Rare shader output is previous frame.

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

Occassionaly the shader will output the previous output image it has been passed not the current one.

Something going on with the frame buffer?

eg:-
a goes in modified a is output
b goes in previous a output is output
c goes in modified c is output

the output is unmodified so the shader isn't reusing the input image, it is bypassing the code and outputting the last frames output.

Happens ~1% of the time

most of the time this isn't much of a problem as you can just generate again, but with long sequences...

https://www.youtube.com/watch?v=axBK9pEOMLQ

it is stopping and starting occasionally because of this.

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

video is 16x slo-mo btw.

[Amorano]

Cool video =D

I think I would need to see the workflow (some cut out version) to figure this out -- or maybe some more information on how many frames? you are generating? Hard to trace without more information.

Technically there is no hold out for the buffer, so, if its getting its "previous" frame that could simply be the ComfyUI execution engine which caches outputs if the inputs don't change -- I do "force" the output, even if the inputs "don't change" but that is probably where I would start to trace this.

I will take more of a look tomorrow.

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

You will also need (though you could bypass it by giving the depth input the diffuse), the lotus depth nodes and model. If comfy doesn't auto download the model... https://github.com/EnVision-Research/Lotus or... Links to huggingface post. "The Lotus disparity models (Generative<https://huggingface.co/jingheya/lotus-depth-g-v2-0-disparity> & Discriminative<https://huggingface.co/jingheya/lotus-depth-d-v2-0-disparity>) are now available, which achieve better performance!" (these are the depth) These output images were in sequence, it took ~30 tries before it borked (hence the load batch), This also occurs when there is just one shader (movie workflow not included) or using load image to select them one by one, the output will be the same as the last output.

…

________________________________ From: Alexander G. Morano ***@***.***> Sent: 09 February 2025 07:45 To: Amorano/Jovi_GLSL ***@***.***> Cc: Z22 ***@***.***>; Author ***@***.***> Subject: Re: [Amorano/Jovi_GLSL] Rare shader output is previous frame. (Issue #7) Cool video =D I think I would need to see the workflow (some cut out version) to figure this out -- or maybe some more information on how many frames? you are generating? Hard to trace without more information. Technically there is no hold out for the buffer, so, if its getting its "previous" frame that could simply be the ComfyUI execution engine which caches outputs if the inputs don't change -- I do "force" the output, even if the inputs "don't change" but that is probably where I would start to trace this. I will take more of a look tomorrow. — Reply to this email directly, view it on GitHub<#7 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BKS64VTSFEHECSK7G3LYFDL2O4BSRAVCNFSM6AAAAABWYOPUAGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBWGEYTCMBSHE>. You are receiving this because you authored the thread.

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

email kept wigging out about file size so here are the example images, they should all be a stereo pair.

[Amorano]

You will also need (though you could bypass it by giving the depth input the diffuse), the lotus depth nodes and model. If comfy doesn't auto download the model...

I think I missed something. I dont need a model -- I need a workflow. If the model itself is the problem, then that is the model's problem.

I need a workflow that makes the "skips" otherwise I cant count that as a problem for my nodes?

Was there some other part of the message that got clipped?

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

It was attached the email along with the shaders, will post it on git comment rather than send a email.

…

________________________________ From: Alexander G. Morano ***@***.***> Sent: 10 February 2025 08:34 To: Amorano/Jovi_GLSL ***@***.***> Cc: Z22 ***@***.***>; Author ***@***.***> Subject: Re: [Amorano/Jovi_GLSL] Rare shader output is previous frame. (Issue #7) You will also need (though you could bypass it by giving the depth input the diffuse), the lotus depth nodes and model. If comfy doesn't auto download the model... I think I missed something. I dont need a model -- I need a workflow. If the model itself is the problem, then that is the model's problem. I need a workflow that makes the "skips" otherwise I cant count that as a problem for my nodes? Was there some other part of the message that got clipped? — Reply to this email directly, view it on GitHub<#7 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BKS64VS66TQZZZHQ5AK6F2L2PBQBZAVCNFSM6AAAAABWYOPUAGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBXGI3DSMJYGU>. You are receiving this because you authored the thread.

[plebghudmgfujygthmjdmfgjsdrgjsdrt]

BugTest.zip