No/Unclear documentation on use of ComfyAutogrowV3

[zouflain]

comfy_script/runtime/nodes.pyi line 8589 defines

class MiniMaxH3ReferenceToVideo:
    '''```
    def MiniMaxH3ReferenceToVideo(
        clip: Clip,
        vae: Vae,
        audio_vae: Vae,
        prompt: str,
        width: int = 1344,
        height: int = 768,
        length: int = 124,
        ref_image_size: MiniMaxH3ReferenceToVideo.ref_image_size | str = 'match',
        **ref_images: ComfyAutogrowV3** | None = None,
        **ref_videos: ComfyAutogrowV3** | None = None,
        **ref_video_audios: ComfyAutogrowV3** | None = None,
        **ref_audios: ComfyAutogrowV3** | None = None
    ) -> tuple[Conditioning, Latent]
    ```

But no documentation exists on how to actually supply a "ComfyAutogrowV3", the object itself is defined only as "..." and every major AI model cannot find nor generate the correct method to supply images from for example img, mask = LoadImage("file_in_inputs.png").

ImageBatch, Image, etc yield only a JSON serialization error, [img] as well.

[Chaoses-Ib]

Just search ComfyAutogrowV3 in nodes.pyi or reference its example workflow. It's the same way as you'd use it in ComfyUI's web UI.

[zouflain]

Thank you for the quick reply! Unfortunately not at all clear how to invoke it during a workflow. Interpreting what's going on in Nodes.pyi requires intimate knowledge of every aspect of comfy_script.

with workflow():
    conditioning, latent = MiniMaxH3ReferenceToVideo(
        clip=clip,
        vae=video_vae,
        audio_vae=audio_vae,
   
        ref_images=None, # It is not clear what to pass here, or how to correctly assemble it (lets use "example_image.png" and "example_image2.png")
        ref_videos=None,
        ref_audios=None, 

        prompt=details["prompt"],
        width=details.get("width", 864),
        height=details.get("height", 480),
        length=details.get("length", 124)
    )

Note that I have searched rather extensively, both via google and on both Comfyui and Comfy_script's github. There is almost no mention whatsoever of HOW to invoke it, and the definition for AutogrowV3 (not ComfyAutogrowV3 is extremely vague).

[Chaoses-Ib]

Have you used ComfyUI's web UI? The type is literally what input type you would see in it.

There is almost no mention whatsoever of HOW to invoke it, and the definition for AutogrowV3 (not ComfyAutogrowV3 is extremely vague).

That's the same for the web UI. If you don't know the pattern already, you better find a example workflow first.

[zouflain]

The web ui takes parameters that are invalid for the MiniMaxH3ReferenceToVideo invocation.

You cannot pass "ref_image_0" as you would in the web UI
`
img, _ = LoadImage("example.png")

conditioning, latent = MiniMaxH3ReferenceToVideo(
    clip=clip,
    vae=video_vae,
    audio_vae=audio_vae,
    
    # Reference Assets (Pass ComfyUI Image/Video/Audio node outputs here)
    ref_image_0 = img,
    ref_videos=None,       # e.g., [LoadVideo_output]
    ref_audios=None,       # e.g., [LoadAudio_output]

    prompt=details["prompt"],
    width=details.get("width", 864),
    height=details.get("height", 480),
    length=details.get("length", 124)
)`

ComfyUI/comfy_api/latest/_io.py", line 1990, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: MiniMaxH3ReferenceToVideo.execute() got an unexpected keyword argument 'ref_image_0'

So it's really unclear how you mean "use it the way you would in the webUI

[Chaoses-Ib]

Sorry, I haven't updated for a while. ComfyAutogrowV3 is a new special built-in type handled in the web UI. You have to use it like this at the moment:

conditioning, latent = MiniMaxH3ReferenceToVideo(
    clip=clip,
    vae=video_vae,
    audio_vae=audio_vae,
    
    # Reference Assets (Pass ComfyUI Image/Video/Audio node outputs here)
    **{ 'ref_images.ref_image_0': img, 'ref_images.ref_image_1': img1, ... },
    ref_videos=None,       # e.g., [LoadVideo_output]
    ref_audios=None,       # e.g., [LoadAudio_output]

    prompt=details["prompt"],
    width=details.get("width", 864),
    height=details.get("height", 480),
    length=details.get("length", 124)
)

I'll make iterable args like ref_images=[img, img1] work, but it may take one or more days as many modules need update. Are you virtual mode or real mode? I can update the one you are using first.

[zouflain]

Thank you so much! I was using virtual mode at the moment, but if

**{ 'ref_images.ref_image_0': img, 'ref_images.ref_image_1': img1, ... },

is valid, please don't feel any pressure to rush, I can use this syntax if needed. Thanks again for your patience and help!