Direct answer: public AI image-generation prices in 2026 are roughly: Google Gemini 3 Pro Image at about USD 0.134/image for 1K/2K and USD 0.24 for 4K; Gemini 3.1 Flash Image at about USD 0.067/image for 1K and USD 0.15 for 4K; Nano Banana 2.1 at about USD 0.034/image for 1K; Stability AI Stable Image Ultra at about USD 0.08/image, SD3.5 Large at 0.065, Core at 0.03, and SDXL from 0.009. OpenAI image models bill by token (e.g., gpt-image-2.5 image input USD 8/million tokens and output USD 30/million tokens), so there is no fixed per-image price. At 1,000/10,000/100,000 images, API costs range from about USD 9 to 13,400; usable-image rate, retouching, brand consistency, and rights review dominate TCO.
This guide is for manufacturing marketing, e-commerce, design, IT, and procurement teams producing product images, catalogues, ads, social assets, and e-commerce detail pages. Data cut-off: 31 October 2026. Prices come from official Google Cloud Vertex AI, Stability AI, and OpenAI pages. Midjourney/Adobe subscription pages did not provide stable public prices during this check and are excluded from the numeric comparison. All prices exclude tax; USD and RMB are not converted and no ranking is implied. This is public-source research and an editorial model, not a quote, and it does not promise image quality, usable rate, savings, or a go-live date.
Per-image price is not cost per usable image: at a 10% usable rate, a USD 0.03 image is USD 0.30. Product shape/colour/logo/text and rights review need human checks; generation speed alone is not acceptance.
What are the official AI image billing units in 2026?
Google bills by output-image tokens/resolution, Stability by credits (1 credit = USD 0.01), and OpenAI by image tokens. Align “per image” vs “per token” first.
There are at least three ways to buy AI image generation: a per-image/per-token API, a credit subscription, and generation credits inside a design tool. Prices and units differ; a single “USD 0.03 per image” number is not a full comparison.
| Service / model | Billing unit | Official list price | Note |
|---|---|---|---|
| Gemini 3 Pro Image | Output image tokens | 1K/2K about USD 0.134/image; 4K about USD 0.24 | 1,120 / 2,000 output tokens; roughly 1MP / 4MP / 16MP |
| Gemini 3.1 Flash Image | Output image tokens | 512 about 0.045; 1K 0.067; 2K 0.101; 4K 0.15/image | By output resolution |
| Gemini 3.1 Flash-Lite Image | Output image tokens | 1K about USD 0.034/image | Low-latency/low-cost tier |
| Gemini Nano Banana 2.1 | Output image tokens | 1K about 0.034; 2K 0.05; 4K 0.113/image | By output resolution |
| Stability Stable Image Ultra | Credit | 8 credits = USD 0.08/image | 1 credit = USD 0.01 |
| Stability SD3.5 Large | Credit | 6.5 credits = USD 0.065/image | Flagship 8B |
| Stability SD3.5 Large Turbo | Credit | 4 credits = USD 0.04/image | Faster |
| Stability SD3.5 Medium / Flash | Credit | 3.5 / 2.5 credits = 0.035 / 0.025/image | Mid/low tier |
| Stability Stable Image Core | Credit | 3 credits = USD 0.03/image | Low cost |
| Stability SDXL 1.0 | Credit | From 0.9 credits = from USD 0.009/image | Legacy/basic generation |
| OpenAI gpt-image-2.5 / 1.5 / mini | Image tokens | Image input 8 / 8 / 2.5 per million tokens; image output 30 / 32 / 8 per million tokens | No fixed per-image price; use the official calculator |
Official list prices in USD, excluding tax, as of 31 October 2026. Google’s per-image values are approximations derived from output tokens and resolution; actual billing is by token. Stability: 1 credit = USD 0.01. OpenAI bills by token and has no fixed per-image price. Midjourney/Adobe subscription pages did not provide stable public prices and are excluded.
What do 1,000, 10,000, and 100,000 images cost per month?
Low tiers are about USD 9–34 per 1,000 images; high tiers about USD 80–134; 100,000 images range from about USD 900 to 13,400, driven by model and resolution.
| Option | 1,000 images | 10,000 images | 100,000 images | Basis |
|---|---|---|---|---|
| Gemini 3 Pro Image (1K/2K) | USD 134 | USD 1,340 | USD 13,400 | 0.134/image |
| Gemini 3.1 Flash Image (1K) | USD 67 | USD 670 | USD 6,700 | 0.067/image |
| Gemini Nano Banana 2.1 (1K) | USD 34 | USD 340 | USD 3,400 | 0.034/image |
| Stability Stable Image Ultra | USD 80 | USD 800 | USD 8,000 | 0.08/image |
| Stability SD3.5 Large | USD 65 | USD 650 | USD 6,500 | 0.065/image |
| Stability Stable Image Core | USD 30 | USD 300 | USD 3,000 | 0.03/image |
| Stability SDXL 1.0 | USD 9 | USD 90 | USD 900 | from 0.009/image |
Editorial model: images × unit price; excludes tax, regeneration, retouching, brand review, storage, and rights/licensing. Google 4K is separate: 3 Pro Image about 0.24/image, 3.1 Flash Image about 0.15, Nano Banana 2.1 about 0.113; the table uses 1K/2K. OpenAI is token-based and not in this table.
Why can a USD 0.03 image become USD 0.30?
If 10 generations produce one usable image, cost per usable image = API price ÷ usable rate; 0.03 ÷ 10% = 0.30 before retouching.
| Usable-rate assumption | API USD 0.03/image | API USD 0.067/image | API USD 0.134/image | Fit |
|---|---|---|---|---|
| 10% | USD 0.30 | USD 0.67 | USD 1.34 | Complex products/strict brand |
| 25% | USD 0.12 | USD 0.268 | USD 0.536 | General marketing |
| 50% | USD 0.06 | USD 0.134 | USD 0.268 | Simple backgrounds/concepts |
Editorial model: per-image API price ÷ usable rate; excludes retouching, review, rights, and storage. Measure usable rate on your own products and brand standards; do not assume a vendor demo’s rate.
Why is the API fee not the image project budget?
Regeneration, selection, retouching, brand consistency, product truth, copyright/portrait/trademark review, and design layout are the main lines; the API only produces candidates.
| Cost line | Included in API price? | How to meter | Common omission |
|---|---|---|---|
| Generation API | Yes | Images × price; tokens | 4K/high resolution, regeneration |
| Usable rate/regeneration | No | Generated images ÷ usable images | Counting “successful generation” instead of “usable” |
| Retouching/compositing/layout | No | Selection, cut-out, colour, copy layout | Counting API only, not design hours |
| Product truth | No | Shape/colour/material/structure/size checks | Generated image does not match the product |
| Brand consistency | No | Style/reference/template/review | Every image has a different style |
| Copyright/trademark/portrait | No | Source, licence, logo/person/font | Generated image contains another logo/celebrity face |
| Storage/CDN | No | Originals/variants/traffic | Keeping every candidate indefinitely |
| Human review | No | Images × review minutes × labour | Assuming generated images can be published |
Editorial guidance, not an accounting taxonomy. Follow corporate finance policy, vendor contracts, and actual invoices. AI image copyright/commercial boundaries are covered separately on this site; this article does not repeat the legal analysis.
How to choose by scenario
Concept exploration/large batches: low-cost models. Final product/ads: high-quality model + retouching. Product truth/rights risk: human review first.
| Scenario | Start with | Why | Do not ignore |
|---|---|---|---|
| Concept sketches/brainstorming | Stability Flash/Core/SDXL | Low per-image cost, many iterations | Usable rate, style consistency |
| Product catalogue/final ads | Gemini Pro/Stability Ultra + retouch | Higher quality/detail for public use | Product truth, brand rules, rights |
| Images with text/logos | Test text/logo accuracy first | Generated text often needs post-check | Typos, fake logos, font licence |
| Fixed brand style | Reference/style control + template | Reduces style drift | Cross-image consistency, review |
| High-volume API integration | Compare per-image/per-token | Low marginal cost, pipeline-ready | 4K, regeneration, storage, concurrency |
| People/client/celebrity images | Consent first, then generate | Portrait/trademark risk is high | Written consent, use/term |
Editorial guidance, not a vendor ranking. Generate 100–300 images with the same product/scenario prompts; compare usable rate, product truth, text/logo, style, and retouching hours.
What should procurement and acceptance measure?
Measure usable rate, product truth, text/logo, brand consistency, rights/portrait, and cost per usable image — not per-image price or a demo alone.
| Acceptance item | Suggested definition | Evidence |
|---|---|---|
| Usable rate | Usable ÷ generated, by scenario/model | Generation record, selection record |
| Product truth | Shape/colour/material/structure/size match | Product photos, comparison sheet |
| Text/logo | Brand/model/label/logo checked | Originals, zoomed checks |
| Brand consistency | Style/palette/composition/template fit | Brand guide, review notes |
| Copyright/trademark/portrait | Source, licence, person/font/logo | Licence files, review record |
| Resolution/format | 1K/2K/4K, ratio, transparent/source file | File spec, export record |
| Cost per usable image | (API + regeneration + retouch + review) ÷ usable images | Invoice, hours, selection record |
| Data/retention | Prompts/images/training use, region, permissions | Contract, configuration, audit |
There is no universal usable-rate threshold; catalogues, ads, social, and concept work differ. High-risk/public content needs human review, not only automated scoring.
Who should read this — and who should not?
For teams producing product images, catalogues, ads, and social assets. Not for teams without product references, brand rules, or rights review that only want the cheapest batch output.
Read if: you work in marketing, e-commerce, design, brand, IT, or procurement; you are comparing Google Gemini image, Stability AI, and OpenAI image models; you need to compare per-image/per-token/credit pricing, usable rate, and retouching. Skip if: you have only a few images and no brand guide; product images must match the real product exactly but you will not do human checks; you want celebrity/client likeness without consent; you treat an AI image as legal/medical/safety evidence. Supports: per-image comparison, 1,000/10,000/100,000-image models, usable rate, TCO, scenario selection, and acceptance metrics.
Lingyi does not rank image vendors or claim any model is “most usable”. If you need image generation integrated with CMS, product catalogues, permissions, and review, see Ai website builder/LingCMS or LingHarness; for a few concept images, a public API/tool is usually faster. To scope product-image standards, usable rate, and acceptance, talk to us.
See our separate notes on AI image copyright and commercial boundaries, AI voice-over cost, and document OCR cost; this article only covers per-image/usable-image generation cost.
FAQ
How do I compare Google’s and Stability’s per-image prices?
Google’s per-image values are approximations from output tokens/resolution; actual billing is by token. Stability is credit-based with 1 credit = USD 0.01, so per-image price is fixed. Compare the same resolution, use case, and usable rate; do not compare 512 with 4K.
Why does OpenAI not have a fixed per-image price?
OpenAI bills by image input/output tokens, e.g., gpt-image-2.5 at USD 8/million image input tokens and USD 30/million output tokens. Tokens per image depend on resolution, quality, and mode; use the official calculator or actual invoice, not someone else’s “per-image” number.
What does a 10% usable rate mean?
One out of ten generations meets the standard. At USD 0.03/image API, cost per usable image is USD 0.30 before retouching and review. Measure usable rate on your own products, prompts, and brand standards.
What does 1,000 images cost per month?
At public prices: Stability SDXL about USD 9, Core 30, SD3.5 Large 65, Ultra 80; Google Nano Banana 2.1 1K about 34, 3.1 Flash 1K 67, 3 Pro 1K 134. Add usable rate, retouching, review, and rights; the table is an editorial model, not a quote.
Can AI images go straight into a product catalogue?
It depends on product truth and brand/legal requirements. Generated images may not match the real shape/colour/material/structure, and text/logos may be wrong; public catalogues need human checks against the product. Copyright/trademark/portrait/font licensing is covered separately; this is not legal advice.
How should we accept image-generation ROI?
Generate 100–300 images with the same prompts and product photos; compare usable rate, product truth, text/logo, brand consistency, retouching hours, and cost per usable image. Do not rely on per-image price, speed, or a vendor demo; high-risk content needs human review.
Method and limitations
This article uses only official pricing/product pages accessible on 31 October 2026. It does not use private contracts, customer invoices, or vendor internal data. Prices, resolution tiers, models, free allowances, and commercial terms can change. The article has no actual enterprise agreement, tax, FX, retouching, or rights-review data, so every monthly cost is an editorial model, not a quote. Google’s per-image values are approximations from output tokens; OpenAI is token-based and not forced into a per-image price; Midjourney/Adobe subscription pages did not provide stable public prices and are excluded. Usable rate, retouching hours, and brand standards are assumptions.
Sources and basis
Data cut-off: 31 October 2026. The following are official pricing/product pages; prices, resolution tiers, allowances, and terms follow the original page and contract.
- Google Cloud, Vertex AI generative AI pricing: cloud.google.com/vertex-ai/generative-ai/pricing. Gemini 3 Pro Image, 3.1 Flash Image, Flash-Lite, and Nano Banana 2.1 output-token/per-image estimates.
- Stability AI, API Pricing: platform.stability.ai/pricing. 1 credit = USD 0.01; Stable Image Ultra/SD3.5/Core/SDXL credit prices.
- OpenAI, API Pricing: platform.openai.com/docs/pricing. gpt-image-2.5/1.5/mini image input/output token prices.