DiffSynth-Studio Adds OPD Workflow: Few-Step Diffusion Training in ~2 GPU-Hours
bdsqlsz · x · 2026-10-15
ModelScope announced that DiffSynth-Studio now offers the OPD training workflow for faster few-step diffusion model training. Key points:
- Fast feedback: with OPD, training convergence typically happens within 2 GPU-hours.
- Fewer sampling steps: distill FLUX.2-klein-base-4B and Qwen-Image-2.1 from 50 to 8 steps, or Qwen-Image-2.1-Turbo from 8 to 3, while preserving generation quality.
- Simple workflow: train on student-generated trajectories with teacher guidance, without pre-generating teacher trajectories or designing an extra discriminator.
- License: released under Apache 2.0.
An example "TurTurbo" Qwen-Image-2.1 LoRA trained with this pipeline and docs links are provided.
More from Multimodal
- FLUX.3 Image Klein ships open in Diffusers: layout control, 2s per image on H100 — ariG23498 · 2026-10-15
- Kling 4.0 hands-on: native 30-second 1080p generation and Omni Reference tested — HeyZoyaKhan · 2026-10-15
- Kling 4.0 generates a cinematic Tarzan riding a giant wolf, hard to tell it's AI — HeyZoyaKhan · 2026-10-15
- Kling 4.0 pitches director-style AI video: choreographed action and world-building, not random clips — alex_verem · 2026-10-15
- Adding motion graphics with just a prompt: the Claude Motion x ElevenLabs MCP workflow — CurieuxExplorer · 2026-10-15
- Google turns dense academic PDFs into animated storybooks with Gemini, Nano Banana and Veo — CurieuxExplorer · 2026-10-15