Qwen based LoRa was trained in Onetrainer, dataset is 50 frames in folk horror genre, was trained for 120 epochs, works with lightning loras aw, working weight is 0.8-1.2
no trigger words. but for prompting i use structure like that:
rural winter pasture, woman with long dark braided hair wearing weathered, horned headdress and thick woolen shawl, profile view, solemn gaze toward herd, 16mm Sovcolor analog grain, desaturated ochre, moss green, and cold muted blues, diffused overcast daylight with atmospheric haze, static wide shot, Tarkovskian composition with folkloric symbolism emphasizing isolation and ancestral presence
domestic interior, young woman with long dark hair wearing white Victorian gown and red bonnet, serene expression lying in glass sarcophagus, 16mm Sovcolor film stock aesthetic with organic grain, desaturated ochre earth tones and muted sepia, practical firelight casting shadows through branches, static wide shot emphasizing isolation and rural dread
foggy forest interior, young woman with dark hair wearing traditional linen dress and white bonnet, holding broomstick with weathered hands, melancholic expression, 16mm Sovcolor film stock aesthetic with organic grain, desaturated palette of muted sepia and faded crimson, diffused overcast daylight filtering through mist, static wide shot emphasizing isolation and rural dread
ritual scene, three figures in traditional embroidered garments and head coverings, arranged in symmetrical formation with raised arms, solemn expressions, Tarkovskian slow cinema composition, muted sepia and faded crimson with religious iconography, golden hour backlighting through wooden structures, locked-off medium shot with folkloric symbolism