v0.28.0

What's Changed

• Add AGENTS.md by @comfyanonymous in #14696
• Add some more stuff to AGENTS.md by @comfyanonymous in #14704
• Fix Qwen3-VL tokenizer crash with custom embeddings by @silveroxides in #14713
• Update AGENTS.md with more stuff. by @comfyanonymous in #14725
• [Partner Nodes] chore(Ideogram): remove IdeogramV1 and IdeogramV2 nodes by @bigcat88 in #14712
• Update AGENTS.md by @comfyanonymous in #14726
• security: fix four vulnerabilities (GHSA-779p-m5rp-r4h4) by @mattmillerai in #14734
• Update AGENTS.md by @comfyanonymous in #14733
• [Partner Nodes] feat(ByteDance): add support for Seed Audio 1.0 by @bigcat88 in #14731
• Update workflow templates to v0.11.2 by @comfyui-wiki in #14741
• [Partner Nodes] chore(StabilityAI): remove StabilityAI nodes by @bigcat88 in #14737
• Update AGENTS.md by @comfyanonymous in #14738
• [Partner Nodes] chore(ByteDance): adjust category name by @bigcat88 in #14752
• Add CLAUDE.md as symlink to AGENTS.md by @robinjhuang in #14757
• fix: error when embedding is loaded with models using llama_template by @silveroxides in #14744
• Try to make coderabbit enforce AGENTS.md by @comfyanonymous in #14759
• fix: Bug when setting transparency in color picker by @alexisrolland in #14764
• [Partner Nodes] fix(logs-auth): mask authorization headers in logs by @bigcat88 in #14774
• Small speedup for text model sampling. by @comfyanonymous in #14773
• Update embedded docs to v0.5.7 by @comfyui-wiki in #14783
• Skip unloading dynamic model patchers in current workflow. by @comfyanonymous in #14799
• ci: add CLA Assistant workflow by @alexisrolland in #14582
• Delete symlink that breaks our updates. by @comfyanonymous in #14803
• fix(Video): don't crash on videos with undecodable audio streams by @bigcat88 in #14746
• feat: Implement basic text overlay node (CORE-137) by @drozbay in #14610
• Update AGENTS.md by @comfyanonymous in #14819
• Support gqa on all attention backends, drop support for pytorch 2.4 by @comfyanonymous in #14772
• Fix crash when using UNetSelfAttentionMultiply by @comfyanonymous in #14823
• add models-directory launch argument by @silveroxides in #9113
• Fix AttributeError in VAE.is_dynamic() for VAEs constructed without a patcher by @j2gg0s in #14826
• [Partner Nodes] feat(ByteDance): add Seedream 5 Pro model support by @bigcat88 in #14832
• Update workflow templates to v0.11.6 by @comfyui-wiki in #14834
• feat(assets): add namespaced model_type tags and align tag semantics by @synap5e in #14511
• Fix qwen3vl reference images when used as a text encode models. by @comfyanonymous in #14845
• feat: allow --comfy-api-base target ephemeral testenvs by @bigcat88 in #14569
• fix: Load3D failing path validation from double path resolution by @jtydhr88 in #14852
• feat(api): expose effective extension filter on /experiment/models by @synap5e in #14797
• Update cla.yml by @alexisrolland in #14851
• Support convrot int4 models. by @comfyanonymous in #14859
• Try to fix the model reloading issue some people have. by @comfyanonymous in #14822
• Fix FP8 activation quantization for >2D activations in mixed_precision_ops by @liminfei-amd in #14643
• Enable comfy-kitchen Triton backend by default on ROCm/AMD by @liminfei-amd in #14862
• Fix black image on turing when using int4 models. by @comfyanonymous in #14864
• Add SeedVR2 support (CORE-6) by @pollockjj in #14424
• [Partner Nodes] feat(Seedream): add widget to disable thinking by @bigcat88 in #14853
• [Partner Nodes] fix(Tencent): restore Tencent3DPartNode FBX output by @bigcat88 in #14867
• Only auto-enable the ROCm comfy-kitchen Triton backend on matrix-core GPUs by @liminfei-amd in #14869
• Save Text Node (CORE-176) by @yousef-rafat in #14102
• feat: add bboxes input to Create Bounding Boxes node by @jtydhr88 in #14724
• CORE-329 feat: add Save 3D (Advanced) node family by @jtydhr88 in #14701
• fix: Undeprecate the text node by @alexisrolland in #14870
• chore: Update preview nodes by @alexisrolland in #14871
• Try to fix some issues with the seedvr VAE. by @comfyanonymous in #14877
• Temporarily disable auto enabling triton by default on AMD. by @comfyanonymous in #14878
• Try to fix flash attention related issue on AMD. by @comfyanonymous in #14880
• Fix SaveVideo description: says images, saves video by @gmhelmold in #14885
• Support PID 1.5 models. by @comfyanonymous in #14894
• [Partner Nodes] feat(client): send ComfyUI Core version in request headers by @bigcat88 in #14910
• fix(image): correct HLG inverse-OETF clamp in hlg_to_linear by @bigcat88 in #14762
• fix(image): support single-channel images in Save Image (Advanced) by @bigcat88 in #14761
• [Partner Nodes] chore(Google): reroute Gemini Image preview models to release versions by @bigcat88 in #14917
• Update embedded docs to v0.5.8 by @comfyui-wiki in #14920
• chore(openapi): sync shared API contract from cloud@491db61 by @comfy-pr-bot in #14815
• Update workflow templates to v0.11.9 by @comfyui-wiki in #14924
• Fix hidream o1 regression. by @comfyanonymous in #14923
• More optimized int8 and int4 on turing. by @comfyanonymous in #14927
• Fix 3D advanced nodes crashing in API mode (no UI) by @jtydhr88 in #14930
• Fix int8 performance regression on 16xx series. by @comfyanonymous in #14941
• [Partner Nodes] feat(sync.so): add support for "sync-3" model by @bigcat88 in #14928
• Fix cached outputs missing from job results when prompt has no client_id by @bigcat88 in #14939
• Bump comfyui-frontend-package to 1.45.21 by @comfy-pr-bot in #14944

New Contributors

• @j2gg0s made their first contribution in #14826
• @liminfei-amd made their first contribution in #14643
• @gmhelmold made their first contribution in #14885

Full Changelog: v0.27.0...v0.28.0