ComfyUI 0.40 adds camera-angle nodes, LoRA stacking and EXR fixes
ComfyUI v0.40.0, released on GitHub on 13 October, adds CameraAngle and CreateCameraInfo nodes, native LoRA stack loaders, MiniMax H3 image decoding and…
Comfy-Org released ComfyUI v0.40.0 on GitHub on 13 October at 21:29 UTC. The release notes list two camera nodes, native LoRA stack loaders, a MiniMax H3 decoding change and fixes to EXR and half-float file handling. AI Weekly reported the release on 14 October, citing the same GitHub changelog. It is one announcement, not two independent confirmations. The points below come from the changelog's pull-request titles and are not results of our own tests.
Camera nodes and LoRA stacking
A new CameraAngle node outputs a camera_info value and a shot prompt. A separate CreateCameraInfo node, with a CameraInfoState widget input, builds camera_info directly. The notes do not say which video models read camera_info or how the shot prompt is worded. Treat these nodes as a way to define camera parameters inside a workflow, not as confirmed camera control over generated video. That depends on the downstream model. The same release also makes every 3D node report its models as a standard 3D output item.
For LoRA workflows, v0.40.0 adds native LoRA stack loader nodes (tracked as CORE-186). It also fixes an issue with bypassed LoRAs on fused code paths, and LoRAs can now replace a model's per-layer attention config.
Video pipeline and file handling
MiniMax H3 single images are now decoded as the first latent of a clip. Saved EXR files are compressed, and EXR outputs written by SaveImageAdvanced get previews in the assets view. A separate fix covers loading half-float files with PyAV. On ingest and scan, the assets system now extracts video metadata into a system_metadata field. Metadata can also be attached to a prompt's websocket messages and its preview image frames.
New partner nodes cover Higgsfield's Genjutsu Motion Transfer, Vidu Q4, Kling Virtual Try-On, Google's Nano Banana 2.1, FishAudio's Drama 3 and a ClarityAI Crystal Upscaler. HeyGen Video 1.0 nodes gain a 2K resolution option. The notes give no pricing, account requirements or details on where processing runs. A partner node in the graph does not mean the model weights are available for local download.
Assets: AI Weekly's headline overstates the default
AI Weekly's headline says the assets system ships on by default. Its summary is narrower, saying the local asset export API is on by default for the Windows standalone build. The changelog shows how that changed during the cycle. One pull request turned assets on by default and added a --disable-assets flag. A later one turned assets off unless --enable-assets is passed. Another turned assets on for the Windows standalone build. Read in order, the final state appears to be opt-in on most installs and on for Windows standalone. Confirm this against the shipped build before relying on it.
Other changes
The notes list a low-VRAM performance improvement for the Qwen 3.5 LLM and enable MTP with images on Qwen3.5. No benchmark figures are given. The release also adds an ARM64 NVIDIA portable build, removes the cryptography package dependency, and bumps the frontend to 1.55.17 and the workflow templates to v0.11.82.
A pull request titled "Add governance enforcement for custom nodes" (#16167) is also included, but the notes do not say what is enforced or which nodes it affects. Pipelines that depend on third-party custom nodes should read that pull request before upgrading.
Frequently asked questions
Does ComfyUI v0.40.0 add camera control for AI video?
ComfyUI v0.40.0, released on GitHub on 13 October, adds a CameraAngle node that outputs a camera_info value and a shot prompt. It also adds a CreateCameraInfo node that builds camera_info directly. The release notes do not say which video models read camera_info. The nodes define camera parameters inside a workflow. Whether they control the camera in generated video depends on the downstream model.
What does ComfyUI v0.40.0 change for LoRA workflows?
ComfyUI v0.40.0 adds native LoRA stack loader nodes, tracked as CORE-186. According to the changelog, it also fixes an issue with bypassed LoRAs on fused code paths and lets LoRAs replace a model's per-layer attention config.
Is the assets system on by default in ComfyUI v0.40.0?
The ComfyUI v0.40.0 changelog suggests the assets system is opt-in on most installs, enabled with the --enable-assets flag, and on by default for the Windows standalone build. This reading comes from the order of pull requests: one turned assets on by default, a later one made them opt-in, and another turned them on for Windows standalone. AI Weekly's 14 October headline describes assets as on by default, which overstates the change. Confirm the behaviour against the shipped build before relying on it.
AI-assisted, reviewed by Codex (25-post review authorized by JJ Rooney, 14 October 2026) on 14 October 2026.
