Qwen-Image-Edit-2511 vs LLaDA-Image-Turbo: hands-on image edit comparison
DiagramAwesome · reddit · 2026-09-08
The author ran 9 challenging edit prompts (object removal with scene reconstruction, local recoloring, mirror-reflection consistency, text rewriting, expression changes) against Qwen-Image-Edit-2511 and LLaDA-Image-Turbo, sampling 3 generations per test and picking the best.
Key findings:
- Qwen remains on top overall, but partial style transfer was disappointing and it basically did nothing when asked to add two obelisks
- Qwen showed odd color variations, e.g. in the water-reflection consistency test
- LLaDA-Turbo is quite good for its size and speed, but often "could do it yet did something completely different"; some of its failures were hilarious
Input images were generated with Krea2; both models were INT8 quantized (qwen-image-edit-2511-int8-convrot, LLaDA-Image-Turbo-INT8), with all prompts published.
More from Multimodal
- Dense instance segmentation via Muse Wonder + SAM 3.1 on Meta's Model API — nikhilaravi · 2026-10-26
- LichtFeld Studio v0.5.4.3 ships community-built SteadySplats, POPSpa and Voxelization — janusch_patas · 2026-10-26
- One prompt to a 2-3 min narrated short: hands-on test of Antigravity video generation — AI_Andrew · 2026-10-26
- Creator makes 10,564-frame animated short entirely with Gemini toolchain — AI_Andrew · 2026-10-26
- FLUX.3 Klein step-distilled to 4-8 steps with CFG baked in — Fit-Structure-6489 · 2026-10-26
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26