From Sketch to Movie Poster: Adding Kontext to Your Art Process

I’m loving Flux Kontext, especially since ai-toolkit added LoRA training. It was mostly trivial to use my original datasets from my \[Every Heights LoRA models\](https://everlyheights.tv/product-category/stable-diffusion-models/flux/) and make matched pairs to train Kontext LoRAs on. After I trained a general style LoRA and my character sheet generator, I decided to do a quick test. This took about 45 minutes.

Overall, Kontext is a super powerful tool, especially when combined with my work from the past three years building out my Everly Heights style/animation asset generator models. I’m thinking about taking all the LoRAs I’ve trained in Kontext since the training stuff came out (Prop Maker, Character Sheets, style, etc.) and packaging it into an easy-to-use WebUI with a style picker and folders to organize the characters you make. Sort of an all-in-one solution for professional creatives using these tools. I can hack my way around some code for sure, but if anybody wants to help let me know.

STEP 4 PROMPT: A 3×3 grid of illustrations featuring the same stylized character in a variety of poses, moods, and locations. Each panel should depict a unique moment in the character’s life, showcasing emotional range and visual storytelling. The scenes should include:

A heroic pose at sunset on a rooftop
Sitting alone in a diner booth, lost in thought
Drinking a beer in an alley at night
Running through rain with determination
Staring at a glowing object with awe
Slumped in defeat in a dark alley
Reading a comic book under a tree
Working on a car in a garage smoking a cigarette
Smiling confidently, arms crossed in front of a colorful mural

Each square should be visually distinct, with expressive lighting, body language, and background details appropriate to the mood. The character should remain consistent in style, clothing, and proportions across all scenes.I’m loving Flux Kontext, especially since ai-toolkit added LoRA training. It was mostly trivial to use my original datasets from my \[Every Heights LoRA models\](https://everlyheights.tv/product-category/stable-diffusion-models/flux/) and make matched pairs to train Kontext LoRAs on. After I trained a general style LoRA and my character sheet generator, I decided to do a quick test. This took about 45 minutes.