Community-trained MOSS Sharpness Slider LoRA: 72 photos, 500 steps, +/-1 focus control
aoleg77 · reddit · 2026-10-27
A Redditor released an independently trained sharpness concept-slider LoRA for MOSS on Hugging Face. They curated 72 varied real photos (including bokeh shots so the subject, not the background, stays in focus), built a contrasting set via Gaussian blur at varying radii, trained a factor-8 slider for 500 steps and picked step 400 to avoid overfitting. Weight +1 gives sharp in-focus images, -1 gives blur; more sharpness trades off depth-of-field control.
More from Multimodal
- ComfyUI Krea2 Prompt Styler Node Revived in a Fork With a Massive Style Library — Winter_unmuted · 2026-10-27
- Single HTML file generates retro poster designs in-browser with zero LLM calls — Afinetheorem · 2026-10-27
- A Redditor Used Copilot to Imagine and Archive Their Own '80s/'90s Winter Streetwear Concepts — Valuable-Sky2134 · 2026-10-27
- Transkun open-source piano transcription model nails MIDI, sustain events included — alexanderchen · 2026-10-27
- fal adds FLUX 3 Image Klein LoRA training: 10 reference images, 3 LoRAs per request — OdinLovis · 2026-10-27
- Guy turns his girlfriend's selfie into a playable AI game ad about popping pimples — CurieuxExplorer · 2026-10-27