ClipCanva

DALL-E GPT Is Retired: Move Stills to ChatGPT Images, GPT Image 2, Then Image-to-Video

OpenAI retired the official DALL-E GPT in ChatGPT on 30 August 2026. What shut down, what still works, and a stills-to-video workflow with ChatGPT Images, GPT Image 2, and image-to-video.

August 31, 2026ClipCanva Editorial

DALL-E GPT Is Retired: Move Stills to ChatGPT Images, GPT Image 2, Then Image-to-Video

The official DALL-E GPT inside ChatGPT was retired on 30 August 2026. Image generation did not disappear. OpenAI’s product release notes tell you to download images you want to keep, then keep creating and editing in ChatGPT Images. User-created GPTs with image generation enabled are not part of that shutdown. The DALL-E 2 and DALL-E 3 API models were already removed on 12 May 2026. ClipCanva is not affiliated with OpenAI, Canva, or Midjourney. Treat this as a desk change: archive the stills you still need, rewrite the brief for GPT Image 2, then send only approved frames to image-to-video.

What actually shut down, and what did not

Do not mix three different sunsets. The ChatGPT GPT, the old Images API, and a 2023 research post are different objects.

Surface Official date Public claim Workflow implication
Official DALL-E GPT in ChatGPT 30 August 2026 OpenAI’s 31 July 2026 product release notes retire the official DALL-E GPT. Download images you want to keep. Continue in ChatGPT Images. User-created GPTs with image generation enabled are not affected. Stop looking for a DALL-E GPT in the picker. The chat image desk is ChatGPT Images.
DALL-E 3 research page Current page The DALL·E 3 post now points readers to ChatGPT Images 2.0 and “Create images in ChatGPT.” Do not treat the 2023 launch article as a live product.
ChatGPT Images 2.0 21 April 2026 OpenAI’s ChatGPT Images 2.0 launch is the current consumer image product. Use this as the ChatGPT stills desk. Plan gates still apply.
Image API Current docs The OpenAI image generation guide documents GPT Image models, including gpt-image-2. Organization verification may be required. New API jobs should name a GPT Image model, not dall-e-3.
dall-e-2 and dall-e-3 API 12 May 2026 OpenAI’s developer deprecation reminder removed those API models and pointed builders to GPT Image replacements. Any remaining dall-e-3 API call is a broken pipeline, not a style preference.

OpenAI has not published a full inventory of what happens to unsaved DALL-E GPT conversation assets after 30 August. If a still is not on disk, in cloud storage, or in a DAM, do not assume ChatGPT will keep serving it.

Pick the desk before you regenerate

The failure mode is using one chat as archive, art director, legal review, and video camera.

Job Where it actually runs What you must supply What the model may invent Fail condition
Recover a keeper still Local disk, DAM, or ChatGPT Images library if it is still there Filename, date, and the brief that produced it Nothing. Recovery is file ops. Regenerating “the same image” from memory and calling it the original
New still in ChatGPT ChatGPT Images Subject, crop, lighting, text rules, forbidden claims Lighting, wardrobe, background clutter, extra type Prompting as if the DALL-E GPT persona were still in the picker
Production still with references ClipCanva GPT Image 2 Locked product, character, or layout reference plus a one-job brief Texture, type, and small geometry Dumping the old DALL-E prompt unchanged and skipping a spelling pass
Layout + brand kit in a design file Canva Dream Lab / Magic Media, or your design tool Brand colors, empty type boxes, export size Decorative type and fake logos Shipping a generated poster with unreadable or invented claims
Motion from an approved frame ClipCanva image-to-video One still you already signed off, plus camera and duration Motion, extra limbs, new copy in frame Animating a draft still that still has wrong SKU text
Multi-shot campaign AI video generator after a script Scene cards, not a pile of unrelated stills Continuity breaks between shots Asking one image model to output the whole film

Canva’s own split is useful as a map, not as a partnership: Dream Lab is stills, Magic Media can generate short clips, and Canva’s May 2026 newsroom update says Image to Video now supports human faces. OpenAI’s split is now the same idea. ChatGPT Images is the stills desk. Video is a later desk.

Rewrite the DALL-E brief before you hit Generate

Old DALL-E prompts were often a style pile. GPT Image 2, on ClipCanva’s model page, is framed as create-and-edit work: a detailed brief, an optional reference, and a spelling pass on any text in the frame. Use prompt ideas if you need a reusable pattern.

Asset job: [product hero | portrait | poster | UI mock | packaging]
Keep from archive: [product geometry / face / logo file — attached, not described from memory]
Change: [one instruction]
Camera: [crop, lens feel, angle]
Light: [window / softbox / overcast / night practicals]
Text in frame: [exact string] OR [no text]
Forbidden: [living-artist style, fake certification, extra fingers, extra logo]
Output: [1:1 / 4:5 / 9:16 / 16:9], still only
Next desk: [edit in place | image-to-video | design file]

Example, product hero that used to live in the DALL-E GPT:

Asset job: 4:5 product hero for a matte black wireless earbud case.
Keep: attached pack shot as geometry reference. Do not restyle the hinge or the charging pins.
Change: move the case onto honed white marble, 3/4 view, 85mm product-lens feel, soft window light from camera left, gentle contact shadow.
Text in frame: none. No invented model number.
Forbidden: extra earbuds, extra logos, fake “studio certified” badge, mirrored type.
Output: still only. Next desk: image-to-video, 5-second slow push-in, no new objects.

If the brief needs a comparison against Google’s image stack, use the existing Imagen 4 vs GPT Image 2 page. Do not invent a winner from a single generate.

Creator checklist: 30-minute migration

  1. Export keepers first. Anything you would reuse in ads, packaging, or a character bible should leave ChatGPT as a file. The release notes’ only hard instruction is to download what you want to keep.
  2. Label the sunset. In the filename or DAM, mark source=dalle-gpt and the original date. A regenerated lookalike is a new asset.
  3. Split API from chat. If a script still calls dall-e-3, that path was already dead on 12 May 2026. Point it at a GPT Image model documented in the image generation guide, and complete org verification if the console requires it.
  4. Rewrite one brief per job. Do not paste a 2024 DALL-E style block into GPT Image 2 and hope the crop, type, and product geometry survive.
  5. Proof text in the still. Packaging, posters, and UI mockups fail on spelling, kerning, and invented SKUs. Read the letters, do not glance at the vibe.
  6. Lock the frame, then move. Only after the still is approved should you open image-to-video. Motion will not fix a wrong label.
  7. Script the clip if it has to sell. A 5-second product push still needs a hook and an end card. Draft that on the AI script generator before you generate five motion takes.
  8. Do not impersonate a retired GPT. Custom GPTs you built can keep image tools. The official DALL-E GPT is the one that left the picker.

Operator card for a still-to-video job

Archive ID: [filename of the DALL-E or ChatGPT still]
New still desk: [ChatGPT Images | ClipCanva GPT Image 2]
Reference attached: [yes/no — what it locks]
Text in frame: [exact / none]
Approved still: [yes/no]
Motion desk: ClipCanva image-to-video
Camera: [push-in | pan | locked-off]
Duration: [seconds]
Must not change: [logo, SKU, face, pack silhouette]
Must not appear: [new type, extra hands, extra product]

If the still is only a mood frame, say so. Mood frames can move. Pack shots, legal marks, and talent likeness cannot.

FAQ

Is DALL-E gone from ChatGPT?

The official DALL-E GPT was retired on 30 August 2026. ChatGPT Images remains the place OpenAI tells you to create and edit images. Custom GPTs with image generation enabled are excluded from that sunset, according to the same release notes.

Can I still call the DALL-E 3 API?

Not as a supported OpenAI API model. OpenAI’s May 2026 deprecation reminder removed dall-e-2 and dall-e-3 from the API. Current docs describe GPT Image models, including gpt-image-2.

Did ChatGPT Images 2.0 replace DALL-E 3?

OpenAI’s DALL·E 3 page now sends readers to ChatGPT Images 2.0. That is the current consumer naming. It is not a promise that every DALL-E 3 look, filter, or old conversation will replay 1:1.

What should I do with stills I already generated?

Download keepers. Store them outside the chat. If you need a new version, attach the file as a reference and write a one-job brief. Do not describe the old image from memory and call the result the original.

Should I animate every recovered still?

No. Animate the frame you would print. Use image-to-video for camera moves on an approved hero. If the clip needs dialogue, a claim, or a CTA, write the script first, then generate video.

What to do on ClipCanva today

Open GPT Image 2 with the archive still attached. Keep product geometry and type rules in the brief. When the still is clean, run image-to-video as a separate job, then continue longer sequences on the AI video generator. The DALL-E 3 model page remains a historical reference for that family; it is not a claim that OpenAI’s official DALL-E GPT is still in ChatGPT.