Collection

Collection generates a batch of style-consistent images from a single brief — forty product shots, a set of social tiles, a stock-style series. It’s not a loop around Image Gen: a planner model first turns your brief into one distinct prompt per image, so the batch varies where you want variety and holds where you want consistency.

Inputs	prompt (text — the brief), image_1 … image_4 (references)
Outputs	images

Parameters

Param	Notes
model	The image model used for every frame.
count	How many images the batch produces.
variation_axes	What should differ across the batch — e.g. angle, scene, time of day — while everything else stays consistent.
planner_model	The LLM that expands the brief into per-image prompts.
planned_prompts	The planner’s output, editable — review and tweak individual prompts before (or after) generating.
review	Optional automatic review pass over the finished batch.
aspect_ratio, quality, output_format	As in Image Gen.
reference_images	Complements the wired reference slots.

The workflow

1. Brief it — one paragraph on subject, look, and purpose. Add reference images to lock a product or character.

2. Set the variation axes — this is the lever that makes a batch feel curated instead of repeated.

3. Run — planning and generation execute as a background job; watch frames land as they finish. Prefer to see the plan first? Plan prompts first writes all the per-image prompts for review — edit or delete any of them — before a cent is spent on images:

   

4. Regenerate the misses — pick the frames that didn’t work and regenerate just those (optionally with a note about what to change — also available through the chat companion). The rest of the batch is untouched.

Tips

• Say what must not vary (“same jacket, same model, same palette”) right in the brief — the planner honors it across every prompt.
• Skim planned_prompts before a large run; fixing one odd plan is cheaper than regenerating after.
• Pin the look with a library style @-tag so follow-up batches match this one.