Skip to main content
All CollectionsImage Generation
Product shots with Custom AIs
Product shots with Custom AIs

Custom model training in Kive. Select good input data, train model and use it to generate high quality images and product photography

Olof Lindh avatar
Written by Olof Lindh
Updated over 2 weeks ago

What is Custom AI?

Custom AI, or model training, is a way to teach Kive's AI new knowledge. Using it, you can train the AI to recognise and reproduce specific products, objects, persons and styles. Custom AIs can be used in Product Shots to create on-brand campaign imagery.

In Early Access. Custom AIs and Product Shots is currently in early access Beta, but will be available to all Pro subscribers soon.

Training a Custom AI model

Image sources

You can use professional product photos or simply take pictures with your smartphone – both work well for training. Just make sure your photos are clear and well-lit.

Tips for selecting images

AI learns from your images. If key details are missing, it will fill in the gaps (often incorrectly). To avoid this, use at least 5 images per subject (10+ images for best results).

  1. Distinct backgrounds

    When training object models, you want to use training images where the object is distinct against the background. Avoid backgrounds that are of the same colour as the object.

  2. Diverse images

    Diverse backgrounds and lighting conditions help the AI recognise your subject in various settings. This improves understanding of size, context, and environment.

  3. Multiple angles

    Add images showing your subject from different key perspectives – front, side, high, low – to teach the AI. Focus on the angles you want to generate in (it's fine to stick with one or two angle if that's all you need, and you don't need to show parts of the object that are not important when you'll be generating – for example the inside of a car if you'll be generating only exterior images). You'll only be able to generate new images from the angles you upload now.

  4. Vary the lighting

    Include shots in different lighting – natural, artificial, or other relevant scenarios – to make your model adaptable.

  5. Prioritise high quality

    Use clear, sharp images with a resolution of at least 1024x1024px. Avoid blurry or pixelated photos, and ensure the subject is in focus.

  6. Close-ups

    If you need the model to learn fine text, logos or labels – be sure to include close ups of those important details.

Example input images

Using a Custom AI model

Once your Custom AI is trained, you can use it by tagging it in your prompt using “@”. For example: "Photo of @Acme Serum".

Examples

photo of @Acme Serumsurrounded by flowers, flowers everywhere, with lush rain forest background, deep in a jungle, lush nature, cinematic scene, ultrasharp photo, cinematic, Cinestill 800T, shot by Roger Deakins, shot on Kodak Portra, beautiful light, dawn

photo of @Acme Serum surrounded by flowers, flowers everywhere, with lush rain forest background, deep in a jungle, lush nature, cinematic scene, ultrasharp photo, cinematic, Cinestill 800T, shot by Roger Deakins, shot on Kodak Portra, beautiful light, dawn

photo of @Acme Serumstanding on a shiny graphite crystal crest, peak of a graphite mountain, dark dusk, blue hour, low light, cold light, shot on ultrasharp 35mm film, shot with Leica, diffracting light, glow, black mist diffusion filter

photo of @Acme Serum standing on a shiny graphite crystal crest, peak of a graphite mountain, dark dusk, blue hour, low light, cold light, shot on ultrasharp 35mm film, shot with Leica, diffracting light, glow, black mist diffusion filter

Did this answer your question?