This study develops a practical framework for evaluating AI-generated brand visuals in a marketing communication context. Using a qualitative case study of an Indonesian sustainable startup (I-NewBee), the research employs photo-elicitation interviews to compare consumer evaluations of brand assets generated by three text-to-image tools (Midjourney, Neural Love AI, and Leonardo AI). Twelve target-market participants assessed anonymized visual sets (A/B/C) produced under a standardized prompt structure, using a semi-structured protocol informed by attention-stage cues from the AISAS model. Expert input from a visual communication design practitioner was used to triangulate judgments on visual quality and brand fit. Data were analyzed through iterative coding and thematic synthesis to identify recurring evaluation dimensions and decision cues. The findings suggest that perceived brand fit is shaped by visual realism, compositional clarity, brand-consistent signals, and message interpretability, while prompt ambiguity and inconsistent visual cues reduce credibility. The paper contributes an actionable evaluation framework expressed as evaluation dimensions and prompt-design considerations for startups seeking to deploy generative AI responsibly in brand communication