Doodle-in LoRA for Qwen-Image-2.1 turns scribbles into photorealistic objects
Gradio · x · 2026-09-30
ML-Intern-lab released a Doodle-in LoRA for Qwen/Qwen-Image-2.1 on Hugging Face: draw a magenta scribble on a photo, name an object, and the model replaces the scribbled region with that object — following the scribble's shape and pose, lit consistently with the scene, leaving everything else untouched.
- Built on Qwen-Image-2.1's single generation+editing model, editing triggered by passing an image
- Trained on 6,042 pairs derived from Open Images V7
- Model card includes checkpoint selection, hit-rate breakdowns, and blind realism votes judged by Qwen2.5-VL-72B
More from Multimodal
- Sora API Shuts Down While Kling Makes Real Revenue: Viral Demos Don't Build Businesses — lmoroney · 2026-10-01
- This AI film cost $22,195 and took 13 days to make: PJ Ace's Nexus Ep 5 — NightsRadiant · 2026-09-30
- Open-source Photoshop relight clone built on Qwen Image 2.1 with turbo LoRAs — linoy_tsaban · 2026-09-30
- Testing Astra High for an AI motion-graphics SaaS product demo — Significant-Type6778 · 2026-09-30
- Minimax H3 ecosystem roundup: ComfyUI 0.38 pixel-seam fix, new LoRAs and a 3060 12GB hybrid — optimisticalish · 2026-09-30
- Clueso MCP tops Product Hunt's Product of the Week for AI video editing — Scobleizer · 2026-09-30