Adobe Unveils Its 8-Step Method for AI Image Prompts

Key Takeaways

1Adobe presents a structured eight-element method for writing AI image prompts

2Natural language and the use of over 100 languages, including French, are emphasized

3It is recommended to iterate gradually and test the portability of prompts between models

Why it matters : Structuring your queries and adapting your approach allows for more relevant and varied results with AI image generators.

Adobe offers an eight-point framework for crafting image prompts, while emphasizing that no formula is universal. Key advice includes using natural language, cross-testing between models, and cautious iterations to avoid over-complexity.

Adobe recommends using simple sentences and negative prompts

The writing of requests now favors natural language over bracket-based syntax or punctuation. A direct formulation is recommended, while still utilizing negative prompts to exclude elements. Firefly supports over 100 languages, including French, and it is advised to alternate between English and other languages depending on the desired outcomes.

Eight components to include for framing the image

Adobe outlines in a guide a method for organizing a prompt into eight components: image type, subject (such as features, expressions, or architectural style), action, viewpoint, lighting, background, color, and style. This guide specifies that this organization leads to higher quality renders. A detailed example illustrates how to combine these parameters to accurately describe the scene, subject, and atmosphere.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Proceed with brief iterations and avoid over-complexity

It is recommended to start with a simple request and then gradually add details, as excessive complexity can produce undesirable effects. Long prompts found online can serve as a learning base, provided they are simplified to understand the models' reactions. Resubmitting the same request can yield different results, as models are sensitive to variations.

Adapt the prompt to the engine and test its portability

Firefly brings together its models and those of partners in a single interface, raising the question of prompt portability. The analogy of the sponge cake describes a common base that needs to be adapted according to the strengths of each engine. Hence, it is beneficial to retry the same prompt across multiple models to compare the outputs. There is no perfect prompt, Olivier Huard reminds us, advocating instead for best practices to achieve satisfactory results more quickly.