How Amorepacific built HERA's virtual holiday campaign with Midjourney
Amorepacific Creatives for the HERA brand
Amorepacific's design team made virtual models for HERA's holiday campaign: keywords from its internal assistant Amore GPT, models generated in Midjourney, upscaling in Magnific and finishing in Photoshop AI. The team says this cut the time and cost of the campaign.
Story written by: the company in the story
The problem
Finding suitable models and shooting a seasonal beauty campaign would normally take considerable time and cost. The team wanted virtual models that express HERA's free-spirited brand image, in several ethnicities, for its holiday campaign. Some AI images did not match the intended vision, so the work also needed careful checking and refining.
How it was built
- 1Choose a season keyword with Amore GPTAmore GPT
- 2Generate the virtual model in MidjourneyMidjourney
- 3Retouch the model in PhotoshopPhotoshop
- 4Rebuild the prompt with Midjourney's Describe featureMidjourney
- 5Upscale in Magnific and finish in Photoshop AIMagnific
- 6Create multi-ethnic variants with Vary (Region)Midjourney
Tool step Person step
1. Choose a season keyword with Amore GPT
The team asked Amore GPT, its internal assistant, for keywords to capture HERA's holiday season and picked 'Winter Awakening' from ten suggestions. They then planned the look: a classic neutral tone, glowing skin, shimmering glitter and silver jewelry.
ToolAmore GPT
Check: The team judges which of the ten keywords best reflects the brand's renewed, modern feel.
2. Generate the virtual model in Midjourney
They prompted Midjourney with the model's physical attributes, such as age, gender and ethnicity, then used its Vary feature to get variations in hairstyles and camera angles.
ToolMidjourney
Check: They review the generated variations and keep the ones that fit the brand image.
3. Retouch the model in Photoshop
The generated AI images were refined and retouched in Photoshop before being used further.
ToolPhotoshop
Check: They refine any parts of the image that are not aligned with the intended vision.
4. Rebuild the prompt with Midjourney's Describe feature
Midjourney's --cref function often gave inconsistent results, so the team uploaded reference images to Midjourney's Describe feature and built new prompts from selected extracted keywords. Shortening and simplifying the prompt worked better than including every detail.
ToolMidjourney
Check: They select which extracted keywords to keep and drop the rest, then compare results as they shorten the prompt.
5. Upscale in Magnific and finish in Photoshop AI
After generating the make-up look images in Midjourney, the team upscaled them using Magnific and enhanced fine details using Photoshop AI.
ToolMagnific
Check: They check the fine details of the upscaled image while enhancing it.
6. Create multi-ethnic variants with Vary (Region)
Based on the campaign image, the team developed models of several ethnicities using Midjourney's Vary (Region) feature and Photoshop, with settings such as --sref and --ar 4:5 for bust shots and close-ups of lips and eyes.
ToolMidjourney
Check: They compare each variant with the intended look and spend extra time refining the ones that miss it.
Before and after
Before
Finding suitable models and expressing the desired look would normally require considerable time and cost.
After
The team says AI tools delivered a high level of completeness in the make-up looks and overall concept, and reduced the time and cost of producing the seasonal campaign (self-reported, no figures given). source
What made it work
- Dropping Midjourney's --cref when it gave inconsistent results, and using the Describe feature to pull keywords from reference images instead.
- Shortening and simplifying prompts instead of including every detail.
- Combining several AI tools, each for the part it does best.
- Refining AI images by hand in Photoshop and Photoshop AI.
- Improving prompt-writing techniques through trial and error.
Follow along
Practice steps suggested by Ai Siftly, based on the account above. We have not run them ourselves.
Write a small brief
Have ready: a photo of a product you own or may use, one audience and one mood, and a written list of product details that must not change. Do not copy the campaign's model likeness.
Example to copy: My tea packaging; gift buyers; calm morning; soft daylight. Keep package colour/shape/label accurate.
You should get: Brief and owned reference image.
Check: Separate creative background choices from factual product claims.
Pick one picture idea
Ask an AI chat for three ways to arrange the product, background and lighting, choose one, and sketch it on paper or in an image tool you already know. Check any AI generation cost before using it.
Example to copy: Suggest three still-life compositions for [brief]. Do not change the product or invent label text.
You should get: One selected concept.
Check: No unsupported product-benefit claims or unlicensed assets.
Check the draft against the real photograph
Compare the draft with your real product photograph and fix shape, colour, logo, words and packaging in an editor you know. If you cannot correct them, keep the image as a rough sketch, not an advert.
You should get: One corrected master image.
Check: Zoom in to inspect labels, shadows and accidental artifacts.
Make controlled variants
Create three variants changing only background, crop or lighting, keeping the product itself fixed.
Example to copy: Preserve the reference product. Change only [variable]; no added text.
You should get: Master plus variant01/02/03.
Check: The same product is recognisable in every frame.
Compare the three versions
Put the images side by side in a document or slide, label what changed in each, and choose a direction. Brand approval and rights checks are still needed before commercial publication.
You should get: A review sheet and chosen image.
Check: Inspect the actual saved files, the intended aspect ratio and permissions.
Try it yourself
Find a matching workflow in the workflow finder. Search words: A I v i r t u a l m o d e l b e a u t y c a m p a i g n.
Sources and notes
- Winter Awakening (Amorepacific Creatives, )
- Results are self-reported by Amorepacific Creatives, with no figures.
- The page text shows 'Jan 22, 2024' but the page address and image upload paths point to 2025, so the publish date is not certain; the campaign described is the 2024 Virtual Holiday Season.
- The Midjourney settings shown (such as --sref and --ar 4:5) are historical and may differ from current versions.
- The follow-along steps are Ai Siftly's suggested reconstruction for practice, not the team's exact process; Ai Siftly did not run this workflow.
- Use a product image you own or may use, and do not copy the campaign's model likeness.
Company's own design site. The page text gives Jan 22, 2024; its URL and image paths show 2025. Qualitative outcome only. Historical tool behaviour and commands are as documented at the source date; current versions may differ. Record dated 2026-10-27.