How to Choose an AI Image Generator: Checklist

Choose an AI image generator by checking six things in order: the licensing terms for your use, the output formats and sizes, the editing features, provenance and content credentials, plan limits, and how the tool handles content policy. This page is an editorial checklist, version 1.0 (2026-10-03). It ranks no tool and we tested none.

The vendor examples below show what kind of information to look for. They are not recommendations.

Why use a checklist?

Image generators differ in what they let you do with the output, not only in how the images look. A tool that makes an attractive image but restricts commercial use, or exports only a format your workflow cannot use, costs time later. Check the paperwork first, then judge quality yourself with your own briefs.

1. Which licensing terms should you read?

What the sources say

• Licenses can depend on who you are. Stability AI's license page says its Community License covers research, non-commercial and commercial use for those under USD 1M in annual revenue. Organizations above that need an Enterprise License to use the Core Models commercially (Stability AI license). We did not review the full agreement.
• The same page says the Community License restricts using the models or their outputs to create competing foundational models.
• The US Copyright Office has a report titled "Copyright and Artificial Intelligence, Part 2: Copyrightability," published January 29, 2025. The page describes it as addressing the copyrightability of outputs created using generative AI (US Copyright Office). We did not read the report's full text and draw no conclusion from it.

What to consider

Read these items in the vendor's terms of service or license, for the plan you will use:

1. Who owns the output, and what rights does the vendor keep? Look for any license the vendor keeps to your prompts and outputs.
2. Is commercial use allowed on your plan? Some terms depend on plan or on your organization's revenue. Look for thresholds.
3. Does the vendor train on your inputs or outputs, and can you opt out? Check separately for free and paid plans.
4. What is restricted? Look for rules on competing models, on recognizable people or brands, and on regulated uses.
5. Do the terms differ by model? One vendor can ship several models with different licenses.
6. Are there indemnity or warranty terms? Some terms offer protection against claims and some disclaim it. Compare the actual wording.

Copyright law on AI-generated works differs by country. We give no legal conclusion. For work central to your business, ask a lawyer.

2. Which output formats and sizes do you need?

What the sources say

OpenAI's image generation documentation shows what a vendor can document. The default format is png, and you can also request jpeg or webp. For transparent backgrounds, it says to set the background to transparent and use png or webp. Custom sizes require width and height to be multiples of 16, an aspect ratio between 1:3 and 3:1, and neither edge above 3840 pixels (OpenAI image generation guide). Other vendors will list different values.

What to consider

List what your workflow needs before you compare:

• File types: png, jpeg, webp, or vector. Check whether a tool exports true vector files or only raster images.
• Resolution and aspect ratios for your channels, such as web, print, social or presentations.
• Transparent backgrounds, if you place images over other designs.
• Whether the free plan changes any of these, for example by lowering resolution or adding a watermark.

3. What editing do you need?

What the sources say

OpenAI's guide lists three capabilities: generating new images from prompts, modifying existing images "partially or entirely" with a new prompt, and editing parts of an image by uploading an image and a mask. Its Responses API also supports multi-turn editing (OpenAI image generation guide).

What to consider

Ask these questions of any tool:

• Can you edit part of an image without regenerating the whole thing?
• Can you upload your own image as a starting point or as a style reference?
• Can you keep a character, product or style consistent across several images? The OpenAI guide lists consistency across generations and precise placement of elements as documented difficulties, so test this yourself if you need it.
• Can you extend, resize or remove objects?
• Does editing work in the product you use, or only in an API?

4. What do content credentials and provenance mean for you?

What the sources say

• C2PA, the Coalition for Content Provenance and Authenticity, describes itself as providing "an open technical standard for publishers, creators and consumers to establish the origin and edits of digital content." It describes Content Credentials as working like a "nutrition label" for digital content (C2PA).
• The Content Authenticity Initiative says Content Credentials can include the content's ingredients, the date, time and location of production, and the device or software used. Which data appears depends on what each implementer supports and what the person applying them includes. It also says identity is optional and that Content Credentials are not intended to prescriptively indicate whether content is "real" (Content Authenticity Initiative).
• OpenAI's developer documentation says it checks two provenance signals on images: C2PA Content Credentials and SynthID, a watermark embedded in supported media. It says editing, converting or sharing a file can remove its metadata, while a SynthID watermark is part of the image and "may survive some transformations." It also says a "not_detected" result does not rule out that content was generated by OpenAI or by another company's model (OpenAI content provenance guide).

What to consider

• Check whether the tool attaches Content Credentials, a watermark, both or neither, and whether you can turn them off on your plan.
• Check whether the tool can read credentials on images you upload.
• Know the limits. OpenAI says editing, converting or sharing a file can remove its metadata, and the Content Authenticity Initiative says identity may not be displayed on some platforms. Treat credentials as evidence of origin when present, not proof when absent.
• If your employer, client or platform requires disclosure of AI-generated images, check their rules, then check whether the tool supports them.

5. What do plan limits look like?

What to consider

We list no plan figures, because plans change and we compare no tools here. Read the vendor's current pricing page and look for:

• The unit. Is usage measured in images, credits, minutes of compute or a monthly allowance? Does a larger image or higher quality setting use more of it?
• What the free plan restricts. Look at resolution, watermarks, queue priority, commercial rights and daily caps.
• Concurrency and speed. Check how many jobs run at once and whether faster modes cost extra.
• What happens at the limit. Does generation stop, slow down, or bill at an overage rate?
• Whether unused allowance rolls over.
• Team features. Look at shared libraries, seat pricing, admin controls and data-handling terms for business plans.
• Price changes. Note the date you read the page. Prices and allowances can change without notice.

6. How does the tool handle content policy and limits?

What the sources say

OpenAI's guide says the system filters all prompts and generated images according to usage policies, and returns a "moderation_blocked" error when a request is blocked. It also lists constraints: complex generations can take up to 2 minutes, text rendering remains imperfect, and the model may have difficulty with precise placement in structured or layout-sensitive compositions (OpenAI image generation guide). These are one vendor's statements and say nothing about others.

What to consider

• Read the usage policy for what is blocked. A blocked brief wastes time if you work with real products, public figures or sensitive topics.
• Look for the vendor's own list of known limitations, and test them with your briefs.

How can you compare tools fairly?

This is a method you can run yourself. We have not run it.

1. Write 5 to 10 briefs from your real work. Include the hardest ones, such as text in an image, a product, or a consistent character.
2. Use the same briefs on each tool. Record the tool, model name, plan, settings and date.
3. Generate the same number of attempts per brief. Pick a number, such as four, and keep it constant.
4. Mark each result usable or not usable for your purpose. Decide your pass criteria before you look.
5. Store the outputs and the prompts. Settings and models change, so keep a record.
6. Work out the cost per usable image. Divide what you spent by the number of usable images. Use the price and allowance from the vendor's page on the date you tested.

Limitations of this page

• It names no best tool and has no tool profiles. We found no basis for ranking.
• The vendor examples come from one or two vendors' pages and are not a survey.
• We could not open some vendors' help pages, including Adobe, Midjourney and parts of OpenAI's help center, so we do not describe their terms.
• We tested no generator. Every question here is something to check, not a finding.
• Terms and documentation change. The statements above are as read on 2026-10-03.

When is this checklist not the right choice?

• You need a final legal answer on ownership. A checklist tells you what to read, not what it means for your situation.
• You need to choose between two specific tools today. Run the comparison method with your own briefs.
• You need exact current prices. Use the vendor's pricing page.
• You are choosing other kinds of AI tools. Our other selection guides cover AI coding assistants and AI automation tools.