The old way was prompting every image yourself, feeding references between tools and keeping eight tabs open. Here you brief an Agent and direct it, which is a different job: you are choosing, not typing.
Direction first, production second. Ask for the look before you ask for deliverables, because everything downstream inherits it.
Attach reference images before you type a word. Words like cozy or premium mean something different to everyone, and an Agent working from adjectives alone will hand you the average of the internet. Three to six pictures of the mood you are chasing do more than a paragraph of description, and they are the difference between a moodboard that looks like your brand and one that looks like stock.
I am attaching [3 to 6] reference images. Treat them as the mood, not as something to copy: take the light, the colour and the texture, never the products, the logos or the compositions.
Build a moodboard for a new brand called [name], a [what it is] for [who it is for].
The feeling is [three adjectives, for example cozy, linen, unhurried]. The palette leans [colours]. The logo should feel [hand drawn / geometric / typographic].
Show me 9 images covering: the product, the space, the packaging, a texture close-up, and one lifestyle shot. Include two logo directions.
Then tell me, in one line, the visual rule you pulled out of my references and are holding across the whole set, so I can keep it for everything that comes after.Once you like the direction, ask for the campaign on top of it. The Agent plans the work, then produces it, and it advances several directions at the same time rather than one image at a time.
Using this moodboard as the locked visual direction, build a launch campaign for [brand].
Give me: a shot list of 6 scenes, the images for each scene, a 15 second video ad cut from them, a voiceover script and its audio, and background music that matches the mood.
Keep the product, palette and lighting rule identical across every asset. Flag anything you could not keep consistent instead of quietly changing it.The agents evaluate their own outputs and keep the brand consistent across the project, which is the part that used to eat your afternoon: checking that shot six still looks like shot one.
A Skill is a repeatable workflow: a saved set of instructions anyone can run to get the same kind of result. Once a campaign works, save the route rather than the output, and the next brand takes minutes.
Turn everything we just did into a reusable Skill called "Brand launch campaign".
It should ask me for: the brand name, what it sells, three feeling words, the palette and the logo style. Then run the same sequence: moodboard, approval, shot list, images, video, voiceover, music.
Before you save it, tell me which parts must stay identical on every run and which parts should change per brand.You do not choose between image models any more. Luma evaluates its available models on performance, speed and output quality, then picks the most appropriate one for each task, or a combination of them. Your job moves up a level, to whether the output is any good.
🎬 Luma · pricing
♻️ Luma Skills
❓ Agents FAQ