What a beautiful image, it’s a pity about the pencil lines on the paintwork/glass of the car. How would you compose a prompt to resolve this?
I had already deleted this chat, but I only need to let the symbiotic relationship between me and my 1995 Microsoft Natural Keyboard acting as an extension of my body let my thoughts flow into the shaping of this image in the underlying mechanisms of the image creator, with every brush stroke of word representing something meaningful to the AI that does not artificially place elements in the depiction simply due to the strength of their underlying embedding weights…
I’ll address your concerns by continuing on further prompting language, where the only expense is our investment of time, letting the AI see our vision instead of filling in an interpretation of our desires.
dalle wide image request: “In this highly-detailed picture of a futuristic prototype automobile to be revealed at the 2027 auto show, we also see the edge-to-edge texture of the fine paper background that the sketch was made on, and the high quality of finely-textured pencil mark strokes outlining the figure of the car, giving the unseen impression that a skilled and masterful artist had drawn the picture, even when limited to using a pencil drawing style. This sleek automobile design is depicted with smooth shaded pencil gradients that appear almost realistic, with smooth reflective glass with specular highlights, and body panels of shapely metal that take on the lighting around the vehicle. The shape of the vehicle gives the impression of a supercar meant for high speeds, with aggressive grille and and cowl hood, a slim waistline with ground effects, and natural rubber tires with large spoked wheels of aluminum that also pick up the shadow and light. A subtle spoiler at the back of the vehicle gives its final racing stance. The car prototype is seen from a low isometric perspective, and the background is rendered in pointillistic style by the skilled pencil artist to give a fine texture and impression of high fidelity that makes this vehicle pop out of the page. Slight touches of watercolor to enhance the appearance do not detract from the quality of the origin being a pencil sketch taking many days to realize.”
An appreciation of the textures within described by prompt:
1 Like
I tried the same thing, this shows how LLMs have difficulty discerning what we want and humans have difficulty communicating what we want.
L
And that AI that gets in the way, rewriting via random token selection and dubious instruction, is bad. I didn’t jailbreak and override before, just to make it within expectations. With “DALLE” GPT you get a different quality of following the tool instructions, also.
Prompt sent after rewriting, and you can compare what yours sent:
In a highly detailed picture, a futuristic prototype automobile is revealed, designed for the 2027 auto show. The background showcases the edge-to-edge texture of fine paper, enhancing the skilled pencil drawing. The vehicle, depicted with smooth shaded pencil gradients, features realistic reflections on its glass surfaces and metal body panels that respond dynamically to lighting. Its design hints at high speeds, with an aggressive grille, cowl hood, slim waistline with ground effects, and natural rubber tires on large spoked aluminum wheels that capture shadows and light. A subtle spoiler adds to the racing stance. Viewed from a low isometric angle, the background is rendered in pointillistic style, providing a high fidelity texture that makes the vehicle stand out. Slight watercolor touches complement the predominantly pencil sketch, showcasing the artist’s mastery in creating a realistic, detailed depiction that took days to realize.
I instead let my GPT loose on sending the text unaltered and relentlessly. Not a misstep to be seen.
Your GPT? So you built a custom version? Can I have the link, please? I’d love to experiment. Thanks.
Have you guys played with this and does anyone know if it will be available via the api: Editing your images with DALL·E | OpenAI Help Center https://help.openai.com/en/articles/9055440-editing-your-images-with-dall-e