One-prompt-trained Qwen-Image-2.1 LoRA turns your doodles into photorealistic objects
Gradio · x · 2026-09-30
Gradio highlighted Qwen-Image-2.1 Doodle-in LoRA, an open-source adapter allegedly designed, trained, and evaluated end-to-end by a "prompt-driven ML Intern."
How it works: draw a rough magenta scribble on a photo, add a short text naming an object, and the model replaces the scribble with that object — following the scribble's shape and pose, lit consistently with the scene, leaving everything else unchanged.
Details:
- Built on Qwen/Qwen-Image-2.1 (one model for generation and editing; editing triggered by passing image=);
- Trained on 6,042 pairs derived from Open Images V7;
- Evals include checkpoint selection on a fixed 48-pair subset, hit-rate breakdowns on 160 pairs, blind realism voting on 100 pairs (judge: Qwen2.5-VL-72B), plus out-of-domain checks;
- Open-sourced on Hugging Face.
Related event: Gradio Open-Sources Doodle-in LoRA for Qwen-Image 2.1(5 posts)→
More from coding & agent
- Generic API-catalog MCP server vs one tool per job: a context-bloat experiment — MandeepKaur9407 · 2026-11-16
- Berkeley paper: LLMs answer with stale values even when the update is in context — ombre-ai · 2026-11-16
- Effective HTML: 4.2k-star agent skills turn wireframes into self-contained interactive prototypes — tom_doerr · 2026-11-16
- skills-gate: one gate skill fronts 148 agent skills at just 389 tokens — Distinct-Pie2389 · 2026-11-16
- Open-weight Karpo-Decision-31B tops Decision Index at 73.9, one forward pass per answer — CodeByPoonam · 2026-11-16
- shadcn boosts @uisdk: one-shot UI rebuild that's instantly E2E-testable by AI — shadcn · 2026-11-16