ChatGPT vs Gemini Image Generation — and When You Need BulkImagen

ChatGPT vs Gemini image generation is the comparison everyone starts with, and for a single great image it is a fair fight: ChatGPT runs OpenAI's newest image model, Gemini runs Google's Nano Banana family, and both are excellent. But the question changes the moment you need fifty images instead of one — a product catalog, a set of ad variations, a storyboard, a coloring book. Then the real question is not which model is better. It is how many times you are willing to type, wait and click Save.

This article compares ChatGPT, OpenAI's Codex, Google Gemini and BulkImagen specifically for that job. All the facts about ChatGPT, Codex and Gemini come from their official help pages and docs as of September 2026, and we say so where something isn't published.

The short answer#

• One image, lots of back-and-forth editing: ChatGPT or Gemini. Conversational editing is what chat apps are built for.
• Many images from a list, with the same settings: BulkImagen. It runs the same families of models — GPT Image and Nano Banana among them — but takes the whole list at once and hands back one ZIP.
• You are a developer and can wait up to a day: the OpenAI or Gemini Batch APIs, at half price, if you are happy to write and maintain the code.

ChatGPT: brilliant at one image, built for conversation#

ChatGPT's image generation is now powered by ChatGPT Images 2.5, released on 8 September 2026 and available on every plan. It is genuinely impressive: any aspect ratio, local edits, uploading your own images, and multi-turn refinement where you simply say "make the background darker".

For bulk work, the limits are structural rather than about quality:

• No list or spreadsheet input. OpenAI's help documentation describes no batch or CSV feature. Each image, or small set, is a chat turn.
• Up to eight images per request — as a connected set. With thinking mode on a paid plan, ChatGPT can produce "a coherent set of up to eight outputs" that build on each other. That is great for a manga page sequence; it is not the same as fifty independent product shots.
• One download at a time. Images land in your library with a Save button per image. There is no bulk download.
• Unpublished limits. The pricing page says "limited" on Free and "unlimited" on Pro (subject to abuse guardrails), but OpenAI publishes no image counts per plan.
• It takes a while. OpenAI's own help page notes a generation "may take a few minutes, depending on the complexity of your request."

None of that is a flaw for how ChatGPT is meant to be used. It is simply a conversation, and conversations are one message at a time.

Codex: it can generate images, but OpenAI points bulk work elsewhere#

OpenAI's coding agent can generate and edit images too — ask in plain language or use $imagegen, in the Codex app, CLI, IDE extension or on the web. It is a nice touch for developers who want a quick icon or placeholder without leaving their editor.

For volume, OpenAI's documentation is candid:

• Image generation in Codex uses your plan's limits "3–5x faster on average" than an ordinary turn, and it is not available on the Free plan.
• For larger batches, OpenAI's advice is to set an API key and generate through the API, so API pricing applies.

In other words, bulk images through Codex means Codex writing a script that calls the Images API — which brings you to the API route, with its own requirements below.

Gemini: generous and great with references, still one image at a time#

Google's Gemini app uses Nano Banana 2 by default, with Nano Banana Pro available as a "Redo with Pro" step on paid plans. It handles multiple reference images well and is widely praised for keeping characters consistent — a real strength.

For bulk work:

• Compute-based limits with no published numbers. Since May 2026, Gemini's allowance refreshes every five hours up to a weekly cap, with higher multipliers on paid plans. Google does not publish how many images that is.
• Downloads one by one, at full size (2K with a Google AI plan, 1K without).
• No list or spreadsheet input in the app.
• Watermarking. Every image carries an invisible SynthID watermark and C2PA metadata that cannot be turned off. The visible watermark can now be switched off in settings in most countries.

Google's other image tools (Whisk and ImageFX, now folded into Flow) let you pick the number of outputs per prompt, but there is no CSV import or bulk export in their help documentation either.

The API route: cheap, powerful, and entirely your job#

Both companies offer real batch processing to developers:

• OpenAI Batch API accepts image generation and edit requests at a 50% discount, with a 24-hour completion window. The standard Images API returns at most 10 images per call, rate limits start at 5 images per minute on the lowest usage tier, and GPT Image models may require organization verification.
• Gemini Batch API also runs image generation at 50% of the standard price, with a 24-hour target turnaround. The image models have no free tier on the API.

If you have an engineer and patience, this is the cheapest way to make images at scale. What you give up: someone writes the script, handles retries and rate limits, stores the files, and builds whatever gallery or download step the rest of the team needs.

BulkImagen: the same model families, built for lists#

BulkImagen was built for exactly the gap between "one image in a chat" and "write your own pipeline". Here is what that looks like in practice.

Paste a list, get a batch. Put one prompt per line, or upload a .csv, .xls or .xlsx, and every row becomes an image. The images render in parallel rather than one after another, and you can close the tab while they do.

Eleven models, one balance. Choose from GPT Image 2 and GPT Image 2.5 from OpenAI, Nano Banana 2, Nano Banana 2 Lite and Nano Banana Pro from Google, Seedream 4.5 and 5.0 from ByteDance, Flux 2 Pro, Grok Imagine and Z-Image. Switch models per batch without a new account or a new subscription — our model comparison shows what each one is best at.

Reference images that multiply. Upload your own photos and choose how they are used: Per image pairs every reference with every prompt (20 product photos × 4 scenes = 80 images in one batch), while Combine images sends several references into each image — a product and a logo, or two characters.

One ZIP at the end. Download the whole batch — or just the images you select — as a single archive, optionally converted to WebP or cropped to a width, with a CSV manifest mapping every file to its prompt and reference.

You see the price before you press Start. The form shows exactly how many images the batch will make and how many credits it will cost. Failed images are retried automatically and refunded if they still fail, so you only pay for images you actually get.

Pay once, no subscription. Credits are bought in packs, never expire, and work on every model. There is no monthly plan to keep alive between projects. See pricing for the current rates.

Wherever you work. The same account and credit balance work on the web, in a Chrome extension side panel, in the iPhone and iPad app, through a REST API, and through an agent skill that lets Claude Code, Cursor or Codex run batches for you — so even the Codex route above gets a bulk-native backend.

Side by side#

Which should you use?#

Use ChatGPT or Gemini when you are exploring an idea, need one hero image, or want to refine a picture by talking to it. They are the best tools in the world for that.

Use the Batch APIs when you have engineering time, very large volumes, and results can wait up to a day.

Use BulkImagen when you already know what the images should be — a product list, a scene list, a set of ad angles, forty coloring pages — and you want them all, in one consistent style, without typing each one or downloading each one. Many people do both: explore the look in a chat app, then paste the final prompt list into a bulk image generator to produce the full set.

FAQ#

Is BulkImagen using the same models as ChatGPT and Gemini? BulkImagen offers GPT Image models from OpenAI and the Nano Banana family from Google, alongside models from other providers. The difference is the workflow around them: a list in, a ZIP out, rather than one chat turn per image.

Can ChatGPT generate 100 images at once? Not in one go. The most it returns per request is a connected set of up to eight images in thinking mode on paid plans, and there is no list or CSV input or bulk download.

Do I need to code to generate images in bulk? Not with BulkImagen — the web app, Chrome extension and iPhone app all take a pasted list or a spreadsheet. Code is only needed if you choose the Batch APIs or want to automate BulkImagen through its API.