# DALL-E 3

> Source: https://aiwiki.ai/wiki/dall_e_3
> Updated: 2026-10-28
> Categories: AI Models, Image Generation, OpenAI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "DALL-E 3." aiwiki.ai, 28 Oct 2026. https://aiwiki.ai/wiki/dall_e_3
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**DALL-E 3** (stylized by OpenAI as **DALL·E 3**) is the third generation of OpenAI's text-to-image system, announced on September 20, 2023 and released to [ChatGPT Plus](https://aiwiki.ai/wiki/chatgpt_plus) and Enterprise subscribers in October 2023, followed by a standalone API endpoint named `dall-e-3`.[^1][^2] OpenAI described it as built "natively on ChatGPT" and said it "understands significantly more nuance and detail than our previous systems, allowing you to easily translate your ideas into exceptionally accurate images."[^1] Compared with its predecessor [DALL-E 2](https://aiwiki.ai/wiki/dall-e), the model produces images that adhere much more closely to long, detailed prompts, renders legible text inside images far more reliably, and is paired with a server-side prompt-rewriting pipeline driven by [ChatGPT](https://aiwiki.ai/wiki/chatgpt).[^1][^3] The training recipe is documented in the OpenAI technical report *Improving Image Generation with Better Captions* by James Betker, Gabriel Goh, Li Jing and colleagues, which credits a bespoke image captioner and synthetic-caption training as the key intervention.[^4] DALL-E 3 was integrated into Microsoft's Bing Image Creator and Copilot products in late 2023[^5][^6] and remained OpenAI's primary image generator until it was superseded in 2025 by GPT-4o native image generation (the API model `gpt-image-1`) and, by December 2025, by `gpt-image-1.5` as the default in ChatGPT.[^7][^8][^9]

<figure class="w-56">
<img src="/images/wd-2026-10-28-openai-logo-2a509640a546-tile.png" alt="OpenAI logo" width="480" height="178" />
<figcaption>OpenAI logo. Source: <a href="https://commons.wikimedia.org/wiki/File:OpenAI_logo_2025_(wordmark).svg">Wikimedia Commons</a>, <a href="https://commons.wikimedia.org/wiki/File:OpenAI_logo_2025_(wordmark).svg#Licensing">Public domain</a>.</figcaption>
</figure>

## Overview

| Item | Value |
| --- | --- |
| Developer | [OpenAI](https://aiwiki.ai/wiki/openai) |
| Predecessor | [DALL-E 2](https://aiwiki.ai/wiki/dall-e) (April 2022) |
| Announcement | September 20, 2023[^1][^2] |
| ChatGPT availability | October 2023 (Plus / Enterprise)[^10] |
| Bing availability | October 3, 2023 (free, via Bing Chat and Bing.com/create)[^23] |
| API model id | `dall-e-3`[^11] |
| API launch | November 6, 2023 (OpenAI DevDay)[^24] |
| Output sizes | 1024x1024, 1024x1792, 1792x1024[^11] |
| Quality settings | `standard`, `hd`[^11] |
| Style settings | `vivid`, `natural`[^11] |
| Provenance | C2PA metadata (added Feb 2024)[^12][^13] |
| Free ChatGPT tier | August 8, 2024 (two images per day)[^25] |
| Successor | `gpt-image-1` (GPT-4o image generation, March-April 2025)[^7][^8] |
| API deprecation | Announced Nov 14, 2025; removed from the API May 12, 2026[^9][^11] |

<figure class="embedvideo">
<iframe src="https://www.youtube-nocookie.com/embed/sqQrN0iZBs0" title="Introducing DALL·E 3"></iframe>
<figcaption>"Introducing DALL·E 3", official video from OpenAI, published September 21, 2023. <a href="https://www.youtube.com/watch?v=sqQrN0iZBs0">Watch on YouTube</a>.</figcaption>
</figure>

## When was DALL-E 3 released? History and timeline

### Background: DALL-E and DALL-E 2

OpenAI's first DALL-E, announced in January 2021, was an autoregressive transformer that produced images by predicting discrete image tokens, conditioned on text.[^1][^26] The original DALL-E paper, "Zero-Shot Text-to-Image Generation", was led by [Aditya Ramesh](https://aiwiki.ai/wiki/aditya_ramesh) with co-authors Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever, and the system was named as a portmanteau of Salvador Dalí and Pixar's WALL-E.[^26] [GLIDE](https://aiwiki.ai/wiki/glide) is OpenAI's 2021 text-guided diffusion image model, an early step in the DALL-E lineage. DALL-E 2, announced on April 6, 2022, replaced this with a CLIP-conditioned [diffusion model](https://aiwiki.ai/wiki/diffusion_model) that operated in unCLIP image-[embedding space](https://aiwiki.ai/wiki/embedding_space), dramatically improving image fidelity but exhibiting weak prompt following: it tended to combine attributes incorrectly, struggled with multi-object scenes, ignored words in long prompts, and was notoriously poor at rendering legible text and detailed hands.[^1][^4] By mid-2023, [Stable Diffusion](https://aiwiki.ai/wiki/stable_diffusion) derivatives and Midjourney v5 had narrowed or closed the aesthetic gap with [DALL-E 2](https://aiwiki.ai/wiki/dall_e_2) while exposing the same compositional weaknesses, motivating OpenAI's next-generation effort.[^4][^21]

Aditya Ramesh continued to lead the DALL-E project through this period, working alongside a team that grew to include researchers focused on safety, evaluation, and data curation.[^26] DALL-E 2 moved from an invitation-only research preview into a beta and was opened to everyone on September 28, 2022; it remained available as an API model (`dall-e-2`) until its scheduled removal from the API on May 12, 2026.[^9][^26]

### Announcement (September 2023)

OpenAI publicly announced DALL-E 3 on September 20, 2023 with a blog post on its corporate website and a research preview campaign showcasing improved prompt following and text rendering.[^1] In the announcement OpenAI emphasized that DALL-E 3 had been built "natively on ChatGPT", meaning the chat model would automatically rewrite short user requests into more detailed image prompts before they reached the image model.[^1] Coverage on launch day from TechCrunch summarized the same three-pronged story: tighter ChatGPT integration, refusals for living-artist styles and named public figures, and an opt-out form for artists who do not want their work used to train future OpenAI image models.[^2]

At launch, OpenAI stated that DALL-E 3 would not generate images in the style of named living artists, would refuse requests depicting named public figures (politicians and celebrities), and would offer artists a form through which they could opt out their works from future training sets.[^1][^2]

### General availability in ChatGPT

DALL-E 3 began rolling out to ChatGPT Plus and Enterprise customers in October 2023, adding image generation to the chat interface.[^10] In an October 19, 2023 blog post, OpenAI confirmed broader availability and disclosed additional safety work: the company said it had trained a provenance classifier (an internal tool) that, in early internal evaluations, was over 99% accurate at identifying unmodified DALL-E 3 images and over 95% accurate after common modifications such as cropping, resizing, JPEG compression, or partial overlay with real imagery.[^10][^27]

The October release arrived as part of a broader expansion of multimodal capabilities in ChatGPT that also included GPT-4 with vision (image input) and a voice mode driven by [Whisper](https://aiwiki.ai/wiki/whisper) for transcription and a new [text-to-speech](https://aiwiki.ai/wiki/text_to_speech_ai) model for output, marking the moment when ChatGPT shifted from a text-only product to a multimodal assistant.[^10][^28] OpenAI rolled out the image tool gradually to Plus and Enterprise subscribers throughout October 2023.[^10]

### Microsoft Bing and Copilot integration

In parallel with the ChatGPT release, Microsoft integrated the model into Bing Image Creator (later renamed "Image Creator from Designer") and into the Copilot consumer surface around the September 21, 2023 Copilot rebrand, making DALL-E 3 available to non-paying users through Bing for the first time.[^5][^6][^29] Microsoft announced on September 21, 2023 that Image Creator would also come to Bing Chat Enterprise, and on October 3, 2023 it made DALL-E 3 generally available to everyone in Bing Chat and Bing.com/create for free.[^5][^23] At that point, Microsoft said Bing Image Creator had already generated more than one billion images since its launch.[^23] Microsoft's September 21 announcement had said DALL-E 3 delivered "more beautiful creations and better renderings for details like fingers and eyes."[^29]

The Bing/Copilot route was strategically important because, unlike OpenAI's paid distribution path, it was free at point of use, financed by Microsoft's existing search [advertising](https://aiwiki.ai/wiki/advertising) business.[^5][^6] On September 21, 2023 Microsoft also rebranded all variants of its Copilot family (including Bing Chat and [Microsoft 365 Copilot](https://aiwiki.ai/wiki/microsoft_365_copilot)) under the unified Microsoft Copilot name, with DALL-E 3 listed as one of the headline new capabilities.[^29]

### API release at DevDay (November 2023)

OpenAI exposed DALL-E 3 as a standalone API endpoint (`dall-e-3`) on November 6, 2023, at the company's first developer conference (DevDay), with parameters for size, quality (`standard` versus `hd`), and style (`vivid` versus `natural`).[^11][^14][^24] The API release shipped alongside the introduction of the Assistants API and the GPT-4 Turbo model family.[^14][^24] Pricing started at $0.04 per generated image for the standard 1024x1024 size.[^18][^24]

Unlike the DALL-E 2 API, `dall-e-3` did not expose endpoints for image editing (inpainting) or image variations: only the generation endpoint `POST /v1/images/generations` was supported.[^11][^24] OpenAI also stated at launch that the API would automatically rewrite user prompts "for safety reasons and to add more detail," a behavior that became one of the most-discussed practical differences between the API and the ChatGPT-facing version.[^24]

### Free tier in ChatGPT (August 2024)

For nearly a year after launch, DALL-E 3 in ChatGPT remained restricted to paying users. On August 8, 2024, OpenAI extended access to free-tier ChatGPT accounts, allowing each free user to generate up to two images per day, a feature previously limited to paid subscribers.[^25] Dataconomy wrote that the change was likely to "democratize creative image generation."[^25]

### Provenance and watermarking

On February 6, 2024 OpenAI announced that images generated by DALL-E 3, both through ChatGPT and through the API, would carry [content-credentials](https://aiwiki.ai/wiki/watermarking) metadata following the C2PA (Coalition for Content Provenance and Authenticity) standard, including a visible "CR" logo and embedded cryptographic provenance information.[^12][^13] OpenAI joined the C2PA Steering Committee three months later, in May 2024.[^30] In May 2026 OpenAI extended this multi-layered approach by adding Google DeepMind's [SynthID](https://aiwiki.ai/wiki/synthid) invisible watermarking to images generated through ChatGPT, Codex, and the OpenAI API.[^15][^31] In November 2023 OpenAI also released the *consistency decoder* used to convert DALL-E 3 latents to pixels as open-source under an MIT license; the decoder was distilled from a diffusion-based decoder down to two sampling steps using consistency distillation and could be used as a drop-in replacement for the Stable Diffusion v1 VAE decoder.[^32][^33]

### Supersession and deprecation

OpenAI's [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) image generation system, announced on March 25, 2025 and exposed in the API as the model [gpt-image-1](https://aiwiki.ai/wiki/gpt_image_1) on April 23, 2025, replaced DALL-E 3 as the default image generator in ChatGPT from launch, with DALL-E remaining available through a dedicated DALL-E GPT.[^7][^8] OpenAI stated that the new model used the GPT-4o architecture rather than the diffusion stack underlying DALL-E 3, allowing it to take both text and reference images as input within the same autoregressive image-generation process and to produce edited outputs that preserved scene identity across turns.[^7] The image-generation rollout drew unusually heavy traffic; OpenAI reported that more than 130 million users created over 700 million images in the first week, and Sam Altman announced temporary rate limits because the company's GPUs were "melting".[^57] A viral trend in which users transformed personal photos into Studio Ghibli-style images contributed to the load and drew renewed attention to copyright questions surrounding the training data.[^34]

On November 14, 2025 OpenAI announced that the `dall-e-2` and `dall-e-3` API model snapshots would be deprecated and removed from the API on May 12, 2026, with OpenAI's deprecations page now listing `gpt-image-2`, `gpt-image-1`, or `gpt-image-1-mini` as replacements; the DALL-E 3 model page states that the model has been deprecated and removed from the API.[^9][^11] On December 16, 2025 OpenAI launched GPT Image 1.5 as the new ChatGPT Images model for all ChatGPT users and in the API, succeeding the GPT-4o-based model that had replaced DALL-E 3 as ChatGPT's default in March 2025.[^7][^35] Developers in OpenAI's community forums objected in threads such as "OpenAI is making a huge mistake by deprecating DALL-E-3", arguing that the diffusion-based output had a distinct aesthetic that the newer autoregressive model did not reproduce.[^36]

## How does DALL-E 3 work? Technical details

### The "Better Captions" recipe

The publicly released OpenAI technical report associated with DALL-E 3, *Improving Image Generation with Better Captions*, by James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, Wesam Manassra, Prafulla Dhariwal, Casey Chu, Yunxin Jiao, and Aditya Ramesh, focuses almost exclusively on the training-data side of the pipeline rather than on the diffusion model architecture itself.[^4] The report's central claim, stated in its abstract, is that "prompt following abilities of text-to-image models can be substantially improved by training on highly descriptive generated image captions," addressing the noisy, short, and often inaccurate alt-text and web captions that have historically been used to train text-to-image systems by instead training a bespoke image captioner and using it to recaption the training dataset with [synthetic captions](https://aiwiki.ai/wiki/synthetic_data).[^4]

Concretely, the authors train an image captioner (described in the report as a CLIP-conditioned language model) and fine-tune it through two stages, producing "short synthetic captions" (SSC) and "descriptive synthetic captions" (DSC). The SSC pass is fine-tuned on a small dataset of captions that describe only the main subject of an image, while the DSC pass is fine-tuned on long, highly descriptive captions that also cover surroundings, background, text found in the image, styles, and coloration.[^4][^37] They then train three text-to-image models on different mixtures of ground-truth and synthetic captions and use a CLIP-based evaluation to compare prompt following.[^4] The report shows monotonic gains in CLIP score as the synthetic-caption ratio increases, with the best results obtained at a 95% synthetic / 5% ground-truth mix; the DALL-E 3 production model is then trained on this mixture.[^4][^37] The ground-truth fraction is included as a regularization device: it exposes the model to the surface statistics of real-world prompts (typical lengths, capitalization, and punctuation patterns) so the trained system does not silently expect every input to look like a DSC paragraph.[^4][^37]

A key ablation in the paper compares image generation quality at the same training compute when only the captions are changed: models trained on 95% descriptive synthetic captions score slightly higher in CLIP score than an otherwise identical model trained on ground-truth captions when evaluated with ground-truth captions, and markedly higher when evaluated with descriptive synthetic captions; this ablation uses CLIP score, not human raters.[^4] The authors argue this strongly suggests that, at frontier scale, *caption quality* rather than *image quality* is now the bottleneck for prompt following, an observation that has shaped subsequent text-to-image research.[^4][^22] The paper states that it "does not cover training or implementation details of the DALL-E 3 model" and does not disclose dataset composition, training compute, or model size; it gives no reason for these omissions.[^4]

### Failure modes documented in the paper

The paper itself acknowledges several limitations that survive into the production model. First, prompt following degrades when user prompts are shorter than typical DSCs, motivating the upstream ChatGPT rewriter described below.[^4] Second, the model has a documented deficit on *spatial reasoning* (phrases such as "to the left of", "underneath", and "behind" are unreliable), which the authors attribute to the same weakness in their synthetic captioner: it "is unreliable at stating object placement."[^4] Third, text rendering is unreliable, with words that have missing or extra characters, which the authors suspect stems from the T5 text encoder seeing whole-word tokens rather than letters.[^4] Fourth, the captioner hallucinates specific details: given a botanical drawing it often invents a plant genus and species even when they are written in the image, and it hallucinates or omits bird species, which leaves DALL-E 3 unreliable for such specific terms.[^4]

### Architecture (what is and is not public)

OpenAI has not published the architecture of the DALL-E 3 image generator. Public materials describe it as a [latent diffusion model](https://aiwiki.ai/wiki/latent_diffusion) with a transformer-based text encoder rather than the autoregressive token model used in the original DALL-E (2021); this contrasts with DALL-E 2, whose architecture was disclosed in a 2022 OpenAI paper.[^1][^4][^26] The Betker et al. report says the production model is "a scaled-up version" of its ablation models "with several other improvements"; its appendix describes those ablation models as text-conditioned U-Net latent diffusion models built on the Rombach et al. VAE with a T5 XXL text encoder, and it refers to "the T5 text encoder we used" for DALL-E 3. It does not fully describe the production model's denoiser or sampler.[^4]

Third-party explainers such as AiCoin summarize the same picture from the report's appendix: a VAE with 8x spatial downsampling (256-pixel training images become 32x32 latents), a T5 text encoder whose outputs are cross-attended to, and timestep conditioning through a modulated GroupNorm.[^4][^38] The only model component released in full is the consistency decoder, which OpenAI open-sourced on November 6, 2023; the decoder reuses the Stable Diffusion v1 latent space and was trained as a diffusion-based decoder, then distilled to two sampling steps via consistency distillation.[^32][^33]

### How does ChatGPT prompt rewriting work?

Because the synthetic captions used to train DALL-E 3 are long and descriptive, short user prompts often underperform. OpenAI's solution is a server-side prompt rewriting layer: when a user asks ChatGPT to "draw" or "create" an image, the chat model is instructed to expand the request into a much more detailed image prompt (typically a paragraph) before invoking the image tool, and to return both the rewritten prompt and the generated image to the user.[^1][^17] The same rewriting also runs by default through the API: developers calling `images.generate` with `model="dall-e-3"` receive a `revised_prompt` field in the response describing what was actually sent to the model.[^11][^17]

The system prompt that ChatGPT uses to perform this rewrite was reverse-engineered by users and published on community sites in late 2023.[^39][^40] According to those reconstructions, ChatGPT is instructed to generate exactly four images per request when no explicit count is provided, to avoid copyrighted artist names from the last 100 years (substituting three style adjectives instead), to avoid named living people including politicians and celebrities, to make each prompt an extremely descriptive and detailed paragraph, and to translate any non-English input into English before rewriting.[^39][^40] The system prompt also imposes an explicit diversity requirement: when generating images of people, the rewriter is told to specify both descent and gender for each person using direct terms.[^39][^40] OpenAI never officially released the system prompt, but the existence of such instructions is consistent with the observable behavior of the ChatGPT image tool and with the contents of the `revised_prompt` returned by the API.[^11][^17]

Rewriting cannot be globally disabled on the API; OpenAI's cookbook says the rewriting "isn't able to be disabled" but that developers can get high fidelity by giving instructions to the rewriter in the prompt, such as asking it not to add any detail and to use the prompt as-is, although in practice users have reported only partial success.[^14] An academic study by Jahani et al. at UC Berkeley, conducted as an online experiment with 1,891 participants randomly assigned to use DALL-E 2, DALL-E 3, or DALL-E 3 with automatic prompt revision, found that automatic LLM-based prompt revision reduced DALL-E 3's advantage over DALL-E 2 by nearly 58%: DALL-E 3 users with revision still beat DALL-E 2 users, but by less than users whose prompts went to DALL-E 3 unchanged, suggesting the rewriting layer can hold back the image model.[^17] A separate analysis from *The Decoder* characterized the rewriting layer as serving a covert moderation role: by transforming prompts that might violate policy into compliant variants, ChatGPT effectively functioned as a safety filter in front of the image model.[^41]

### Image quality and text rendering

In human side-by-side evaluations in the Betker et al. technical report, raters preferred DALL-E 3 over Midjourney 5.2 and [Stable Diffusion XL](https://aiwiki.ai/wiki/sdxl) on prompt following and style (on a 170-caption "DALL-E 3 Eval" set) and on coherence (on 250 randomly sampled MSCOCO captions); on prompt following, DALL-E 3 scored 153.3 Elo against -104.8 for Midjourney 5.2 and -189.5 for SDXL.[^4] The paper benchmarks against Midjourney v5.2 and SDXL 1.0 and reports DALL-E 3 winning on prompt adherence, style consistency, and overall coherence across human raters.[^4][^37] The report also shows DrawBench examples in which GPT-4 "upsampled" short prompts into detailed ones, letting DALL-E 3 correctly render compositions it would otherwise get wrong.[^4]

For automated evaluation, the report uses CLIP score on MSCOCO captions, GPT-4V judgments on Google's DrawBench prompts, and T2I-CompBench; DALL-E 3 beat DALL-E 2 and SDXL on each (for example 70.4% versus 49.0% and 46.9% on short DrawBench prompts), and OpenAI said it uploaded all samples and prompts from these comparisons to GitHub, though not the captioner.[^4] Later systems narrowed or reversed the gap: Google's 2024 Imagen 3 evaluation placed Imagen 3 ahead of DALL-E 3 on most prompt-alignment comparisons, around the time Black Forest Labs released [FLUX.1](https://aiwiki.ai/wiki/flux_1).[^49] A hands-on review found DALL-E 3 could render short labels and headings of around three to five words before text degenerated into gibberish, and that GPT-4o image generation was far more reliable at rendering text in 2025.[^42][^7]

## How much did the DALL-E 3 API cost? API and pricing

### Endpoint

`dall-e-3` was invoked through the OpenAI Images API at `POST /v1/images/generations`. Required and supported parameters at launch included:[^11]

| Parameter | Allowed values | Notes |
| --- | --- | --- |
| `model` | `dall-e-3` | |
| `prompt` | up to ~4000 characters | |
| `n` | 1 | Multi-image generation not supported in a single call.[^11] |
| `size` | `1024x1024`, `1024x1792`, `1792x1024` | |
| `quality` | `standard`, `hd` | `hd` doubles the price at 1024x1024 and raises it 1.5x at the larger sizes.[^18] |
| `style` | `vivid` (default), `natural` | `vivid` produces hyper-real images; `natural` is less stylized.[^11] |
| `response_format` | `url`, `b64_json` | |

Unlike `dall-e-2`, the `dall-e-3` endpoint did not support image editing (inpainting), image variations, or batched multi-image returns; the `/v1/images/edits` and `/v1/images/variations` endpoints remained DALL-E 2 only.[^24][^43] OpenAI's ChatGPT web interface did expose an inpainting "select an area" tool for DALL-E 3 outputs, but that capability was not exposed in the API.[^43][^44]

### Pricing

OpenAI listed per-image pricing for `dall-e-3` on its API pricing page. The published rates were:[^18]

| Quality | 1024x1024 | 1024x1792 or 1792x1024 |
| --- | --- | --- |
| Standard | $0.040 | $0.080 |
| HD | $0.080 | $0.120 |

Image generation is billed per request rather than per token; the prompt and revised prompt are not separately metered.[^18]

### Rate limits

OpenAI's help-center article says the `dall-e-3` rate limit depends on the account's usage tier.[^19] OpenAI's model page lists limits from 500 images per minute at Tier 1 to 10,000 at Tier 5, with no free-tier access, while each request could return only one image (`n` is fixed at 1 for `dall-e-3`).[^11] Free-tier ChatGPT users were further limited to two images per day after the August 2024 expansion.[^25]

### Programmatic interface example

A typical Python call using the OpenAI Python SDK looked like the following (paraphrased from the OpenAI cookbook):[^14]

```python
from openai import OpenAI
client = OpenAI()

response = client.images.generate(
    model="dall-e-3",
    prompt="A poster for a chess tournament in a 1960s Saul Bass style",
    size="1024x1792",
    quality="hd",
    style="vivid",
    n=1,
)
image_url = response.data[0].url
revised = response.data[0].revised_prompt
```

Developers were responsible for downloading the URL within the response TTL (1 hour) and for surfacing or suppressing the `revised_prompt` field as appropriate for their application.[^11][^14]

## Is DALL-E 3 safe? Safety mitigations

### System card and red teaming

OpenAI published a separate DALL-E 3 system card on October 3, 2023, documenting the safety stack and the external red-teaming process used to evaluate the model before release.[^45] The system card describes a multi-tiered safety system covering training-data filtration, prompt classifiers, output classifiers, and policy-driven refusals. Red teaming was carried out by both internal and external experts and probed four "dual-use" risk domains aligned with CBRN (chemical, biological, radiological, nuclear) categories, focusing on whether image generation could meaningfully accelerate harmful workflows; OpenAI concluded that DALL-E 3 did not provide a substantive uplift in these domains.[^45] The red team work also targeted graphic content, sexual imagery, and the model's ability to be coaxed into producing misleading or impersonation images of named individuals.[^45]

The system card also discloses that DALL-E 3 inherits known biases from web-trained image data, including a default tendency to produce images of "individuals who appear White, female, and youthful" in the absence of explicit demographic prompting, and notes that the ChatGPT prompt rewriter (which forces gender and descent attributes) was added in part to counteract this default.[^45][^46]

### Pre-training and policy filters

The DALL-E 3 system card and OpenAI's launch documentation describe a layered safety stack. Training data was filtered to remove explicit content and to reduce the prevalence of violent and graphic imagery; prompt-side classifiers refuse requests that target named living artists or that depict public figures by name; and output classifiers screen generated images before they are returned to the user.[^1][^2][^45] The classifier architecture combines a frozen CLIP image encoder for feature extraction with a small auxiliary model for safety score prediction, and the same classifier family is used for both pre-generation prompt screening and post-generation image screening.[^45]

### Provenance: C2PA, watermarking, and the detection classifier

Beginning February 6, 2024, OpenAI added C2PA "Content Credentials" metadata to every image returned by DALL-E 3 through both ChatGPT and the API.[^12][^13] The metadata is signed and lists the originating model and the generation timestamp; OpenAI also adds a small visible "CR" badge in the upper left corner of ChatGPT-served images.[^12][^13] OpenAI has cautioned that C2PA metadata is "not a silver bullet": the data is preserved by compliant tools but can be stripped by simple operations such as screenshotting or re-saving an image.[^13][^30]

OpenAI complemented C2PA with two additional measures. First, in May 2024 the company announced it had joined the C2PA Steering Committee;[^30] in May 2026 it added Google DeepMind's SynthID invisible watermarking to its images, which Google says is designed to withstand modifications such as cropping, filters, and lossy compression that can strip C2PA metadata.[^15][^31] Second, in May 2024 OpenAI announced a DALL-E 3 detection classifier through a research access program that, in internal tests, correctly identified roughly 98% of DALL-E 3 images and produced false positives on under 0.5% of human-created images, with degraded accuracy of 5-10% incorrect classification for images from non-OpenAI models.[^30]

### Bing, Copilot and the 2024 deepfake incident

When Microsoft integrated DALL-E 3 into Bing Image Creator and Copilot in late 2023, it added an additional safety layer of prompt and output filters that, in some categories, are stricter than the OpenAI defaults.[^5][^6] In late January 2024, sexually explicit non-consensual [deepfake](https://aiwiki.ai/wiki/deepfake) images of the musician Taylor Swift circulated widely on X and 4chan; investigative reporting by 404 Media traced the origin to a Telegram community using Microsoft Designer (and indirectly DALL-E 3 through Designer's image generator).[^20][^47] The bypass relied on a prompt-rewriting loophole in which users inserted descriptors between first and last names (for example "taylor 'singer' swift" or "jennifer 'actor' aniston") to evade the name-aware filter.[^47] Microsoft subsequently tightened its filters to close the loophole; asked about the images, CEO Satya Nadella said "we have to act."[^20][^48]

Separately, a Microsoft AI engineering leader named Shane Jones disclosed in late January 2024 that he had identified safety vulnerabilities in DALL-E 3 in early December 2023, including methods to bypass content filters to produce violent or explicit images, and that Microsoft's legal department had instructed him to delete a December 14, 2023 LinkedIn open letter calling on OpenAI to suspend the model.[^20][^48] Jones subsequently wrote letters to U.S. Senators Patty Murray and Maria Cantwell, Rep. Adam Smith, and Washington State Attorney General Bob Ferguson asking that DALL-E 3 be removed from public use until the issues were addressed.[^48] OpenAI responded that the techniques described did not in fact bypass its safety systems, and Microsoft tightened Designer's filters in the wake of the Taylor Swift incident.[^20][^48]

## How does DALL-E 3 compare to Midjourney and Stable Diffusion?

DALL-E 3 was released into a crowded text-to-image market. The most direct contemporaries were Midjourney v6 (released as an alpha in December 2023), Stability AI's SDXL (released July 2023), and (from early 2024) [Stable Diffusion 3](https://aiwiki.ai/wiki/stable_diffusion_3) and Google's [Imagen 3](https://aiwiki.ai/wiki/imagen_3).[^21][^49] Independent comparisons in 2024 and 2025 generally agreed on the following picture:

| Capability | DALL-E 3 | Midjourney v6 | SDXL 1.0 |
| --- | --- | --- | --- |
| Prompt following on long prompts | Strongest of the three[^21] | Improved over v5 but weaker than DALL-E 3[^21] | Weak without prompt-engineering scaffolding[^21] |
| Text-in-image rendering | Reliable only for short strings of a few words[^4][^42] | Improved in v6; less reliable than DALL-E 3[^21] | Generally poor without LoRAs[^21] |
| Aesthetic / photorealism | Strong but often visibly "AI" smoothed[^21] | Strongest aesthetic in 2024 community reviews[^21] | Tunable via fine-tunes and ControlNet[^21] |
| Customization | None (closed model)[^11] | None (closed) but supports `--sref` style references[^21] | Open weights; supports LoRA / [diffusion](https://aiwiki.ai/wiki/diffusion_model) tooling[^21] |
| Interface | [ChatGPT](https://aiwiki.ai/wiki/chatgpt), Bing/[Copilot](https://aiwiki.ai/wiki/microsoft_copilot), API[^10][^5] | Discord and web app[^21] | Local, third-party UIs, hosted APIs[^21] |
| Cost per image | $0.04-$0.12 via API[^18] | Subscription ($10-$120/month)[^21] | Free if self-hosted; metered on hosts[^21] |

Google's Imagen 3, released in 2024, surpassed DALL-E 3 on several public benchmarks, though only narrowly on OpenAI's own DALL-E 3 Eval prompt-alignment test (1,078 versus 1,077 Elo), while losing to Midjourney v6 on visual appeal.[^49] On the DOCCI benchmark (photographs with detailed captions averaging 136 words), Imagen 3 beat the second-best model, Midjourney v6, by 114 Elo points with a 63% win rate.[^49][^58] Stable Diffusion 3, introduced in February 2024 and released in June 2024 with the MMDiT (Multimodal Diffusion Transformer) architecture, also closed the gap on text rendering and prompt adherence and was available with open weights.[^51]

Industry reviews in 2024-2025 frequently recommended using the systems together: DALL-E 3 for tasks demanding faithful prompt adherence (signage, layouts with specified text), Midjourney v6 for aesthetic-led illustration, and SDXL or Stable Diffusion 3 when controllability and custom fine-tunes mattered.[^21][^42]

## Why does DALL-E 3 matter? Significance

DALL-E 3 mattered for three reasons that go beyond image quality. First, it normalized the pattern of pairing a frontier text-to-image model with an LLM-driven prompt rewriter, which has since become standard practice in Imagen 3, FLUX.1, and most commercial image APIs.[^7][^21] Second, the *Better Captions* paper popularized the use of LLM-generated synthetic captions as a training-data lever, an approach that has been adopted (and credited) in subsequent open-source image-model papers; Google's Imagen 3 paper, for example, explicitly cites Betker et al. as motivation for its own use of synthetic captions.[^4][^22][^49] Third, by routing image generation through ChatGPT, DALL-E 3 was the first widely used image generator whose primary interface was a conversation rather than a prompt box, foreshadowing the multi-turn, refinement-style interface that GPT-4o native image generation made canonical in 2025.[^7]

Beyond technical impact, DALL-E 3's release accelerated industry-wide consolidation around two strategic patterns. The first was bundling image generation into general-purpose chat assistants rather than offering it as a standalone product: in the year after DALL-E 3, Google added image generation to Gemini, Anthropic shipped image input (but not output) in Claude, and xAI added image generation to Grok.[^21] The second pattern was the introduction of C2PA Content Credentials as the de facto provenance standard, which Adobe, Microsoft, Google, and OpenAI all adopted within an 18-month window of DALL-E 3's launch.[^12][^15][^30] DALL-E 3's role as the image generator behind Microsoft Copilot also brought it into Windows 11, whose September 2023 update Microsoft said had more than 150 new features, including Copilot.[^29]

## What are the limitations of DALL-E 3? Criticisms

### Loss of low-level control

Because DALL-E 3 ran behind ChatGPT's rewriter, users frequently complained that they could not get the exact prompt they wrote into the image model. The rewriter often added unrequested details (lighting, mood, ethnicity) and sometimes dropped key constraints, particularly negation ("no humans in the image").[^17][^41] OpenAI added partial workarounds in API documentation (the literal "use the prompt AS-IS" preamble) but never exposed a true disable flag.[^17][^14] The system prompt's hard-coded diversity instructions also produced unexpected outputs (for example, racially diverse historical figures when historical accuracy was the user's intent), echoing a similar controversy that affected Google's Gemini image generator in February 2024.[^39][^46]

### Closed model

Neither DALL-E 3 weights nor the bespoke captioner used for training were released. The Betker et al. report withholds dataset composition, model size, training cost, sampler configuration, and most architectural details of the production model, without stating a reason.[^4] This contrasts with SDXL and FLUX.1, whose weights were openly released, and complicates third-party safety auditing.[^4][^21] The consistency decoder is the only component of the DALL-E 3 stack that OpenAI open-sourced.[^32][^33]

### Provenance fragility

C2PA Content Credentials embedded in DALL-E 3 images survive only round-trips through C2PA-aware tools; screenshots, common social-media uploads, and most re-encoders strip the metadata, limiting its real-world utility as an authentication signal.[^13][^15][^30] OpenAI's own DALL-E 3 detection classifier reportedly degrades on cross-model images (with 5-10% misclassification when shown outputs from other generators) and was offered only to researchers through an access program; in May 2026 OpenAI previewed a separate public verification tool that checks images for its C2PA and SynthID signals.[^15][^30]

### Misuse vectors

The Taylor Swift Designer/Copilot deepfake episode in January 2024 exposed that even policy-compliant generators, when chained behind a permissive front-end with name-aware prompt rewriting, can be coaxed into producing non-consensual sexual imagery of named individuals through prompt obfuscation (for example, by inserting a fictional profession between first and last name).[^20][^47] Shane Jones's December 2023 disclosure indicated that Microsoft had been warned about such vulnerabilities months earlier, and his subsequent letters to U.S. senators and the Washington Attorney General called for DALL-E 3 to be pulled from public use.[^20][^48] OpenAI maintained that the specific bypass paths described did not exceed its policy thresholds, and Microsoft updated Designer's filters in late January 2024.[^20][^47]

### Copyright controversies

OpenAI was not named in the original *Andersen v. Stability AI* class action filed in January 2023, which targeted Stability AI, Midjourney, and DeviantArt over the use of the LAION-5B dataset.[^52] The opt-out form announced alongside DALL-E 3 in September 2023 covered only future models and asked artists to upload each image they wanted excluded, with a description of each piece.[^2][^53] Following the GPT-4o image generation launch in March 2025 and the viral Studio Ghibli style trend (more than 700 million images of all kinds were generated with the new feature in about a week), the French law firm DDG wrote that "it seems clear that DALL-E 3 has been trained on thousands of images taken from Ghibli and Disney films, without authorisation or licence," but the legal status of style imitation under U.S. copyright law remained unsettled.[^34][^54] The broader *New York Times v. OpenAI* lawsuit, filed December 27, 2023 against OpenAI and Microsoft, alleged that OpenAI had unlawfully used Times articles to create AI products.[^55]

### Aesthetic homogenization

Reviewers consistently flagged a recognizable "DALL-E 3 look": soft, painterly, heavily saturated, with smoothed faces. This was attributed to the dominance of descriptive synthetic captions in the training set and to the default `vivid` style; switching to `natural` partially mitigated it but never matched Midjourney v6's aesthetic variety.[^21] The constraint that DALL-E 3 would refuse to generate images in the style of named living artists, while reasonable as a policy choice, also narrowed the achievable style space relative to Stable Diffusion fine-tunes that had no such restriction.[^2][^21]

### Persistent failure modes: faces, hands, fingers, and text

Although DALL-E 3 markedly improved on DALL-E 2 in rendering hands and faces, third-party testing in 2024 and 2025 documented continuing failure modes.[^42][^56] A December 2025 review found that DALL-E 3 rarely produced incorrect finger counts but could still struggle with multiple hands in complex scenes, and another review found that on-image text longer than a few words still degenerated into gibberish.[^42][^56] The first review attributed AI image models' long-standing hand problem to a training-data imbalance: hands occupy a small part of most photographs compared with faces.[^56]

### Throughput and latency

OpenAI's cookbook noted that `hd` quality often added about 10 seconds to generation time, and a third-party pricing guide lists standard DALL-E 3 generations at roughly 8-15 seconds.[^14][^18] Combined with the one-image-per-request limit, this made DALL-E 3 a poor fit for use cases requiring large image volumes, such as synthetic dataset construction for training other models, programmatic ad-creative generation, or game-asset pipelines.[^19][^21]

## Related work

DALL-E 3 sits in a lineage that began with OpenAI's original DALL-E (2021), an autoregressive transformer over discrete image tokens, and continued with DALL-E 2 (2022), a CLIP-conditioned diffusion model with the unCLIP architecture.[^1][^26] Adjacent commercial systems of the same era include Google's [Imagen](https://aiwiki.ai/wiki/imagen) family and later Imagen 3 (which adopted similar caption-augmentation tactics), Stability AI's Stable Diffusion line including SDXL and Stable Diffusion 3, and Black Forest Labs' FLUX.1 models.[^4][^21][^49][^51] On the OpenAI roadmap, DALL-E 3 was eventually superseded by GPT-4o native image generation in 2025, exposed in the API as gpt-image-1.[^7][^8] Related research strands include caption augmentation, [prompt engineering for image generation](https://aiwiki.ai/wiki/prompt_engineering_for_image_generation), and image provenance via the C2PA / content credentials specification.[^22][^12][^15]

## See also

- [DALL-E (original 2021 model and DALL-E 2)](https://aiwiki.ai/wiki/dall-e)
- [gpt-image-1](https://aiwiki.ai/wiki/gpt_image)
- [GPT-4o](https://aiwiki.ai/wiki/gpt_4o)
- [OpenAI](https://aiwiki.ai/wiki/openai)
- [ChatGPT](https://aiwiki.ai/wiki/chatgpt)
- [Sam Altman](https://aiwiki.ai/wiki/sam_altman)
- [Microsoft Copilot](https://aiwiki.ai/wiki/microsoft_copilot)
- [Midjourney](https://aiwiki.ai/wiki/midjourney)
- [SDXL (Stable Diffusion XL)](https://aiwiki.ai/wiki/sdxl)
- [Stable Diffusion](https://aiwiki.ai/wiki/stable_diffusion)
- [Stable Diffusion 3](https://aiwiki.ai/wiki/stable_diffusion_3)
- [Imagen](https://aiwiki.ai/wiki/imagen)
- [Imagen 3](https://aiwiki.ai/wiki/imagen_3)
- [FLUX.1](https://aiwiki.ai/wiki/flux_1)
- [Adobe Firefly](https://aiwiki.ai/wiki/adobe_firefly)
- [Latent diffusion model](https://aiwiki.ai/wiki/latent_diffusion)
- [Diffusion model](https://aiwiki.ai/wiki/diffusion_model)
- [Whisper](https://aiwiki.ai/wiki/whisper)
- [SynthID](https://aiwiki.ai/wiki/synthid)
- [AI watermarking](https://aiwiki.ai/wiki/ai_watermarking)
- [Synthetic data](https://aiwiki.ai/wiki/synthetic_data)
- [Deepfake](https://aiwiki.ai/wiki/deepfake)
- [Prompt engineering for image generation](https://aiwiki.ai/wiki/prompt_engineering_for_image_generation)
- [Text-to-image models](https://aiwiki.ai/wiki/text-to-image_models)
- [AI image generation](https://aiwiki.ai/wiki/ai_image_generation)

## References

[^1]: OpenAI, "DALL·E 3", OpenAI Blog, 2023-09-20. https://openai.com/index/dall-e-3/. Accessed 2026-05-24.
[^2]: Kyle Wiggers, "OpenAI unveils DALL-E 3, allows artists to opt out of training", TechCrunch, 2023-09-20. https://techcrunch.com/2023/09/20/openai-unveils-dall-e-3-allows-artists-to-opt-out-of-training/. Accessed 2026-05-24.
[^3]: Encord, "OpenAI's DALL-E 3 Explained: Generate Images with ChatGPT", Encord Blog, 2023-09-26. https://encord.com/blog/openai-dall-e-3-what-we-know-so-far/. Accessed 2026-05-24.
[^4]: James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, Wesam Manassra, Prafulla Dhariwal, Casey Chu, Yunxin Jiao, Aditya Ramesh, "Improving Image Generation with Better Captions", OpenAI Technical Report, 2023. https://cdn.openai.com/papers/dall-e-3.pdf. Accessed 2026-05-24.
[^5]: Sarah Perez, "Microsoft Bing to gain more personalized answers, support for DALL-E 3 and watermarked AI images", TechCrunch, 2023-09-22. https://techcrunch.com/2023/09/22/microsoft-bing-to-gain-more-personalized-answers-support-for-dalle-e-3-and-watermarked-ai-images/. Accessed 2026-05-24.
[^6]: Microsoft, "AI Art Improvements with DALL-E 3 in Microsoft Copilot", Microsoft Bing, 2023-10. https://www.microsoft.com/en-us/bing/do-more-with-ai/image-creator-improvements-dall-e-3. Accessed 2026-05-24.
[^7]: OpenAI, "Introducing 4o Image Generation", OpenAI Blog, 2025-03-25. https://openai.com/index/introducing-4o-image-generation/. Accessed 2026-05-24.
[^8]: OpenAI, "Introducing our latest image generation model in the API", OpenAI Blog, 2025-04-23. https://openai.com/index/image-generation-api/. Accessed 2026-05-24.
[^9]: OpenAI, "Deprecations", OpenAI API Documentation, 2025-11-14. https://platform.openai.com/docs/deprecations. Accessed 2026-05-24.
[^10]: OpenAI, "DALL·E 3 is now available in ChatGPT Plus and Enterprise", OpenAI Blog, 2023-10-19. https://openai.com/index/dall-e-3-is-now-available-in-chatgpt-plus-and-enterprise/. Accessed 2026-05-24.
[^11]: OpenAI, "DALL·E 3 Model", OpenAI API Documentation, 2024. https://developers.openai.com/api/docs/models/dall-e-3. Accessed 2026-05-24.
[^12]: OpenAI, "C2PA in ChatGPT Images", OpenAI Help Center, 2024-02-06. https://help.openai.com/en/articles/8912793-c2pa-in-chatgpt-images. Accessed 2026-05-24.
[^13]: PetaPixel, "AI Images Generated on DALL-E Now Contain the Content Authenticity Tag", PetaPixel, 2024-02-08. https://petapixel.com/2024/02/08/ai-images-generated-on-dall-e-now-contain-the-content-authenticity-tag/. Accessed 2026-05-24.
[^14]: OpenAI, "What's new with DALL·E 3?", OpenAI Cookbook, 2023-11. https://cookbook.openai.com/articles/what_is_new_with_dalle_3. Accessed 2026-05-24.
[^15]: OpenAI, "Advancing content provenance for a safer, more transparent AI ecosystem", OpenAI Blog, 2026-05-19. https://openai.com/index/advancing-content-provenance/. Accessed 2026-05-24.
[^16]: Genra.ai, "OpenAI Retires Its Image Models on May 12: What Replaces Them", Genra Blog, 2025-11. https://genra.ai/blog/dall-e-retired-may-2026-what-replaces-it. Accessed 2026-05-24.
[^17]: Matthias Bastian, "ChatGPT's automatic prompt rewriting reduces DALL-E 3's performance, study finds", The Decoder, 2024-08-10. https://the-decoder.com/chatgpts-automatic-prompt-rewriting-reduces-dall-e-3s-performance-study-finds/. Accessed 2026-05-24.
[^18]: TokenMix, "DALL-E API Pricing 2026: $0.04-$0.12/Image vs Flux $0.03", TokenMix Blog, 2026-01. https://tokenmix.ai/blog/dall-e-api-pricing. Accessed 2026-05-24.
[^19]: OpenAI, "What are the rate limits for Image Generation?", OpenAI Help Center, 2024. https://help.openai.com/en/articles/6696591-what-s-the-rate-limit-for-the-dall-e-api. Accessed 2026-05-24.
[^20]: Todd Bishop, "Microsoft AI engineer says company thwarted attempt to expose DALL-E 3 safety problems", GeekWire, 2024-01-30. https://www.geekwire.com/2024/microsoft-ai-engineer-says-company-thwarted-attempt-expose-dall-e-3-safety-problem/. Accessed 2026-05-24.
[^21]: Super Allen, "AI Image Generation 2025: DALL-E 3 vs Midjourney v6 vs Stable Diffusion XL", Super Allen Blog, 2025. https://www.superallen.org/blog/2025/ai-image-generation-2025. Accessed 2026-05-24.
[^22]: The Batch (DeepLearning.AI), "OpenAI Researchers Improved Text-to-Image Prompt Following with Generated Captions", DeepLearning.AI, 2023-10. https://www.deeplearning.ai/the-batch/openai-researchers-improved-text-to-image-prompt-following-with-generated-captions/. Accessed 2026-05-24.
[^23]: Microsoft Bing Blog, "DALL-E 3 now available in Bing Chat and Bing.com/create, for free!", Bing Blogs, 2023-10-03. https://blogs.bing.com/search/october-2023/DALL-E-3-now-available-in-Bing-Chat-and-Bing-com-create-for-free. Accessed 2026-05-24.
[^24]: Kyle Wiggers, "OpenAI launches DALL-E 3 API, new text-to-speech models", TechCrunch, 2023-11-06. https://techcrunch.com/2023/11/06/openai-launches-dall-e-3-api-new-text-to-speech-models/. Accessed 2026-05-24.
[^25]: Emre Çıtak, "Free ChatGPT Users Can Now Use DALL-E 3", Dataconomy, 2024-08-09. https://dataconomy.com/2024/08/09/dall-e-3-free-chatgpt/. Accessed 2026-05-24.
[^26]: Wikipedia, "DALL-E", Wikipedia, 2026-05. https://en.wikipedia.org/wiki/DALL-E. Accessed 2026-05-24.
[^27]: Carl Franzen, "OpenAI brings DALL-E 3 image generator to ChatGPT for Enterprise, teases classifier", VentureBeat, 2023-10-19. https://venturebeat.com/ai/openai-brings-dall-e-3-image-generator-to-chatgpt-for-enterprise-teases-classifier/. Accessed 2026-05-24.
[^28]: InfoQ, "OpenAI Announces ChatGPT Voice and Image Features", InfoQ, 2023-10. https://www.infoq.com/news/2023/10/chatgpt-voice-image/. Accessed 2026-05-24.
[^29]: Yusuf Mehdi, "Announcing Microsoft Copilot, your everyday AI companion", Microsoft Blog, 2023-09-21. https://blogs.microsoft.com/blog/2023/09/21/announcing-microsoft-copilot-your-everyday-ai-companion/. Accessed 2026-05-24.
[^30]: OpenAI, "Understanding the source of what we see and hear online", OpenAI Blog, 2024-05-07. https://openai.com/index/understanding-the-source-of-what-we-see-and-hear-online/. Accessed 2026-05-24.
[^31]: Google DeepMind, "SynthID: A toolkit for watermarking AI-generated content", DeepMind, 2024. https://deepmind.google/models/synthid/. Accessed 2026-05-24.
[^32]: OpenAI, "openai/consistencydecoder: Consistency Distilled Diff VAE", GitHub Repository, 2023-11. https://github.com/openai/consistencydecoder. Accessed 2026-05-24.
[^33]: Hugging Face, "Consistency Decoder VAE", Hugging Face Diffusers Documentation, 2023-11. https://huggingface.co/docs/diffusers/en/api/models/consistency_decoder_vae. Accessed 2026-05-24.
[^34]: Amanda Silberling, "Studio Ghibli hasn't commented on OpenAI's onslaught of AI copies, but the fan subreddit has", TechCrunch, 2025-04-03. https://techcrunch.com/2025/04/03/studio-ghibli-hasnt-commented-on-openais-onslaught-of-ai-copies-but-the-fan-subreddit-has/. Accessed 2026-05-24.
[^35]: Rebecca Bellan, "OpenAI continues on its 'code red' warpath with new image generation model", TechCrunch, 2025-12-16. https://techcrunch.com/2025/12/16/openai-continues-on-its-code-red-warpath-with-new-image-generation-model/. Accessed 2026-05-24.
[^36]: OpenAI Developer Community, "OpenAI is making a huge mistake by deprecating DALL-E-3", Community Forum, 2025-11. https://community.openai.com/t/openai-is-making-a-huge-mistake-by-deprecating-dall-e-3/1367228. Accessed 2026-05-24.
[^37]: Shreyansh Singh, "Paper Summary #12 - Image Recaptioning in DALL-E 3", Personal Blog, 2024-02-18. https://shreyansh26.github.io/post/2024-02-18_dalle3_image_recaptioner/. Accessed 2026-05-24.
[^38]: AiCoin, "Interpreting OpenAI's most powerful text-to-image model - DALL·E 3", AiCoin, 2023. https://www.aicoin.com/en/article/371500. Accessed 2026-05-24.
[^39]: spdustin, "ChatGPT-AutoExpert: DALL-E system prompt", GitHub Repository, 2023-10. https://github.com/spdustin/ChatGPT-AutoExpert/blob/main/_system-prompts/dall-e.md. Accessed 2026-05-24.
[^40]: Matthias Bastian, "DALL-E 3's system prompt reveals OpenAI's rules for generative image AI", The Decoder, 2023-10. https://the-decoder.com/dall-e-3s-system-prompt-reveals-openais-rules-for-generative-image-ai/. Accessed 2026-05-24.
[^41]: The Decoder, "Prompt transformation makes ChatGPT OpenAI's covert moderator for DALL-E 3", The Decoder, 2023-10. https://the-decoder.com/prompt-transformation-makes-chatgpt-openais-covert-moderator-for-dall-e-3/. Accessed 2026-05-24.
[^42]: Opace, "ChatGPT Image Generation: GPT-4o vs Dall-E 3 & Other Models", Opace Blog, 2025. https://opace.agency/blog/chatgpt-image-generation-gpt-4o-vs-dall-e-3-and-others. Accessed 2026-05-24.
[^43]: OpenAI Developer Community, "DALLE3 Inpainting: Editing your images with DALL·E", Community Forum, 2024. https://community.openai.com/t/dalle3-inpainting-editing-your-images-with-dall-e/705477. Accessed 2026-05-24.
[^44]: FavTutor, "DALL-E 3 Inpainting Update: New Features coming to their Editor", FavTutor Blog, 2024. https://favtutor.com/articles/dall-e-3-editor-interface-inpainting/. Accessed 2026-05-24.
[^45]: OpenAI, "DALL·E 3 System Card", OpenAI Technical Report, 2023-10-03. https://cdn.openai.com/papers/DALL_E_3_System_Card.pdf. Accessed 2026-05-24.
[^46]: OpenAI, "Reducing bias and improving safety in DALL·E 2", OpenAI Blog, 2022. https://openai.com/index/reducing-bias-and-improving-safety-in-dall-e-2/. Accessed 2026-05-24.
[^47]: Amanda Silberling, "The Taylor Swift deepfake debacle was frustratingly preventable", TechCrunch, 2024-01-30. https://techcrunch.com/2024/01/30/the-taylor-swift-deepfake-debacle-was-frustratingly-preventable/. Accessed 2026-05-24.
[^48]: AIWI (AI Whistleblower Initiative), "Case Study: Shane Jones", AIWI.org, 2024. https://aiwi.org/case-studies/shane-jones/. Accessed 2026-05-24.
[^49]: The Batch (DeepLearning.AI), "Google's Imagen 3 Outperforms Rivals in Text-to-Image Benchmarks", DeepLearning.AI, 2024. https://www.deeplearning.ai/the-batch/googles-imagen-3-outperforms-rivals-in-text-to-image-benchmarks/. Accessed 2026-05-24.
[^50]: Unite.AI, "Google Imagen 3 vs. The Competition: A New Benchmark in Text-to-Image Models", Unite.AI, 2024. https://www.unite.ai/google-imagen-3-vs-the-competition-a-new-benchmark-in-text-to-image-models/. Accessed 2026-05-24.
[^51]: Hugging Face, "Stable Diffusion 3", Hugging Face Diffusers Documentation, 2024. https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_3. Accessed 2026-05-24.
[^52]: Christopher Hayman, "Andersen v. Stability AI: The Landmark Case Unpacking the Copyright Risks of AI Image Generators", NYU Journal of Intellectual Property & Entertainment Law, 2024. https://jipel.law.nyu.edu/andersen-v-stability-ai-the-landmark-case-unpacking-the-copyright-risks-of-ai-image-generators/. Accessed 2026-05-24.
[^53]: Bloomberg Law, "Dall-E 3 Is Stoking an Artist Revolt Against AI Scraping", Bloomberg Law News, 2023. https://news.bloomberglaw.com/artificial-intelligence/dall-e-3-is-stoking-an-artist-revolt-against-ai-scraping. Accessed 2026-05-24.
[^54]: DDG (De Gaulle Fleurance), "AI & Copyright: Legal Insights from the Ghibli-Style DALL·E 3 Buzz", DDG, 2025. https://www.ddg.fr/actualite/ai-intellectual-property-what-lessons-can-be-learned-from-the-global-buzz-surrounding-the-generation-of-videos-in-the-style-of-studio-openai-ghibli. Accessed 2026-05-24.
[^55]: Cordelia Smith, "How a New York Times copyright lawsuit against OpenAI could potentially transform how AI and copyright work", The Conversation, 2024-01. https://theconversation.com/how-a-new-york-times-copyright-lawsuit-against-openai-could-potentially-transform-how-ai-and-copyright-work-221059. Accessed 2026-05-24.
[^56]: Vertu, "Has AI Fixed the Hand Problem? Midjourney V7, DALL-E 3 & SDXL Accuracy (2025 Check)", Vertu AI Tools, 2025. https://vertu.com/ai-tools/the-2025-reality-check-on-ai-generated-hands-and-fingers/. Accessed 2026-05-24.
[^57]: Graham Barlow, "'Our GPUs are melting' - OpenAI puts limits on image creation and delays rollout to free accounts", TechRadar, 2025-03-28. https://www.techradar.com/computing/artificial-intelligence/our-gpus-are-melting-openai-puts-limits-on-image-creation-and-delays-rollout-to-free-accounts. Accessed 2026-11-02.
[^58]: Imagen-Team-Google, "Imagen 3", arXiv:2408.07009, 2024. https://arxiv.org/abs/2408.07009. Accessed 2026-11-02.

