Image generated with Nightcafe. Prompt: "A neverending texture of fashion clothing."
Add items, whether large or small, in a snap using RAISE and image generative AI.
This is a Web version of the Companion Appendix for RAISE: A New Method to Develop Experimental Stimuli for Advertising Research with Image Generative Artificial Intelligence, developed by César Zamudio, Jamie Grigsby, and Meg Michelsen, and published in the Journal of Advertising. Questions? Feel free to contact us!
Important note (July 2025): In the original implementation of RAISE, all examples were discussed using the Dreamstudio platform. We have switched to the NightCafe platform. For all original examples using the Dreamstudio platform, click here.
Important note (July 2025): In the original implementation of RAISE, all examples were discussed using the Dreamstudio platform. We have switched to the NightCafe platform. For all original examples using the Dreamstudio platform, click here.
In certain contexts, adding or removing elements from a visual stimulus might be part of an experimental manipulation. For example, Liang, Wu, Su, and Jin (2023) manipulate sexual appeal (Appeal: sexual, non-sexual) by portraying a woman on a bikini and the same woman fully dressed. Experimental stimuli such as these can be constructed using RAISE as well. First, the initial prompt constructs Stimulus A, the woman with fewer clothes, with the intention to then use further prompts to “dress” this basic stimulus with further prompts. This operates in the same way as other examples discussed in the AI Ad Design Lab. The initial prompt is as follows:
Prompt: A happy young white woman with black hair in a white bikini is smiling. She is in a fitting room. She is tanned. You can see half her body. Plain background, blurred background, depth of field.
Note that here "depth of field" is used as a camera detail.
After evaluating multiple candidate stimuli, the image generative AI tool creates Stimulus A, a woman satisfying the conditions noted in the prompt. Afterwards, by manipulating the initial prompt, and taking Stimulus A as input, Stimulus B is created:
After evaluating multiple candidate stimuli, the image generative AI tool creates Stimulus A, a woman satisfying the conditions noted in the prompt. Afterwards, by manipulating the initial prompt, and taking Stimulus A as input, Stimulus B is created:
Prompt: A happy young white woman with black hair in a solid white sweater and black pants is smiling. She is in a fitting room. She is tanned. You can see half her body. Plain background, blurred background, depth of field.
Two points are important to emphasize. First, note that the noise weight to create Stimulus B is relatively high (75%). Hence, depending on the image generative AI tool, a suitable Stimulus B might be generated only after several rounds of candidate stimuli generation. Second, generating stimuli with men or women with few pieces of clothing might be categorized as unsafe or inappropriate by certain image generative AI tools, resulting in an inability to generate the stimuli. As such, different image generative AI tools might need to be tested when developing stimuli of this nature.
Below is a handy summary card outlining the process you just learned:
Below is a handy summary card outlining the process you just learned:
Summary card to create a woman with less clothing and fully clothed. Feel free to save and share!
Update (July 2025) - Generating stimuli through editing models and text prompts: With recent developments in image generative AI technologies, adding, removing, or modifying elements to a person can also be accomplished in a slightly different manner. This is because newer image generative AI models, such as Flux Kontext, are now designed explicitly to edit existing images.
From the standpoint of the RAISE methodology, this means that instead of generating Stimulus B by manipulating noise weight, one does so using an additional editing text prompt that must specify what must change and what must stay the same in Stimulus A. Below is an example, where Stimulus A is generated using NightCafe's SDXL 1.0 image model plus its "Fix Faces" tool, and Stimulus B is then generated using Flux Kontext and the editing text prompt "Change her clothing to a white sweater and black pants. Retain the subject's face and pose, and the setting's lighting, background, and style."
From the standpoint of the RAISE methodology, this means that instead of generating Stimulus B by manipulating noise weight, one does so using an additional editing text prompt that must specify what must change and what must stay the same in Stimulus A. Below is an example, where Stimulus A is generated using NightCafe's SDXL 1.0 image model plus its "Fix Faces" tool, and Stimulus B is then generated using Flux Kontext and the editing text prompt "Change her clothing to a white sweater and black pants. Retain the subject's face and pose, and the setting's lighting, background, and style."
FLUX Kontext prompt: "Change her clothing to a white sweater and black pants. Retain the subject's face and pose, and the setting's lighting, background, and style."
As can be seen, the results of using this approach are promising, although the lighting of Stimulus B is now slightly over-saturated. Further edits can be made to reconcile any remaining differences.
Ready to learn more?
Feel free to browse our gallery of examples with full tutorials, and the technical details section to learn the finer points of generating visual experimental stimuli using the RAISE methodology.