Visual Engine: on-brand images, delivered daily.

The situation

The client publishes visual content daily: characters, environments, posters and key art. She has a recognisable style and a clear preference for OpenAI's GPT Image models, but every session started from zero. Prompts had to be rewritten, results drifted away from her look, and a large part of each day went to regenerating and rejecting images instead of publishing them.

What we built

Instead of another prompt template, we built a dedicated engine inside the Vyndexo workspace that carries her style permanently.

• Style base. 340 reference images supplied by the client were analysed and condensed into a permanent style profile that every generation is anchored to. Style strength is adjustable up to maximum.
• Model choice. The engine runs on GPT Image 2.0 and GPT Image 2.5 (Flare and Sunburst) because those are the models she prefers. The architecture is model-independent and can be pointed at any image model.
• Creation types. Characters, environments, mix-ups and posters, covers and key art can be combined per setup. A Custom type lets her add permanent guidance in her own words: recurring characters, mood, colours, composition and things to avoid.
• Two ways to produce. A scheduled mode delivers 1 to 8 images per day at the times and weekdays she chooses, in her own timezone. A manual mode lets her enter her own prompts, with the engine applying her style on top.
• Output control. Quality, resolution, opaque or transparent background, and an aspect-ratio mix (3:4 primary, 16:9, 4:3, 2:3, 9:16) so each daily set already fits every channel.
• Delivery. Each connected design account receives one permanent project for the setup, so finished images land where she publishes from, not in a download folder.

Rules and exception handling

• Scheduled runs follow her weekdays and times in her local timezone; days she unticks are skipped.
• Failed generations are never charged and are shown in the setup instead of silently producing nothing.
• Changing models, ratios or guidance only affects future runs; images already delivered stay as they are.

Delivered scope

• Style analysis of 340 reference images and a permanent style profile
• Scheduling engine with per-day quantity, times, weekdays and timezone
• Manual prompt mode with the same style anchoring
• Multi-generator and multi-ratio support with quality and background controls
• Delivery to five connected design accounts
• Hand-over and ongoing operation as part of her Vyndexo workspace

Outcome facts

• Live and in daily use since October 2026; client anonymised at her request, based in Romania.
• Input: 340 reference images supplied by the client, condensed into one permanent style profile.
• Output: 1 to 8 images per scheduled day, on the weekdays and times she sets, in five aspect ratios; manual prompts any time.
• Delivery: five connected design accounts, each with its own permanent project.
• Not measured: time saved per day. We publish that figure only once the client has recorded it and approved the statement.

Result

The client now receives her daily set automatically and only opens the engine when she wants something specific. Keeping her style consistent is no longer a daily task. Kroventra publishes measured figures only when the client's records support them; this case describes delivered scope and working behaviour, not a guaranteed outcome.

Published anonymously at the client's request; she approved this description. See all work.