Compare
Nano Banana vs ChatGPT Images vs Midjourney vs Flux: which should you use?
No single winner: pick by the job. According to each vendor (checked Oct 23, 2026), Nano Banana 2.1 suits edits with many references, ChatGPT Images 2.5 suits chat-style edits and transparent backgrounds, Midjourney V8.2 suits stylised art, and FLUX 3 suits layout control. We have not tested them.
By Found Motive · Published · Latest claim verified . We have not tested these tools ourselves. How to read the labels.
We re-check every claim on this page each Tuesday. Last check: Oct 23, 2026. Next check: Tue, Oct 27, 2026.
Disclosure. Found Motive has no affiliate links on this page and earns nothing from the tools on it. Where we do have an affiliate link, it is marked next to the link and does not change what we say. How we handle affiliate links.
Which is best: Nano Banana, ChatGPT Images, Midjourney or Flux?
No single winner; it depends on the job. According to each vendor (checked Oct 23, 2026), Nano Banana 2.1 suits edits with up to 14 reference images, ChatGPT Images 2.5 suits conversational edits and transparent backgrounds, Midjourney V8.2 suits stylised art, and FLUX 3 suits precise layout with bounding boxes. We have not tested them.
VENDOR-CLAIM Google DeepMind: Nano Banana 2.1 model card (opens in new tab) · OpenAI: Introducing ChatGPT Images 2.5 (opens in new tab) · Midjourney docs: Version (opens in new tab) · Black Forest Labs: release notes (opens in new tab) · Verified
How do Nano Banana, ChatGPT Images, Midjourney and Flux compare?
Every cell is a vendor claim with its source and date. "Not stated" means the sources we read do not say, so we did not fill it in from memory. There are no prices or plan rows by design.
| Feature | Nano Banana 2.1Google | ChatGPT Images 2.5OpenAI | Midjourney V8.2Midjourney | FLUX 3 ImageBlack Forest Labs |
|---|---|---|---|---|
| Current version | Nano Banana 2.1, model card published Oct 6, 2026. API id gemini-nano-banana-2.1. | ChatGPT Images 2.5, announced Sep 8, 2026. API models gpt-image-2.5-flare (default) and gpt-image-2.5-sunburst (more precision). | V8.2, the default version since Jul 24, 2026. | FLUX 3 Image, launched Oct 1, 2026. FLUX 3 Image Klein followed on Oct 15, 2026. |
| Edits an existing image | Yes. Google describes it as an image generation and conversational editing model. | Yes. Edits by instruction across several turns, and you can place comments directly on an image. | Yes, with the Edit Model: instruction edits, inpainting and outpainting. | Yes. Generate and edit through one endpoint. |
| Reference images | Up to 14, with consistency for up to 4 characters and fidelity for up to 10 objects, according to Google. | Reference photos are supported. The sources we read do not state a count. | Up to 4 image references at once with the Edit Model. | Up to 10 reference images. |
| Highest output size | 4K. Output comes in 1K, 2K and 4K (default 1K). | API: neither edge over 3,840 px and up to 4K in total pixels. The in-app limit is not stated. | HD images at 2,048 px (2K) without upscaling, in V8.1 and V8.2. | Up to 4K. |
| Layout or sketch control | Mask and doodle-based editing. Google lists partial instruction following in masked and doodle editing as a known limit. | Sketch: type @Sketch to draw a rough layout and ChatGPT turns it into an image. Templates cover formats such as Poster and Merch. | Inpainting and outpainting regions with the Edit Model. | Bounding boxes place each element and can confine an edit to an area. |
| Transparent background | Not stated in the sources we read. | Yes. The API supports background: transparent, and OpenAI says Images 2.5 handles transparent backgrounds. | Not stated in the sources we read. | Not stated in the sources we read. |
| Grounding in web search | Yes: grounding with Google Web and Image Search. | Not stated in the sources we read. | Not stated in the sources we read. | Not stated in the sources we read. |
| Where you use it | Gemini app, Google AI Studio, Gemini API, Search AI Mode, Google Ads, Google Flow and Google Stitch. | ChatGPT on every tier, ChatGPT Work, Codex and the Images API. | midjourney.com and Discord. | Black Forest Labs' API. Klein weights are on Hugging Face and load in ComfyUI. |
| Open weights | No. The sources we read describe a hosted model. | No. Hosted in ChatGPT and the API. | No. Hosted on midjourney.com and Discord. | Klein only, under the FLUX Kommunity License. Larger organizations need a commercial license or the API. |
| Provenance marker | SynthID watermark. | C2PA metadata and invisible watermarking. | Not stated in the sources we read. | Not stated in the sources we read. |
What is Nano Banana 2.1 good at?
Nano Banana 2.1 is Google's current image model, published October 6, 2026. According to Google, it generates and edits images at 1K, 2K and 4K, accepts up to 14 reference images and can ground results in Google Search and Image Search. Google lists blurry small text at 1K and imperfect character consistency as known limits.
Strengths
Up to 14 reference images, with consistency for up to 4 characters and fidelity for up to 10 objects, according to Google.
Grounding with Google Web and Image Search, so images can reflect current facts.
Limits
Google says small text is "often blurry in 1k" and long paragraphs render poorly.
Character consistency between input and output is "not always perfect", according to the model card.
OFFICIAL Google DeepMind: Nano Banana 2.1 model card (opens in new tab) · Gemini API: Nano Banana 2.1 (opens in new tab) · Gemini API: image generation (opens in new tab) · Verified
What is new in ChatGPT Images 2.5?
ChatGPT Images 2.5, released September 8, 2026, is OpenAI's current image model, available on every ChatGPT tier, according to OpenAI. It edits by instruction across several turns, lets you comment on parts of an image, and accepts a rough Sketch as a guide. Developers get gpt-image-2.5-flare for speed and gpt-image-2.5-sunburst for precision.
Strengths
OpenAI says Images 2.5 is "better at editing only what you've asked for" and follows editing instructions more reliably across multiple turns.
The API supports transparent backgrounds, and neither edge may pass 3,840 px.
Limits
OpenAI lists its own limits: complex prompts can take up to 2 minutes, text placement can still struggle, and character and brand consistency is not guaranteed.
Outputs carry C2PA metadata and invisible watermarking.
OFFICIAL OpenAI: Introducing ChatGPT Images 2.5 (opens in new tab) · OpenAI API: image generation guide (opens in new tab) · Verified
What does Midjourney V8.2 do best?
Midjourney V8.2, the default since July 24, 2026, focuses on aesthetics, image quality and Personalization, according to Midjourney. Its Edit Model, announced August 27, 2026, handles instruction edits, inpainting, outpainting and up to four image references. HD images render at 2048 pixels. You use it on Midjourney's website or in Discord.
Strengths
V8.2 is "an update focused on aesthetics, image quality, and Personalization", according to Midjourney's docs.
The Edit Model replaces Omni Reference, Character Reference and the Retexture tool.
Limits
Inpainting and outpainting on HD images downscale the result to SD.
No free plan: "All Midjourney plans are subscriptions."
OFFICIAL Midjourney docs: Version (opens in new tab) · Midjourney: Version 8.2 (opens in new tab) · Midjourney: Edit Model for V8 (opens in new tab) · Verified
What is FLUX 3 and who makes it?
FLUX 3 Image, launched October 1, 2026 by Black Forest Labs, generates and edits through one endpoint, takes up to ten reference images, outputs up to 4K and lets you place elements with bounding boxes. FLUX 3 Image Klein (October 15, 2026) is a 5.6B open-weights version you can run yourself. Quality comparisons are Black Forest Labs' own tests.
Strengths
Generate, edit and combine up to ten reference images, and draw bounding boxes to place each element.
Klein has 5.6B parameters and open weights on Hugging Face, with the same features as FLUX 3 Image.
Limits
Black Forest Labs' "ranks first" claims for Klein are internal evaluations, not independent tests.
Klein uses the FLUX Kommunity License, which sets commercial terms by company size, so read it before client work.
OFFICIAL Black Forest Labs docs: FLUX 3 Image (opens in new tab) · Black Forest Labs: release notes (opens in new tab) · Black Forest Labs: FLUX 3 Image Klein (opens in new tab) · Verified
Which should you pick for your job?
| Pick | …if you |
|---|---|
| Nano Banana 2.1 |
|
| ChatGPT Images 2.5 |
|
| Midjourney V8.2 |
|
| FLUX 3 |
|
These picks repeat each vendor's own claims. We have not tested the tools, so we do not rank them on quality.
Free prompts to test all four
Untested templates: run the same prompt in each tool and compare text, layout and edits side by side.
Text in the image UNTESTED TEMPLATE
A square poster for a neighbourhood bake sale. Headline: "SATURDAY BAKE SALE". Subline: "10 am, Elm Street Hall". Hand-drawn croissant illustration, warm cream background, no other text.
Edit a photo you already have UNTESTED TEMPLATE
Upload a product photo. Prompt: "Change only the background to a plain light grey studio sweep. Keep the product, its label and its shadow exactly as they are."
Place elements by layout UNTESTED TEMPLATE
A kitchen scene in 16:9. Left third: a blue kettle on the stove. Centre: a wooden table with a bowl of lemons. Right third: a window with a green plant. Soft morning light.
Put it to work
What we have not verified
- Midjourney API availability and plan terms: its plan pages were not re-read for this page.
- ChatGPT Images in-app maximum resolution, reference-image count, and which model variant the ChatGPT app uses for a given prompt.
- Transparent-background, search-grounding and provenance support for the tools marked "Not stated".
- Text-rendering quality, speed and realism rankings: no neutral source exists, vendor claims only, and we have not tested.
- What the Gemini app's model picker is called today, even though Google's model card lists the Gemini app as a Nano Banana 2.1 channel.
- Whether "Flux" in a search means Black Forest Labs' model. It is probable but ambiguous, so we write "FLUX (Black Forest Labs)" when it matters.
Quick answers
Which is best: Nano Banana, ChatGPT Images, Midjourney or Flux?
No single winner; it depends on the job. According to each vendor (checked Oct 23, 2026), Nano Banana 2.1 suits edits with up to 14 reference images, ChatGPT Images 2.5 suits conversational edits and transparent backgrounds, Midjourney V8.2 suits stylised art, and FLUX 3 suits precise layout with bounding boxes. We have not tested them.
What is Nano Banana?
Nano Banana is Google's name for its Gemini image models. The current one is Nano Banana 2.1, published on October 6, 2026, which generates and edits images up to 4K, takes up to 14 reference images and can ground images in Google Search. It sits in the Gemini app, Google AI Studio, the Gemini API and Google Flow.
Is Flux the same as Midjourney or ChatGPT?
No. FLUX is a separate family of image models from Black Forest Labs, a different company from Midjourney and OpenAI. The current flagship is FLUX 3 Image, launched October 1, 2026, and FLUX 3 Image Klein (October 15, 2026) is a smaller version with open weights that you can download and run yourself.
Which one can edit an image I already have?
All four can. ChatGPT Images 2.5 edits by instruction and lets you comment on parts of an image. Nano Banana 2.1 does conversational editing with reference images. Midjourney's Edit Model handles instruction edits, inpainting and outpainting. FLUX 3 Image generates and edits through one endpoint, with bounding boxes to place changes. All per the vendors, checked Oct 23, 2026.
Which one is best for text inside the image?
We can't rank them, because we haven't tested text rendering and no vendor gives a neutral comparison. Google says Nano Banana 2.1 improves text rendering but admits small text can look blurry at 1K. OpenAI says Images 2.5 handles complex layouts. Test your own wording in each tool before committing to one.
Can I run any of these on my own computer?
Only one. FLUX 3 Image Klein, released October 15, 2026, has open weights that you can download from Hugging Face and run on your own GPU, under the FLUX Kommunity License, which sets commercial terms by company size. Nano Banana, ChatGPT Images and Midjourney are hosted services, so you use them through their apps or APIs.
More comparisons
Get the weekly roundup
Playbooks, new features and tool comparisons, in one short email every Thursday.