Advanced prompting techniques

Prompt patterns beyond the defaults — writing copy from product images, priming with example responses to mimic existing copy, draft-then-critique chains, and chaining a lifestyle image prompt into a video prompt.

Updated July 1, 2026

For general copy — descriptions, headlines, reasons to buy — the default prompts are usually enough, with maybe a tweak to the instructions. But a few situations call for more deliberate prompt design. This article covers four techniques:

  • Writing copy based only on product images, when you don't have usable existing text.
  • Priming a prompt with an example response so Aimée mimics your existing copy exactly.
  • A draft-then-critique chain that improves output quality at the cost of more computing power.
  • Chaining a lifestyle image prompt into a video prompt, so Aimée animates an image it just generated.

See Working with prompts first if you haven't used the prompt editor before.

Writing copy from product images#

Use this when you lack usable product descriptions but do have images.

  1. Create a new prompt and choose Write descriptions by looking at the product image as the type.
  2. Notice that its step is built differently from the default text prompt: alongside the usual Product name, Short product title, Brand name, and Product category placeholders, it includes up to three Product image placeholders — the AI-selected, background-removed images, not the raw feed Image link — feeding actual product photos into the step.
  3. Because the step attaches real images instead of only text, Aimée runs them through an image-understanding model as part of generating the copy, letting it describe what it actually sees in the photos alongside the product's specifications.

Mimicking your existing copy exactly#

This uses a priming technique: show the model a real example of input and desired output before giving it the actual product to process.

  1. Create a new prompt and choose Write descriptions as the type.
  2. Open the ... menu on the Instructions step and choose Add prompt step.
  3. In the new step, don't use placeholders — paste in the literal product information for one real, representative product. Match the same formatting your final step's placeholders will produce, for example:
Product name: Logitech Pro X TKL Lightspeed - GX Red linear
Short product title: Wireless Logitech gaming keyboard
Brand name: Logitech
Product category: Electronics > Electronics Accessories > Computer Components > Input Devices > Keyboards
Product description: """A championship-trusted wireless gaming keyboard designed for the highest levels of competitive play..."""
  1. Open the ... menu on that new step and choose Add prompt step again. Click the step's Input/Example-response icon to toggle it to Example response.
  2. In this new step, paste the exact copy you want Aimée to produce for that example product — written in your own voice and format.
  3. Use Generate to preview the prompt on a different product. Aimée now has a concrete before/after example to imitate, and should closely match your tone and formatting for every other product.

Writing and critiquing a draft#

This "chain of thought" pattern gives the model several instructions in sequence, letting it reflect on and improve its own output. It generally produces stronger copy, at the cost of more computing power (and more credits) per product.

  1. Create a new prompt and choose Write descriptions as the type.
  2. Change the Instructions to something like "Write a first draft for... we'll critique and improve it in later steps."
  3. Open the ... menu on the input step and choose Add prompt step. Instruct this step to critique the draft, for example: "Reflect on the draft and list ways it could be improved. Focus on natural-sounding language and on any claims not supported by the product information."
  4. Add one more step the same way, instructing the model to write a final version that incorporates the critique and drops any unsupported claims.
  5. Use Generate to preview — you can inspect each step's output individually, including the intermediate draft and critique, which makes it much easier to see why the final output turned out the way it did.

Chaining a lifestyle image into a video#

Aimée can animate a lifestyle image she generated in one prompt as the source for a second, video-generating prompt — useful when you want a moving version of a specific AI-generated scene rather than a video generated from scratch.

  1. Create (or reuse) a prompt of type Generate lifestyle images, scoped with Use as so its output is tagged the way you want (for example, to a specific product set).
  2. Create a second prompt of type Animate lifestyle images to video. Its source-image step is pre-wired to pull from a lifestyle image rather than product text — this is what lets it source specifically from AI-generated lifestyle images rather than feed images.
  3. Run the first prompt (or let it run automatically) so a lifestyle image exists for a product, then run the second. The video prompt will use that image as its starting frame.
  4. Preview with Generate on a product that already has a generated lifestyle image, so you can confirm the animation looks right before applying it at scale.

Because video generation typically has a materially higher credit cost than image or text generation, check the credit estimate shown after previewing before running this across a large product set — see AI image and video generation and Understanding credits.

Related articles#