By Lucas Desfossez · Rawrr project operator
Local AI can help designers explore visual directions on their own workstation. The useful question is not just “can it make an image?” but “can I control the result, repeat the workflow and use the output in a real design process?”
From a brief to a direction
Start with a clear design constraint: silhouette, composition, material, palette or atmosphere. Use a reference you have permission to process. Generate a small set of variations, compare them against the brief, and select what to develop. Keep the original reference separate from generated alternatives so reviewers can tell what changed.
Concept art, product ideas and visual development
Image generation can support mood exploration and concept iteration. Video can help test movement or a visual treatment, while 3D-oriented workflows can support spatial exploration. Generated visuals are not validated engineering drawings, accurate material specifications or production-ready geometry by default. Human review remains part of the job.
Use LoRAs for a more intentional visual identity
A compatible LoRA can adapt a base model toward a particular visual style or subject. Record the base model, LoRA version, strength and generation settings when comparing results. A LoRA does not guarantee consistent geometry, exact branding or identical characters across every output. Check rights for both training references and the resulting adaptation.
Keep confidential inputs within an approved workflow
For unreleased products, client concepts and NDA work, decide where processing is allowed before importing assets. Fully local inference does not need to upload those inputs to an external generation provider. Plugins, backups and deliberate sharing can create separate data paths. Use public stand-ins until the workflow is approved.
Evaluate results, not just a feature list
- Does the result preserve the constraints in your brief?
- Can you reproduce the workflow after a model update?
- Can you trace which inputs, models and licences were used?
- Does the output fit your next editing or review step?
- Are processing time and memory use practical on your workstation?
Rawrr’s direction
Rawrr is being built around local-first image, video and 3D creation with personal LoRA workflows. Supported features, hardware and export capabilities will be confirmed during beta; this guide describes evaluation criteria, not a promise that every workflow is already supported.