ComfyUI 0.22: audio, panorama and Hydit workflow review

This review tests ComfyUI 0.22 templates that run models locally, excluding workflows that require paid tokens or API connections. The presenter shows how to filter ComfyUI templates to find recent additions, then tries audio generation and several image workflows.

In the Stable Audio 3 tests, generation finishes despite reported errors. The presenter finds the music repetitive and lacking complexity, and says the tested models produce music without vocals.

Qwen panorama generation produces apartment interiors the presenter considers useful for VR and 3D environments. Tests with science fiction scenes and people fare worse. Resolution is another constraint: a 2K panorama leaves relatively few pixels in a narrow viewing angle. Converting a panorama into geometry produces visible mesh and texture breaks, so the presenter treats that result as experimental. A separate image workflow extracts depth, color and normal maps for further editing or rendering work.

The Hydit tests have mixed results. Loading a custom image into the FP8 workflow produces an unresolved grid artifact. The other workflow successfully upscales an image and adds prompted fuzzy fur. The transcript calls this model FP16, while the supplied chapter title calls it FP6. These are observations from the presenter's tests, without hardware requirements or a confirmed fix for the grid issue.