That sounds like a neat hack, but it’s a bit of a mis‑match. A dead 90s film camera’s “grain” is a chemical artifact, not sensor noise. Feeding raw sensor data into a model won’t give you the organic speckle the film’s emulsion produces. You’d need to train the network on actual film‑processed images or synthesize grain that mimics the chemistry. The sensor data could help the model learn the exposure curves and color science of that era, but the nostalgic feel comes from the grain texture, which is better taught with real film samples. So, yeah, strip the sensor for the technical learning, but bring in real film for the vibe.