Best DALL-E Alternatives in 2026 (Is DALL-E Still Available?)

DALL-E is no longer the current image model from OpenAI. As reported, the DALL-E 2 and DALL-E 3 APIs were retired on 12 May 2026 and the ChatGPT DALL-E GPT on 30 August 2026, while image generation continues in ChatGPT through the newer GPT Image models. The closest like-for-like replacement is GPT Image 2, which on VdoBloom costs 3 credits at 1K, 5 at 2K and 8 at 4K.

Updated October 2026.

Disclosure: VdoBloom is our platform.

What happened to DALL-E

OpenAI’s own pages are the authority here, so confirm the current status in OpenAI’s help centre before you plan around it. What was reported by secondary sources is a two-step retirement: the DALL-E 2 and DALL-E 3 APIs went on 12 May 2026, and the ChatGPT DALL-E GPT followed on 30 August 2026. Image generation did not stop with it. ChatGPT still makes images, through the newer GPT Image line.

That distinction matters when you are choosing a replacement. If what you used was DALL-E inside ChatGPT, the product still exists and only the model behind it changed. If what you used was the DALL-E API in a script or a workflow, the endpoint is gone and the replacement has to be a different API and a different model id. Nothing below requires a ChatGPT account.

Pick by job: text in images, photoreal, editing, illustration

There is no single best replacement, because DALL-E was used for four quite different jobs. Choose by the job, then compare models inside that job.

Text inside the image. Signs, labels, packaging and posters need a model that tracks letters. GPT Image 2 is the strongest fit here, and Seedream 5.0 Pro is the alternative when the job is closer to a designed layout.
Photoreal people and products. Seedream 5.0 Pro at Standard or High, and Nano Banana 2 at up to 4K, both handle skin, fabric and product surfaces. Nano Banana 2 is flat-priced at 15 credits at any resolution.
Editing an existing image. FLUX 2 Pro image-to-image accepts one to eight inputs at 1K or 2K, and FLUX Kontext Pro handles a single targeted change for 7 credits.
Illustration and concept art. Any of the generators works; pick on price and on how literally the model follows a detailed prompt. A 1K draft on GPT Image 2 is 3 credits, which makes iteration cheap.

The GPT Image family: the closest like-for-like replacement

If the old workflow was a DALL-E prompt with a described scene and a requested style, GPT Image 2 is the natural successor. It takes long, literal prompts, covers 11 aspect ratios including auto, and prices by resolution rather than by a flat fee. At 3 credits for a 1K image you can run ten variations for 30 credits before deciding which one deserves a 4K finish at 8 credits.

GPT Image 2.5, released 2026-09-08, is the same family with two variants, Flare and Sunburst. Flare is the faster default; Sunburst trades some speed for edit precision and fidelity. Both take identical inputs and cost exactly the same on VdoBloom: 3 at 1K, 5 at 2K and 8 at 4K. The 2.5 generation extends editing from 10 reference images to 16 and fixes a 2.0 restriction: a 1:1 image on GPT Image 2 cannot be rendered at 4K, while 2.5 drops that restriction and clamps four extra aspect ratios to 1K instead of failing.

The full model comparison is in the GPT Image 2.5 guide and GPT Image 2 versus Nano Banana 2. Model pages: GPT Image 2 and GPT Image 2.5 Flare.

Nano Banana 2, Seedream 5 and FLUX options

These three cover the jobs where GPT Image falls short: many references, designed layouts and precise edits.

Model	What it is for	Credits
Nano Banana 2	Compositing with up to 14 references, ratios from 1:8 to 8:1	15 flat, 1K to 4K
Seedream 5.0 Pro	Layout-led design, generation and editing on one page, up to 10 references	10 Standard, 19 High
FLUX 2 Pro image-to-image	Edits across one to eight source images	12
FLUX Kontext Pro	A single targeted edit that keeps the frame	7

Nano Banana 2 is the one to reach for when a scene is assembled from parts: a character from one image, wardrobe from another, a product from a third. Its fifteen aspect ratios include extreme strips at 8:1 and 1:8, and the price does not change with resolution. Seedream 5.0 Pro is the layout model, with a coordinate- and mask-precise edit path and five ratios. FLUX 2 Pro keeps the source framing with an auto ratio, which is what you want for a targeted change rather than a recomposition. The model pages are Nano Banana 2, Seedream 5.0 Pro and FLUX 2 Pro image-to-image.

The same prompt on several models: prompts to try and what to look for

Run one prompt on three models before committing to any of them. These are prompts to try, not outputs, and the point is to compare four things: text accuracy, subject count, light and texture.

Text test: “A matte black coffee bag on a white counter, label reading NORTH ROAST in clean sans-serif capitals, soft daylight from the left, product photography.” Look for spelling, letter spacing and whether the label sits flat on the bag.
Count test: “Three ceramic bowls of different heights on a pale oak table, one holding lemons, one empty, one with a linen cloth; morning light, quiet interior.” Count the bowls and the lemons. Extra objects are the usual failure.
Texture test: “Close portrait of a woman in her forties against a grey studio wall, hard side light, visible skin texture, no retouching, neutral expression.” Look at pores, hair edges and whether the light stays directional.
Layout test: “A vertical poster with the headline FIELD NOTES at the top, a single photograph of tall grass in the middle, and a footer line reading Volume Two; generous margins, two colours.” Check that the hierarchy and both text blocks survive.

Compare at the same resolution and the same aspect ratio. A model that wins at 1K may lose at 4K, so finish the winner once at the resolution you will deliver.

Commercial use and watermarks

Before you publish anything, read the platform’s terms for the model you used and the terms of the service you are moving from. VdoBloom shows the credit price before every generation and refunds a failed generation, but commercial rights depend on your account and the model’s provider rules, so check the current terms rather than assuming.

Our notes on AI video and image commercial rights explain what to look for. This is not legal advice: if the image is going into an ad campaign, a product listing or client work, check the rules for your market and the platform’s own licence text. “Free to generate” is not the same as “free to use commercially”, and the two are often confused.

Frequently asked questions

Is DALL-E still available in 2026?

As reported, the DALL-E 2 and DALL-E 3 APIs were retired on 12 May 2026 and the ChatGPT DALL-E GPT on 30 August 2026. Image generation continues in ChatGPT through the newer GPT Image models; confirm the current state in OpenAI’s help centre.

What is the closest replacement for DALL-E?

GPT Image 2 is the closest like-for-like replacement for prompt-led generation. It takes detailed literal prompts, covers 11 aspect ratios, and costs 3, 5 or 8 credits at 1K, 2K and 4K.

Which alternative is best for text in images?

GPT Image 2, with Seedream 5.0 Pro as the layout-led alternative. Test one short label first; letterforms are the fastest way to see whether a model will hold your text.

Can I compare models without spending much?

Yes. Draft the same prompt at 1K on each model, then finish only the winner. A GPT Image 2 1K draft is 3 credits, and a Seedream 5.0 Pro Standard run is 10.

Do I need a subscription to use these models?

No. VdoBloom sells one-time credit packs that never expire, and no subscription is needed. New accounts get 10 free credits with no card, which covers three 1K GPT Image 2 generations.

Try the prompt tests in the image generator, then move the winner into a 4K finish. The model lineup and live prices are on the models page and the pricing page. New accounts start with 10 free credits and no card, so the comparison can begin without buying anything.