34MB Kodachrome LoRA for MOSS Image trained on 551 4x5 film scans, one-shot generations
aoleg77 · reddit · 2026-10-26
Reddit user aoleg77 released a Kodachrome-style LoRA for MOSS Image on Hugging Face: a factor-16 LoKR of just 34MB, trained on 551 scans of 4x5 Kodachrome sheet film (not 35mm, so no film grain).
- One-shot generations on the base model transfer Kodachrome's color response and sharpness — no Turbo, no SFT, and no blur even with an empty negative prompt.
- Trigger word is "Kodachrome" / "Kodachrome film", but the author recommends phrasing like "shot on Kodachrome film" to avoid the model literally rendering the trigger text into images.
- Sample shots (a railway worker photo) show the desired look transferred perfectly.
More from Multimodal
- One sref code, three animals: Midjourney style refs nail a unified sporty cartoon look — sergeantsref · 2026-10-26
- Cozy Halloween Midjourney set with reusable sref code 623781479 — miilesus · 2026-10-26
- Hyperrealistic AI-generated street clip shows how hard fakes are to spot now — Dayna6380- · 2026-10-26
- LTX-2.5 runs in ComfyUI: text-to-video talking influencer demo — comfy-king-999 · 2026-10-26
- Claude plugin auto-creates music videos with perfect beat sync and AI-rendered scenes — AIandDesign · 2026-10-26
- SDXL + Z-Image Turbo: a 2-step hybrid workflow in ComfyUI with CMK Flow — Away-Sheepherder-578 · 2026-10-26