Skip to content
View original post on X: ModelScopeOfficial· 46/100AI score46/100

ModelScope adds OPD training workflow to DiffSynth-Studio for few-step diffusion models

AISummary

ModelScope has added an OPD training workflow to DiffSynth-Studio for training few-step diffusion models.

It reports training convergence typically within 2 GPU-hours and distillation of FLUX.2-klein-base-4B and Qwen-Image-2.1 from 50 to 8 steps, and Qwen-Image-2.1-Turbo from 8 to 3 steps, while keeping generation quality.

The workflow trains on student-generated trajectories with teacher guidance, and the code is released under the Apache 2.0 license.

Post on XView on X
ModelScopeVerified on X
@ModelScope2022

Train your own few-step diffusion models faster, with OPD training workflow now offered in DiffSynth-Studio! Try example “TurTurbo” Qwen-Image-2.1 LoRAs trained with this pipeline:
🤖 https://modelscope.ai/models/DiffSynth-Studio/Qwen-Image-2.1-TurTurbo
📚 Docs: https://diffsynth-studio-doc.readthedocs.io/en/latest/Training/Distillation.html

⚡ Fast training feedback: with OPD, we are seeing training convergence typically 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 upholding generation qualtiy.
🧠 Simple workflow: train on student-generated trajectories with teacher guidance, without pre-generating teacher trajectories or designing an additional discriminator.
📜 Release under Apache 2.0 license.

Source: ModelScope · x.comPublished · added here

Topics

Open-source ecosystemDeployment & engineeringData & trainingProduct updates