着几个惊喜：

  • 按张定价（例如高清导出约 0.20–0.50 美元/张）听起来便宜，直到你乘以几百张图和反复重跑。
  • 订阅上限："无限批量"套餐已被悄悄砍成固定月限额——卖家报告过改套餐后批量功能被整个移除。
  • 积分经济：抠图消耗的积分可能和放大不同，月积分会过期。
  • 迁移风险：工具关停或改价时，你的工作流和预付积分会被搁浅。

这些都不说明云端 AI 不好——对一周只处理十张图的卖家来说，它没问题。问题在于规模化：成本、重跑和依赖都随目录规模增长。

3. 改变算法的那笔账：本地且免费

这是大多数卖家从没算过的对比：

  • 本地浏览器工具：每张 0 美元、无限次运行、不上传、不记积分。唯一成本是你的时间——而批量处理让时间成本趋近于零。
  • 质量差距：对纯色背景的产品照——目录工作的大头——本地基于颜色的抠图与外包结果不相上下。复杂材质仍需要人或更好的模型。
  • 混合打法：90% 的常规图片走本地流水线，外包只留给招牌图和复杂材质。许多卖家报告外包预算下降 70–80%。

4. 一套现实的决策框架

按数量和难度选，而不是按习惯选：

  • 少于 10 张且产品复杂：外包。你的时间比那 20 美元更值钱。
  • 10–500 张常规图：本地批量流水线。结果相同，成本趋近于零。
  • 上新的招牌图：专业拍摄或精修，刻意做预算。
  • 其余全部：保持本地、保持无限，把差价存起来。

能低成本规模化图片生产的卖家，不选单一方法——他们按难度和数量拆分工作。理解每条路径的真实单张成本，正是让这种拆分变得理性而不是被动的原因。

5. 一个算到底的例子：200 SKU 目录

给决策放上真实数字。一个 200 SKU 的卖家，每个 SKU 需要一张主图加一张细节图，共 400 张图。外包按混合价每张 15 美元算，目录成本 6000 美元，加上约两周的返工和修改轮次。云端 AI 按每张 0.30 美元导出、失败重跑，积分成本约 150–300 美元——听起来便宜，直到订阅档位砍掉月度批量、让你排进更慢的队列。

同样的目录走本地批量流水线，每张成本为零。主要开销是一次性建立一致的拍摄装置，以及跑批量通道的时间——对标准化的产品照来说，是一个下午，不是两周。卖家省下 5700–6000 美元，而且每次上新都继承同一条免费流水线，而不是重启成本时钟。

这些都不说明外包从不值得。对珠宝品牌来说，每张图都是真正的艺术指导，外包账单买到的是批量工具无法匹敌的结果。关键是：常规目录量的默认答案应该是免费本地流水线，外包只留给真正需要人眼的图片。

6. 被忽略的成本：你自己的时间

有一项成本很容易忘：你的工时。如果手动修图每张要 15–30 分钟，你的时间值 30 美元/小时，那么"免费"工具仍然让你每张损失 7.5–15 美元的时间。这就是为什么批量处理不是便利功能——它是"一本目录花真金白银"和"一本目录只花一个下午"之间的区别。

为一次批量运行计时。当本地流水线一小时内处理 200 张、而手动方法要 50–100 小时时，算法自己就解出来了。能低成本规模化图片生产的卖家不选单一方法——他们按难度和数量拆分工作，并把自身时间计入价格。