ComfyUI v0.40.0
retained text
# Links * [**Official changelog**](https://docs.comfy.org/changelog#v0-40-0) * [**GitHub release**](https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.40.0) # New Open-Source Model Support * [**Qwen-Image 2.1 Turbo**](https://github.com/Comfy-Org/workflow_templates/pull/1328): The 8-step accelerated Qwen-Image 2.1 checkpoint, with new text-to-image and image edit templates # New Node Updates * [**LoRA stack loaders**](https://github.com/Comfy-Org/ComfyUI/pull/16309): Apply several LoRAs in order from one node, each with its own file, strength, and on/off switch, to either the diffusion model or the text encoder * [**Camera Angle**](https://github.com/Comfy-Org/ComfyUI/pull/16251): Pick a horizontal angle, vertical angle, and zoom to output a `camera_info` and a plain-English shot description for prompts * [**Create Camera Info**](https://github.com/Comfy-Org/ComfyUI/pull/14964): Build a `camera_info` in orbit, look-at, or quaternion mode for Render Splat and the other 3D camera nodes * [**Latent Operation Blend**](https://github.com/Comfy-Org/ComfyUI/pull/16923): A new blend operation, plus start and end percent on `LatentApplyOperationCFG` * [**LoRA attention config**](https://github.com/Comfy-Org/ComfyUI/pull/16880): LoRAs can replace a model's per-layer attention config, so turbo LoRAs can switch layers to sol or kitchen attention # Partner Node Updates * [**Fish Audio Drama 3**](https://github.com/Comfy-Org/ComfyUI/pull/16999): Fish Audio's voice direction model on the Text to Speech node, with plain-language directions, word tags, and up to 5 voices * [**Higgsfield Genjutsu Motion Transfer**](https://github.com/Comfy-Org/ComfyUI/pull/16913): Rebuild the cast, location, and style of a video from reference images while keeping its motion and audio * [**HeyGen Video 1.0**](https://github.com/Comfy-Org/ComfyUI/pull/16876): Added 2K resolution to the HeyGen Video 1.0 nodes # Performance & Stability * [**Qwen-Image 2.1 low VRAM**](https://github.com/Comfy-Org/ComfyUI/pull/16870): comfy-aimdo 0.5.6 and fewer compiler rogues when the KV cache is offloaded to CPU * [**Qwen 3.5 LLM**](https://github.com/Comfy-Org/ComfyUI/pull/16820): Faster in low VRAM conditions * [**Qwen 3.5 MTP**](https://github.com/Comfy-Org/ComfyUI/pull/16922): MTP decoding now also runs on prompts that contain images * [**comfy_attention**](https://github.com/Comfy-Org/ComfyUI/pull/16806): Extended to the remaining models, and [model files can now use kitchen sol attention](https://github.com/Comfy-Org/ComfyUI/pull/16831) * [**AMD gfx1151**](https://github.com/Comfy-Org/ComfyUI/pull/16726): Faster attention when keys and values do not fit in L2 cache * [**Saved EXR**](https://github.com/Comfy-Org/ComfyUI/pull/16948): EXR files are now saved with compression * [**GLB textures**](https://github.com/Comfy-Org/ComfyUI/pull/16867): Faster saving with PNG compression level 4 * [**3D outputs**](https://github.com/Comfy-Org/ComfyUI/pull/16200): Every 3D node now reports its models as a standard `3d` output item * [**cryptography removed**](https://github.com/Comfy-Org/ComfyUI/pull/16988): Dropped the cryptography package dependency # Bug Fixes * [**Bypass LoRAs**](https://github.com/Comfy-Org/ComfyUI/pull/16862): Fixed bypass LoRAs on fused Linear paths * [**Qwen 3.5 MTP offload**](https://github.com/Comfy-Org/ComfyUI/pull/16877): Fixed MTP embedding offload * [**MiniMax-H3 single images**](https://github.com/Comfy-Org/ComfyUI/pull/16881): Single images are now decoded as the first latent of a clip * [**Half float video**](https://github.com/Comfy-Org/ComfyUI/pull/16949): Fixed loading half float files with PyAV * [**Dynamic inputs**](https://github.com/Comfy-Org/ComfyUI/pull/16377): Validation and lazy execution now work with dynamic inputs * [**Asset dates**](https://github.com/Comfy-Org/ComfyUI/pull/16810): Scanned assets are dated by their file modification time # Asset Library * [**Enable flag**](https://github.com/Comfy-Org/ComfyUI/pull/16929): The asset system stays off unless ComfyUI starts with `--enable-assets`, and the output folder is scanned again when it is on * [**Windows portable**](https://github.com/Comfy-Org/ComfyUI/pull/16928): The asset system is on by default in the Windows standalone build * [**Asset export API**](https://github.com/Comfy-Org/ComfyUI/pull/16578): The asset export API now works locally * [**Video metadata**](https://github.com/Comfy-Org/ComfyUI/pull/15180): Video metadata is extracted on ingest and scan * [**EXR previews**](https://github.com/Comfy-Org/ComfyUI/pull/16898): Previews for EXR assets, written by `SaveImageAdvanced` for outputs # Developer Experience * [**Prompt metadata**](https://github.com/Comfy-Org/ComfyUI/pull/16763): Attach metadata to a prompt's websocket messages and [preview image frames](https://github.com/Comfy-Org/ComfyUI/pull/16809) * [**Custom node governance**](https://github.com/Comfy-Org/ComfyUI/pull/16167): Governed builds can restrict ComfyUI to approved custom node packs; normal installs are unchanged * [**Content Security Policy**](https://github.com/Comfy-Org/ComfyUI/pull/15565): The CSP now allows `blob:` and `data:` sources # Installation * [**ARM64 NVIDIA portable**](https://github.com/Comfy-Org/ComfyUI/pull/16985): A new Windows portable build for ARM64 devices with NVIDIA GPUs # Workflow Templates * [**Templates v0.11.82**](https://github.com/Comfy-Org/ComfyUI/pull/17001): Updated the bundled workflow templates, with these new templates: * [**Qwen Image 2.1 Turbo: Text to Image**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_turbo_t2i.json): 1024×1024 images in 8 sampling steps, with an optional Qwen 3.5 prompt enhancer * [**Qwen Image 2.1 Turbo: Image Edit**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_turbo_image_edit.json): Edit an image from two reference images with the Turbo checkpoint * [**LTX-2.5 DFR: Image to Video**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_ltx2_5_dfr_i2v.json): Three-stage image-to-video with spatial and optional temporal upscaling, plus audio generated in the same pass * [**Higgsfield Genjutsu: Motion Transfer**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/api_higgsfield_genjutsu_motion_transfer.json): Recast a video from 1 to 8 reference images while keeping its motion, camera movement, timing, and audio * [**Fish Audio Drama 3: Two-Speaker Dialogue**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/api_fishaudio_drama3_dialogue.json): A dialogue scene between two voices in one script * [**Fish Audio Drama 3: Directed Performance**](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/api_fishaudio_drama3_directed.json): A single voice directed with plain-language directions and word tags * The Luma Ray 2 image-to-video and text-to-video templates are deprecated, following the [Ray 2 node deprecation in v0.39.0](https://github.com/Comfy-Org/ComfyUI/pull/16741). The Ray 2 endpoints shut down on October 24
document fetched 2026-10-15 00:45 · canonical https://reddit.com/r/comfyui/comments/1x5nc6x/comfyui_v0400