From 325a95051bb20787d3db9faba9b4f62b2a63c43a Mon Sep 17 00:00:00 2001 From: Lev Novitskiy <57654885+leffff@users.noreply.github.com> Date: Tue, 4 Nov 2025 03:38:07 +0300 Subject: [PATCH] Kandinsky 5.0 Docs fixes (#12582) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * add transformer pipeline first version * updates * fix 5sec generation * rewrite Kandinsky5T2VPipeline to diffusers style * add multiprompt support * remove prints in pipeline * add nabla attention * Wrap Transformer in Diffusers style * fix license * fix prompt type * add gradient checkpointing and peft support * add usage example * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: Álvaro Somoza * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: Álvaro Somoza * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: Álvaro Somoza * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: Álvaro Somoza * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: Álvaro Somoza * remove unused imports * add 10 second models support * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * remove no_grad and simplified prompt paddings * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * moved template to __init__ * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * moved sdps inside processor * remove oneline function * remove reset_dtype methods * Transformer: move all methods to forward * separated prompt encoding * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * refactoring * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * refactoring acording to https://github.com/huggingface/diffusers/commit/acabbc0033d4b4933fc651766a4aa026db2e6dc1 * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * Update src/diffusers/pipelines/kandinsky5/pipeline_kandinsky.py Co-authored-by: YiYi Xu * fixed * style +copies * Update src/diffusers/models/transformers/transformer_kandinsky.py Co-authored-by: Charles * more * Apply suggestions from code review * add lora loader doc * add compiled Nabla Attention * all needed changes for 10 sec models are added! * add docs * Apply style fixes * update docs * add kandinsky5 to toctree * add tests * fix tests * Apply style fixes * update tests * minor docs refactoring * refactor Kandinsky 5.0 Vide docs * Update docs/source/en/_toctree.yml --------- Co-authored-by: Álvaro Somoza Co-authored-by: YiYi Xu Co-authored-by: Charles Co-authored-by: Sayak Paul Co-authored-by: github-actions[bot] --- docs/source/en/_toctree.yml | 4 ++-- .../api/pipelines/{kandinsky5.md => kandinsky5_video.md} | 8 ++++---- 2 files changed, 6 insertions(+), 6 deletions(-) rename docs/source/en/api/pipelines/{kandinsky5.md => kandinsky5_video.md} (90%) diff --git a/docs/source/en/_toctree.yml b/docs/source/en/_toctree.yml index 8103c01643..251eb25899 100644 --- a/docs/source/en/_toctree.yml +++ b/docs/source/en/_toctree.yml @@ -529,8 +529,6 @@ title: Kandinsky 2.2 - local: api/pipelines/kandinsky3 title: Kandinsky 3 - - local: api/pipelines/kandinsky5 - title: Kandinsky 5 - local: api/pipelines/kolors title: Kolors - local: api/pipelines/latent_consistency_models @@ -656,6 +654,8 @@ title: Text2Video-Zero - local: api/pipelines/wan title: Wan + - local: api/pipelines/kandinsky5_video + title: Kandinsky 5.0 Video title: Video title: Pipelines - sections: diff --git a/docs/source/en/api/pipelines/kandinsky5.md b/docs/source/en/api/pipelines/kandinsky5_video.md similarity index 90% rename from docs/source/en/api/pipelines/kandinsky5.md rename to docs/source/en/api/pipelines/kandinsky5_video.md index a98a0826b7..533db23e1c 100644 --- a/docs/source/en/api/pipelines/kandinsky5.md +++ b/docs/source/en/api/pipelines/kandinsky5_video.md @@ -7,9 +7,9 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o specific language governing permissions and limitations under the License. --> -# Kandinsky 5.0 +# Kandinsky 5.0 Video -Kandinsky 5.0 is created by the Kandinsky team: Alexey Letunovskiy, Maria Kovaleva, Ivan Kirillov, Lev Novitskiy, Denis Koposov, Dmitrii Mikhailov, Anna Averchenkova, Andrey Shutkin, Julia Agafonova, Olga Kim, Anastasiia Kargapoltseva, Nikita Kiselev, Anna Dmitrienko, Anastasia Maltseva, Kirill Chernyshev, Ilia Vasiliev, Viacheslav Vasilev, Vladimir Polovnikov, Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina, Tatiana Nikulina, Polina Gavrilova, Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Denis Dimitrov +Kandinsky 5.0 Video is created by the Kandinsky team: Alexey Letunovskiy, Maria Kovaleva, Ivan Kirillov, Lev Novitskiy, Denis Koposov, Dmitrii Mikhailov, Anna Averchenkova, Andrey Shutkin, Julia Agafonova, Olga Kim, Anastasiia Kargapoltseva, Nikita Kiselev, Anna Dmitrienko, Anastasia Maltseva, Kirill Chernyshev, Ilia Vasiliev, Viacheslav Vasilev, Vladimir Polovnikov, Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina, Tatiana Nikulina, Polina Gavrilova, Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Denis Dimitrov Kandinsky 5.0 is a family of diffusion models for Video & Image generation. Kandinsky 5.0 T2V Lite is a lightweight video generation model (2B parameters) that ranks #1 among open-source models in its class. It outperforms larger models and offers the best understanding of Russian concepts in the open-source ecosystem. @@ -92,7 +92,7 @@ pipe = pipe.to("cuda") pipe.transformer.set_attention_backend( "flex" -) # <--- Set attention backend to Flex +) # <--- Sett attention bakend to Flex pipe.transformer.compile( mode="max-autotune-no-cudagraphs", dynamic=True @@ -115,7 +115,7 @@ export_to_video(output, "output.mp4", fps=24, quality=9) ``` ### Diffusion Distilled model -**⚠️ Warning!** all nocfg and diffusion distilled models should be inferred without CFG (```guidance_scale=1.0```): +**⚠️ Warning!** all nocfg and diffusion distilled models should be infered wothout CFG (```guidance_scale=1.0```): ```python model_id = "ai-forever/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers"