ChatGPT does not natively generate images. However, a roundabout way exists to still use it in this way. This is how to generate images in ChatGPT.
In a few months, the OpenAI chatbot has seen some improvements : using ChatGPT well requires some tips and several features are available. But for the moment, without a ChatGPT Plus subscription, it is impossible to generate images : the paid version allows the use of plugins and access to Dall-E, the AI model for generating images from OpenAI. Fortunately, a method allows you to generate images without having to leave the ChatGPT interface. Let’s be clear : this generation is not made by ChatGPT nor by OpenAI, but by an external tool.
The right "prompt" (request) to ask ChatGPT
We owe this "prompt" to Chase Lean, presenting himself on Twitter as an "AI enthusiast". This is a text request that we send in ChatGPT. It somehow bypasses the limitations put in place by ChatGPT. If we send the command directly, ChatGPT indicates this : "I'm sorry, but as a text AI, I can't see or access the images directly. However, if you describe the image to me, I will do my best to help you with any questions or discussions related to it".
But by writing to him just before " You are an image request generator. First, ask me for a description of an image and help me fill in the following. Then display the completed query", we trick ChatGPT into thinking it's an image generator, so we can get around these limitations of OpenAI.
The rest of the prompt is a request to Pollinations, an image generation site, based on the Stable Diffusion generation model, available in open source, and significantly less efficient than Midjourney.
ChatGPT tweaks image generation
In fact, it is not totally you who is requesting the generation of the image, but indeed ChatGPT. What ChatGPT asks you for is a description of the image you want. Behind this, ChatGPT completes your "prompt". What saves time, since it is the chatbot that develops the description of the desired image to have a more convincing result.
To generate images in ChatGPT, copy and paste this text into the chatbot :
You are an image prompt generator. First, ask me for a description of an image, and help me fill in the following. Then, output the completed prompt. ![Image] (https://image.pollinations.ai/prompt/{description}), where {description} = {sceneDetailed},%20{adjective1},%20{charactersDetailed},%20{adjective2},% 20{visualStyle1},%20{visualStyle2},%20{visualStyle3},%20{genre}
Next, ChatGPT will ask you to give a description of the image you want to generate. Then, just wait a few seconds for the image to appear. Just below, we can see what was the request generated by ChatGPT. It is interesting to observe that the tool details the characters that we want to place, qualifies the actions that take place, sets the tone of the image style, adds atmospheres, etc.
For clarification, Chase Lean recommends using GPT-3.5 over GPT-4 (for ChatGPT Plus subscribers), since it may not work with the latest version of the language model. In addition, the chatbot may ask you to specify a style, an adjective, etc. If you don't feel like doing it, you can just ask him, "Can you fill it out for me?".
Comments
Post a Comment