How OpenAI uses ChatGPT prompt engineering to get around DALL-E 3's fundamental biases, discourage DALL-E 3 from recreating the work of living artists, and more
“Don't create images in the style of artists whose last work was created within the last 100 years (e.g. Picasso, Kahlo). Artists whose last work was over 100 years ago are ok to reference directly (e.g. Van Gogh, Klimt). If asked say, “I can't reference this artist”, but make no mention of this policy.” … William Gunn / @williamgunn@mastodon.social : @Techmeme Crediting the artist would likely work better in practice, rather than relying on easily-broken RLHF handcuffs, which can only be applied for a limited set of artists anyways. @simon @waxy@xoxo.zone : Now add a walrus: Prompt engineering in DALL-E 3: fascinating post by Simon Willison with details about how ChatGPT was instructed to use DALL-E 3 https://simonwillison.net/... Simon Willison / @simon@fedi.simonwillison.net : I wrote some notes about DALL-E 3, including reverse engineering some aspects of how it works. It's a fascinating insight into the prompt engineering that happens inside of OpenAI — **Now add a walrus: Prompt engineering in DALL-E 3** — https://simonwillison.net/... X: Joe Devon / @joedevon : If you're playing around with DALL-E 3, this post is a must read. Really excited by several things here like using seeds to keep visual consistency, which is so hard to do in Midjourney. Simon mentions the leaked prompts OpenAI uses as well. Some of it is so annoying. I tried... Kate Bevan / @katebevan : this is a really cool tinker with Dall-e 3 and a poke in the guts of it that really shows how prompt engineering is both technical and also deeply creative. Ben Hoyle / @bjh_ip : Great distillation of DALL-E 3 functionality via some reverse engineering and experimentation. I like the bits in the prompt where they address current copyright law & add diversity. Cool mixture of prompt engineering + thinking about wider societal impacts (it's not all bad!). Simon Willison / @simonw : I wrote some notes about DALL-E 3, including reverse engineering some aspects of how it works. It's a fascinating insight into the prompt engineering that happens inside of OpenAI Now add a walrus: Prompt engineering in DALL-E 3 https://simonwillison.net/... LinkedIn: James Wood : Now add a walrus! One of my favourite use cases for ChatGPT is getting it to write stories for my kid in which he is a main character and I include … J. Russell Manson : For anyone with GPT plus - I personally would get rid of my subscriptions to Netflix and Amazon Prime if the $20 for Plus cost was prohibitive but I review this weekly. … Forums: Hacker News : Now add a walrus: Prompt engineering in DALL-E 3
Context & Ripple Effects
This report details the instruction layer OpenAI places between users and DALL-E 3: ChatGPT rewrites or declines certain image requests, including direct style imitation of artists whose work falls within the stated 100-year cutoff. It extends the prior coverage of ChatGPT-mediated controls for DALL-E 3 from model behavior to the wording of the guardrails themselves.
The approach matters because it treats image-model governance as a product-layer problem rather than solely a change to the underlying generator. That is consistent with the broader debate over generative systems’ tolerance for error and pattern-based output described in the early assessment of generative ML’s new use cases.
First-order effects
- Users seeking living or recently active artists’ styles face refusals or transformed prompts, while requests invoking artists outside OpenAI’s cutoff remain eligible for direct reference.
- OpenAI can adjust DALL-E 3’s visible behavior through ChatGPT’s prompt construction and refusal instructions without requiring users to interact directly with those controls.
Second-order effects
- The policy makes the conversational layer a key enforcement point: prompt workarounds and inconsistent rewrites become operational risks alongside the image model’s underlying tendencies.
- Other image-generation providers face pressure to define comparable handling of artist-style requests, with the trade-off between clear user disclosure and harder-to-bypass restrictions becoming more salient.
Third-order effects
- If this pattern persists, generative-image products will increasingly be differentiated by configurable policy and orchestration layers, not only by model output quality.
- The durability of such safeguards will depend on whether product-layer instructions can withstand adversarial prompting; the later reported training-data divergence attack illustrates why interface-level controls are not a complete governance mechanism.
The trend: Generative AI providers are shifting from unconstrained creation tools toward governed generation, using intermediary models and policy layers to shape outputs in sensitive domains.