Dev finetunes an open model to remove video subtitles and watermarks
keithhon · reddit · 2026-09-15
Finding watermark removal still hard with existing tools, the author finetuned an open-source model to tackle it. A 10-second video takes half a minute to a few minutes depending on the number of overlays (not yet accelerated). It also works on still images, and a live demo is available on Hugging Face Spaces.
More from Multimodal
- ElevenLabs Brings Voice, Music, Image and Video Generation to Its MCP Server — nikola_mr64990 · 2026-09-15
- PolloAI-generated GTA 6 style video of Los Santos looks playable — thetripathi58 · 2026-09-15
- Manycore launches Aholo Lux3D: text or image to 3D assets in as fast as 20 seconds — JaynitMakwana · 2026-09-15
- Creator Builds Childhood Heroes Tribute With GPT, fal and three.js — OdinLovis · 2026-09-15
- Video Character Swap with Higgsfield Genjutsu: Full Prompt and Workflow — CodeByPoonam · 2026-09-15
- Consistent AI Image Editing Needs Heavy Hand-Written Constraints, Not Casual Prompting — eyishazyer · 2026-09-15