PBRify - Fast, Ethical Upscaler & PBR Generator!

[limpid flame]

ok coo we on same page lol

[vagrant moon]

only thing to clarify

the zip would contain the ComfyUI folder, and they just drag it into the directory above it?

PBRify_Remix_ComfyUI_ONLY.zip\ComfyUI\models\upscale_models

[limpid flame]

ye

[vagrant moon]

i actually don't think i'm gonna do this part

i feel like it'd be easier to screw up the dragging part than placing it in an exact directory that i specify

[limpid flame]

makes sense

you do have that advantage when i patch i change alot of stuff lol

[vagrant moon]

yeah i'm sure lol

[limpid flame]

for the seesr update it will require a full redownload i had to change the whole environment

[vagrant moon]

is there a way to clear or empty a node?

like that

i don't want that to be in the saved template

just closing and reopening the tab clears it

@limpid flame alright there's only one issue left now, lol. it's the same one chaiNNer had

if there's no alpha layer, this workflow will add one

is there any way you know of to disable alphas if they're not actually present?

[limpid flame]

i will look into it

since comfy will error if the return type is not filled, it may become a Boolean as well and bypass merge if not present

ig default image if none could be full white?

[vagrant moon]

afaik that's what it does already

separating the generated alpha and viewing it = white image

[limpid flame]

o

will have to experiment with it can yo usend that test image

[limpid flame]

well i mean in comfy

[vagrant moon]

ya

[limpid flame]

i think its set to black

[vagrant moon]

this is the result from comfy

[limpid flame]

but direct from get textures node

[vagrant moon]

ah

input

output in Remix

it is black @limpid flame

not sure how it turns white lol

current workflow

[limpid flame]

ty i will work on it this weekend

[vagrant moon]

thank ya

i'll upload this workflow separately and add it to my steps for now, if anyone wants to use it

i don't want to bug Nvidia about it until it's finalized lol

[quaint jungle]

It looks scary but I'll give it a go lol

[vagrant moon]

i simplified it even more for ya

it's all in these steps now: https://github.com/Kim2091/PBRify_Remix?tab=readme-ov-file#comfyui

i linked the latest alpha compatible workflow at the relevant step

[quaint jungle]

It's the comfyui stuff that looks confusing to me lol. But I'm sure I'll figure it out when I mess with it

[limpid flame]

wdym this looks p ez to me

cut the green wire

[vagrant moon]

ya i understand

thankfully though, you don't have to mess with comfy at all other than launching it and pressing "queue prompt"

[quaint jungle]

as part of the new comfyUI stuff the textures get assigned automatically right?

@vagrant moon

[vagrant moon]

yep

[quaint jungle]

[vagrant moon]

you just select the texture in the toolkit and press queue, then it'll upscale and assign all of them

[quaint jungle]

is there a way to specify layer so I can have an "AI-Placeholders" one I can toggle

[vagrant moon]

the ONLY issue is that you have to manually modify displacement afterward every time, and it defaults to a full strength of 1

[quaint jungle]

nvm

[quaint jungle]

not gonna use height so that's fine

unless

[vagrant moon]

i mean, it's the same output as before

it's just on by default here

[quaint jungle]

gotcha

well I'm gonna try out the latest builds and see how that goes

of remix I mean. Been having issues with RC lately and those perf improvements would be nice too

[limpid flame]

@vagrant moon can you test this commit before i pull

https://github.com/Night1099/ComfyUI-RTX-Remix

[vagrant moon]

sure. what am i looking for?

[limpid flame]

fix for alpha returning as opaque by defauilt

[vagrant moon]

hm

422 Client Error: Unprocessable Entity for url: http://127.0.0.1:8011/ingestcraft/mass-validator/queue/material

  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_data
    return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\execution.py", line 74, in map_node_over_list
    results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-RTX-Remix\nodes\common.py", line 63, in wrapper
    return (instance.context,) + self.func(instance, *args, **kwargs)
                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-RTX-Remix\nodes\ingestion.py", line 139, in ingest_texture
    check_response_status_code(r)
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-RTX-Remix\nodes\utils.py", line 58, in check_response_status_code
    raise err
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-RTX-Remix\nodes\utils.py", line 54, in check_response_status_code
    response.raise_for_status()
  File "C:\Users\Kim\Projects\Tools\ComfyUI_windows_portable\python_embeded\Lib\site-packages\requests\models.py", line 1024, in raise_for_status
    raise HTTPError(http_error_msg, response=self)```

[limpid flame]

ok hold up my fork wasnt synced up one sec

ok now it should be good sorry

[vagrant moon]

no worries

[vagrant moon]

same error unfortunately

i noticed it doesn't preview the texture at any stage in the pipeline

[limpid flame]

how are you updating it

i dont have any errors

[vagrant moon]

i deleted the entire ComfyUI-RTX-Remix node folder and git cloned the new version

[limpid flame]

and restarted comfy i assume

[vagrant moon]

yes

[limpid flame]

hmmmmmmm

whats odd is that error message doesnt touch the file i edited lol

do you have remix open 😝

[vagrant moon]

i do

huh

so i opened my older Barnyard project (the one i'd been using for about 4 months) and it's just not working in there. it just worked a couple nights ago with the normal nodes though

opened a newer project and it's fine

[limpid flame]

wonder if you reclone base repo is it same error

also why is my commit showing a whole file change im annoyed

[vagrant moon]

the error isn't there

actually, it is

but it previews the changes in the chain

ah i think i know. you said you set it to full transparency all the time? wouldn't that have the effect of textures being completely invisible?

[vagrant moon]

i'm just extremely confused. it was just working 2 days ago

[limpid flame]

o

[vagrant moon]

but it seems that it's not due to your changes

[limpid flame]

wait so what change are you requesting

[vagrant moon]

not to run the upscale process for alpha if there's no alpha layer on the input texture

if i put in an RGB texture, it spits out RGBA with a white alpha layer

waste of space + it can cause issues + wasted time upscaling nothing

[limpid flame]

ah, that'd have to be a custom bool bypass thing hmmm

[vagrant moon]

i'm guessing that's too much to add?

realistically the extra alpha layer isn't that bad, it's just annoying to deal with

[limpid flame]

No not hard, but would be a custom node, or if we can add it to their node pack

[peak heart]

There anyway to add a check if the file already exists and skip it?

[vagrant moon]

what method are you using? chaiNNer? or REST API

[proud robin]

Not sure exactly what happened but my ai_upscaled.usda just deleted itself, all the ingested assets are still there. My remix was running slow during the ComfyUI queue so I saved mid upscale, then on next open it hints ""..sublayer path is invalid and was cleaned up /aiupscale_textures.usda"

Double checked the project folder and confirmed its gone lol, any way to restore a backup?

[vagrant moon]

😦

unfortunately if it's gone, i don't know how you could possibly restore it

that's moreso a question for @peak dagger

[proud robin]

Thankfully its just Delta Force, the captures are only a couple hundred assets, lesson learned though haha

save often, just beta things

[vagrant moon]

good to know for sure lol. i'll back up mine more often

sorry that happened though

[proud robin]

Its cool but i was like damn it's really gone haha

I could only imagine how someone would react on a bigger capture

[vagrant moon]

oh btw, you should use the workflow file here: https://github.com/Kim2091/ComfyUI-RTX-Remix/raw/main/workflows/rtx_remix_pbrify_remix_workflow.png

it supports transparency on textures

[proud robin]

Yeah after getting over the initial hurdles the REST PBRify stuff is pretty neat

I will say a simple 1024 x 1024 couch texture defeated it though 😃

[vagrant moon]

yeah... that's due to the way comfyui handles some things related to upscaling unfortunately

the original chaiNNer version is far more optimized, but is less feature rich

[proud robin]

I do like the ease of use, at first it was intimidating with all the stuff out there but its actually really easy

[vagrant moon]

if you want to have far faster + lighter weight upscaling, you can switch to the SPANV4 model for upscaling instead of the DAT2 model

it's right here in the chain

just change the top one to the same as the bottom one

quality is definitely worse, but not drastically so

[proud robin]

Is there any models that can get higher than 4X? The game I'm working with has some tiny textures and I'd love to blast 4K at it for fun

[vagrant moon]

none that are worth using. anything above 4x with these types of models is a waste unfortunately

they just can't reconstruct that much detail

[proud robin]

Gotcha, the current models do some good work with the 64 x 64 terrain textures from Delta Force as is

[vagrant moon]

🙂 i'm glad

the DAT2 model is a lot more capable for that kind of thing, but also uses significantly more VRAM and is roughly 10x slower than the SPANV4 model

i'm working on an upgrade right now that should have the same (or better) quality as DAT2, while being much much faster

hopefully it'll just be a week from now

anyway, i'm happy this is working for ya. afaik you're the first person so far to use the REST API version extensively

[proud robin]

Yeah I was actually using it across a whole level and it beats the hell out of manually exporting and re-ingesting
Also I just started using remix a few days ago too lol

Thanks for the work you put into all this, definitely some amazing groundwork for something great

[peak heart]

Chainner

Better late than never lol.

[vagrant moon]

then yes. in the "Save Image" nodes (they're buried in each stack, sorry), there's a setting to skip images that exist

however this won't prevent the upscaling process for those existing images unfortunately

[peak dagger]

Normally if we clean anything up all we do is remove the reference in the stage, we don't delete the file, so that most likely came from the ComfyUI graph or another source if the file was actually deleted

[nova hound]

Hi!

I noticed PBRIfy and PBRFusion 2 produce different roughness maps; in fact, they seem to be inverted when compared to eachother. So either PBRify or PBRFusion2 is producing inverted roughness maps?

ie, left side is PBRify and right side is PBRFusion2.

Here's a comparison between the two (scroll down for roughness maps)
https://onnoj.github.io/DeusExEchelonRendererPages/pbrcomparison/?setA=01_pbrify&setB=02_pbrfusion2

Here's an in-game comparison of PBRify and PBRify with inverted roughness maps:
https://onnoj.github.io/DeusExEchelonRendererPages/comparison/?setA=13_v0.3.9_pbrify&setB=13_v0.3.9_pbrifyinvroughness

[peak dagger]

PBRify seems closer to what I would expect artists to produce in your comparison 😛

[vagrant moon]

that's a good question. the difference simply comes down to how the model is interpreting roughness

PBRFusion is generally more robust with PBR generation and usually does a better job interpreting these things. it's based on diffusion and is a lot more capable than the types of models i'm using

[peak dagger]

In the comparison look here and you'll see the "inverted" PBRify textures are more accurate

[nova hound]

Yeah, but scroll down

[peak dagger]

Left is PBRify and Right is PBR Fusion

[peak dagger]

Oh there's more !

[vagrant moon]

yeah i was just about to say you can scroll lol

i just noticed

that's a pretty cool website

[peak dagger]

Yeah didn't know that was a thing either!

[vagrant moon]

PBRFusion generally does a better job imo. PBRify's roughness generation really struggles when an input has a dark portion, it'll just interpret it as smooth no matter what

[peak dagger]

I would say generally looks like PBRify just likes to make things shinier. Which is not bad, because you can just use Photoshop or any image altering software after and just reduce the exposure to make things more mat if you want

[nova hound]

versus

[vagrant moon]

i'm going to be training a new set of models coming up, hopefully they're more robust

[nova hound]

So far I feel that they both excel at different things

oh wait, I misread, sorry

[limpid flame]

Roughness has been the hardest map to produce because it really does require the model to know what it's looking at, I will say I'm not happy with the current outputs and have artificially made them to be more rough then not

[vagrant moon]

haha that's what i was doing initially too

roughness maps are really difficult

[nova hound]

yeah, one comparison is PBRIfy versus PBRify w/inverted roughness (where I think inverted roughness looks better). The screenshot I posted is from PBRify versus PBRFusion 2

[vagrant moon]

PBRFusion's PBR maps tend to be more defined. so although Night is adjusting them to be more rough after it's generated, it still has better highlights than mine

this is a sample from the PBRFusion screenshot

mine isn't able to reproduce the reflectivity on the edges quite as well

i'm going to be retraining the entire set of models on a new architecture with an entire new (and much better) dataset. it should improve this

[vagrant moon]

but yeah. thankfully just editing it a bit in post should produce a good result

[limpid flame]

I agree with this ppl like to use it as a batch replacement(which is fine), in practice i use as a tool, editing the outputs in post

As it can definitely get things very wrong

[vagrant moon]

yep. any time i use something like this, i edit it afterward to improve it. it's less than a minute per edit a lot of the time, but drastically improves the quality 🙂

[limpid flame]

Yep even as simple as an invert

[vagrant moon]

ya. some of the inverted outputs Onno shared are more correct, but others aren't. i do wish it was as simple as just running this and having the outputs be correct 100% of the time, but i think that'll be quite a while away

[limpid flame]

This would be a long project but, since I have prompt model in the workflow now, I could pass the clip to the controlnet for roughness trained on correct clip of prompts for what it should do for roughness in different materials to get it more correct?

I have kind of disregarded the power of prompts in the models

But also to be fair a model that recognizes them correctly has only been a recent development

[vagrant moon]

yeah. tbh i'd probably wait a bit longer until there's a model that can reliably detect it 95% of the time

but having that would pretty much render PBRFusion the go-to 😛

[limpid flame]

It'd be too powerful, even with clip correct like that though, the adherence to color bias would still be stronger, I'm going to start testing on flux soon when controlnet training has been flushed out

[vagrant moon]

ah. seems there are more factors than i realized lol

[limpid flame]

There always is 👀

I'm confused is he saying that that brick wall shouldn't be rough or is he referring to the game screenshots lol

[vagrant moon]

i think he was asking which is more correct for the actual map

[limpid flame]

Gotcha the embed preview threw me off lol

[vagrant moon]

great example of PBRify just making anything that was black in the albedo shiny

really wish i could find a way around that

[limpid flame]

Color bias has been extremely annoying, the only way I even began to train it out was with 1 million image pairs 👀

And it still has some

[vagrant moon]

no matter what i did, i wasn't able to remove it

then again i didn't try even 50,000 pairs lol

i only had about 3000 to work with

[limpid flame]

It's okay even depth anything v2 has some color bias

And that's trained with actual depth flags lol

[peak dagger]

Screenshots only! The overall map look more appealing to me because of the nice reflections 😛 Regardless of the expected physical look haha! Definitely not "accurate" but more eye-candy looking map

This map being one of the clearest examples of what I mean

[vagrant moon]

makes sense 😛

[proud robin]

Thanks for the insight, after a few days of testing it I found that when my system gets choked for VRAM in the middle of an upscale using the DAT2 upscale model ComfyUI somehow deletes my aiupscaled usda file, can confirm it happens on the 'Upscale texture (using model)' stage. I have not had this issue when using SPANV4 though, while using the SPANV4 model I can choke the VRAM down to where remix says I have literally 0 Bytes left of VRAM and it never causes the slowdown / aiupscaled usda delete. The DAT2 related crashes seemed to happen more on the PBRify with alpha support workflow, the same workflow with SPANV4 for both models has yet to cause issues though

[vagrant moon]

i think the reason it'd happen more is because it's running two different models right after each other, which would increase VRAM load

but to me that absolutely sounds like a bug in ComfyUI or in the addons. nothing i modified in the workflow could be causing this. all i did was set it to use an upscaling model rather than resizing for the diffuse

[proud robin]

Who knows, I have a feeling its not caused by Remix but again just a feeling. Something definitely eats VRAM like crazy in Remix after about 40 or 50 prompts. I'll go from almost 6 GB free down to literal bytes until I close and re open Remix, which it goes back to eating VRAM by the time I've done another 10 or 15 prompts

I'll definitely have to test this on another game / project before I can say what the cause is

[vagrant moon]

well, no matter what, that file shouldn't be deleted

i don't know what could be causing it. maybe an error finishing a write to the file and it's just... gone?

[proud robin]

Is there any way I can log this? ComfyUI logs seemed pretty unhelpful

[vagrant moon]

to clarify, do you mean restarting comfyui? or the remix toolkit

[proud robin]

Just the remix toolkit, all of the used VRAM comes back after relaunching the toolkit, without closing ComfyUI

[vagrant moon]

huh

i'll pay more attention to VRAM load next time i test this

[vagrant moon]

if you run Remix from the lightspeed.app.trex.bat file in %LocalAppData%\ov\pkg\rtx-remix-2024.4.1, it'll open a terminal window with a log

[proud robin]

Got it, I'll see if I can generate anything useful, might take a little while. Also in case anyone runs into the same thing: If you notice the slowdown / VRAM eating with the project open and know it deleted the aiupscaled usda, don't close the project or you'll lose the layer, you can export the aiupscaled layer to another folder and manually re-add it to the project folder

[quaint jungle]

Going off the embed here this is the wrong way of going about it. PBRify is actually super accurate here in terms of contrast, but not brightness. Inverting it actually is making the difference between the brick and gaps more inaccurate though the overall brightness is more accurate to a real brick roughness map. If you took PBRify's output there and ran it through levels/brightness adjustment to bias it more towards white, you'd get the best possible result

[vagrant moon]

A new version is WIP. Thanks to the donations I received, I was able to purchase additional content to use in the dataset.

In addition, @limpid flame helped (did most of it lol) put together a script for InstaMAT (texture design software) to bake textures. This will greatly improve PBRify's ability to handle and interpret degraded game textures.

Previously PBRify was trained only on albedos, which are textures with no lighting information. Baked textures contain all of the lighting information in a single image.

This is an example from the new dataset:

Albedo | Baked

I plan to use MoSR (a newer, faster architecture) for the new models. I will be (hopefully) retraining the entire PBRify set, PBR generation and upscaling. I also intend to give an unbaking model another go, now that the baked texture generation is more robust 🙂

[whole shadow]

good to see that the pbrify is being updated

[sweet dagger]

WOHOOOO!! Nice work Kim!!

[vagrant moon]

it seems like my AI layer was deleted too @peak dagger

[peak dagger]

That simply means the layer didn't exist on your HDD when you loaded your project

Something external most likely deleted the layer before you reloaded the project. Are you using a ComfyUI graph?

[vagrant moon]

i am, using the PBRify one on the official repository

however, i saw nothing in the comfyui logs about it

Matt above said this:

Thanks for the insight, after a few days of testing it I found that when my system gets choked for VRAM in the middle of an upscale using the DAT2 upscale model ComfyUI somehow deletes my aiupscaled usda file, can confirm it happens on the 'Upscale texture (using model)' stage.

i haven't been able to confirm it, but it makes sense. so it's likely a bug of some sort in comfy 😦

[peak dagger]

I think it's most likely the design of the ComfyUI graph. If you look there's probably a Delete Layer node

It was probably executed for some reason and because the generation didn't succeed, you never got the updated texture in the new layer in the end

[vagrant moon]

lol you're right. it's at the beginning of the chain

it seems to be executing all the time though

and i'm not sure why it's enabled by default 😦

i'll make a PR tomorrow to disable it. thanks Pete

[peak dagger]

Np!

For reference I think our demo graph had that so that you would fetch the textures from the capture layer and not the auto upscale layer.

If you want to be able to execute the graph multiple times you'll probably need a similar system, muting layers would probably work too

[vagrant moon]

hi Pete. i'm sorry but i don't really know what you mean with that last part. i'm having a hard time wrapping my head around this 😦

would this retain the layer, and insert any new modifications into the existing layer?

[peak dagger]

Your graph will have a Get Textures node. That grabs the textures from the stage in Remix, but if some textures were replaced, it'll grab those. So you want to make sure you only grab the textures that come from the capture layer, right?

[peak dagger]

You can probably simplify that by never deleting layers, and only creating a layer of it doesn't exist

That would also be "safer"

[vagrant moon]

hm

i believe this is correct then?

i have the REST API documentation, but i don't see any for the ComfyUI nodes in particular other than the descriptions in the readme

[vagrant moon]

so this is the top part connecting to if_false

so i'd set it to create instead of insert

[peak dagger]

ComfyUI basically only calls the REST API so it should be quite close

[peak dagger]

This mostly okay, just just need the switcher to make sure you only create the layer if it doesn't exist

[vagrant moon]

right. i believe i have it set up correctly now

i'll give it a test 🙂

yep! it all works

i restarted comfy and the toolkit after upscaling and the layer is retained

it also made a new one successfully 🙂

thank you for the help

[peak dagger]

Np!

[proud robin]

Good to know I'm not crazy lol, have the workflows been updated on the ComfyUI-RTX repo?

[vagrant moon]

i made a pull request, but it hasn't been merged yet

for now, you can use this one: https://media.githubusercontent.com/media/Kim2091/ComfyUI-RTX-Remix/refs/heads/main/workflows/rtx_remix_pbrify_remix_workflow.png

it should work 🙂

[proud robin]

Sweet, thanks again for helping with this, you and NV_Pete are legends

[vagrant moon]

new release tomorrow! it'll only be one model, a new upscaler that's roughly 8x faster than the current one, but with superior quality

the improvements of that model cascade to PBR generation as well. it all looks a decent bit better 🙂

[vagrant moon]

PBRify 1.7.1 Update

Download
ComfyUI Only Version

This update brings a new upscaling model, 4x-PBRify_RPLKSRd_V3. This model is roughly 8x faster than the current DAT2 model, while being higher quality. It produces far more natural detail, resolves lines and edges more smoothly, and cleans up compression artifacts better.

As a result of those improvements, PBR is also much improved. It tends to be clearer with less defined artifacts.

Download use it in chaiNNer, download this release: https://github.com/chaiNNer-org/chaiNNer-nightly/releases/tag/2024-10-29
If you want to use it in ComfyUI, follow the steps below

Comparisons

---

Important

The current version of ComfyUI at the time of writing hasn't been updated to the latest spandrel library.

To update ComfyUI to be compatible with the new model:

1. Navigate to your ComfyUI installation directory
2. go into python_embedded
3. Run this command: .\python.exe -m pip install spandrel --upgrade

🚀 If you like my work, please support me on Ko-fi! 🚀

[whole shadow]

Holy shit

Time to download and test it

[vagrant moon]

it's also far lighter on VRAM, so hopefully there will be no issues with your more limited amount

for a 512x512 texture it should only use ~1.5 GB vs the 4-5 GB DAT2 required

[sweet dagger]

Uh roh, I did a clean install of comfyui, the ComfyUI only version doesn't have the workflow.png

I accidentally had the old workflow that deleted the ai_upscale layer if the upscale was interrupted halfway through

but it's no worries - I wasn't far in progress lol

[vagrant moon]

it's not supposed to. it's packaged in with the comfy nodes provided by Nvidia

i also have steps to install it properly here: https://github.com/Kim2091/PBRify_Remix?tab=readme-ov-file#comfyui

[sweet dagger]

Ahhhh that's right, I forgot nvidia made one, thank youu

[peak dagger]

Fun fact you can also use the ComfyUI Manager to install the Remix nodes & it'll copy over the workflows as well

[vagrant moon]

ah cool

[sweet dagger]

Oh how I love software

thanks!

[peak dagger]

Yeah It should hopefully simplify your install instructions 😛

[sweet dagger]

Ohh cool, it also helps keep things up to date

[peak dagger]

Yeah and if you open a workflow for which you're missing nodes it can "normally" download all the missing nodes for you

It can also download some models but I've found it to be less reliable for that

[vagrant moon]

that's great

[sweet dagger]

Oh wait I lied, it was the the pbrfusion workflow that deleted the layer.

If you get the chance, do you think you could assist Nightraven to adjust the delete layer to how you fixed it kim?

[vagrant moon]

sure

[vagrant moon]

could you send the workflow so i don't need to download the whole zip?

[sweet dagger]

Yeah!

[vagrant moon]

looks like i can't work with it as i'm missing some nodes

and i don't have enough space right now for the download 😦

[sweet dagger]

That's ok! I tried adjusting it earlier but I failed If I make it an exact match to how PBRify is setup would it work?

[vagrant moon]

the main changes to mine were in the first block

this one

i'm not sure if night based it on the PBRify workflow or if he made it from scratch

[sweet dagger]

Ahh looks like it was! Got it all done/tested, it seems to work Does it just add to your existing aiupscale_textures.usda layer and not create something like aiupscale_textrues_01 or anything like that?

[vagrant moon]

yeah. once an ai layer is there, it'll just open and add to it 🙂

[sweet dagger]

Hmm, I went to "batch" ingest mym entire scene by turning off the return_selection on the Get Textures node, but got this error, not sure why

[limpid flame]

Def a spandrel problem

[sweet dagger]

Maybe updating pressing the "Update All" on the ComfyUI manager menu wasn't a good idea

[limpid flame]

O

def dont do that on pbrfusion lol

here it shouldnt have mattered

[sweet dagger]

I'll clean install them both and see if that fixes it - and not press update

[vagrant moon]

did you update spandrel?

[sweet dagger]

Ahhhhhh, I figured the "update all" was going to do all that for me (scary it doesn't), manually doing that step fixed it, thanks y'all

[limpid flame]

@vagrant moon you could try deepbumps normal to height via MTB node pack

[vagrant moon]

o.o

[vagrant moon]

i can't find that node in their node pack

at least the info on their wiki or in the code

[limpid flame]

https://github.com/melMass/comfy_mtb/wiki/nodes-deep-bump

[vagrant moon]

unfortunately that'd interfere with the "ethical" thing

[whole shadow]

Where can i get the latest Alpha version of chainner?

[vagrant moon]

https://github.com/chaiNNer-org/chaiNNer-nightly/releases

[proud robin]

The fact it made it look like an actual ceiling tile is amazing, this stuff drops my jaw every time

This is with the default 0.05 height map as well

[vagrant moon]

that's a great result 🙂

[jolly quest]

Is there a simple way to batch process bmp files to just upscale them with no map generation with the latest version of PBRify? Or should I just use the older model and chainner for now?

[vagrant moon]

mhm, hang on

First install the latest chaiNNer: https://github.com/chaiNNer-org/chaiNNer-nightly/releases/download/2024-07-19/chaiNNer-windows-latest-nightly.2024-07-19.exe

1. Download this: https://github.com/chaiNNer-org/spandrel/archive/refs/tags/v0.4.0.zip
2. Open it
3. Navigate to %AppData%\chaiNNer\python\python\Lib\site-packages\spandrel
4. In the zip file, navigate to spandrel-0.4.0.zip\spandrel-0.4.0\libs\spandrel\spandrel\
5. Extract the contents to the spandrel folder in the chaiNNer install
6. In the zip file, navigate to spandrel-0.4.0.zip\spandrel-0.4.0\libs\spandrel_extra_arches\spandrel_extra_arches\
7. Extract the contents to the spandrel_extra_arches folder in the chaiNNer install

overwrite if it prompts, then start chaiNNer & use the attached chain file

if you don't want to do all of that, you will just have to use the DAT2 model instead. it's slower and doesn't look as good though

[jolly quest]

Thanks!

[vagrant moon]

https://github.com/chaiNNer-org/chaiNNer-nightly/releases/tag/2024-10-29

latest chaiNNer supports the new PBRify models

@jolly quest if you didn't do the steps above already

[whole shadow]

Finally

[ancient loom]

holy crap

what

i come back to this after a while

i do an upscale and wonder "uhh did something go wrong?" because it did it in like a second

no it just kinda actually upscaled in a second

[vagrant moon]

yep 🙂

new model is just that good

[ancient loom]

one of the best things about having adhd is with stuff like this

i lose interest for a while, i come back and there's so much that's improved since the last time i used it

[craggy sparrow]

Hi

I have some issues with chaiNNer

Traceback (most recent call last):
File "C:\Users\Hosnt\AppData\Local\chaiNNer\app-0.24.1\resources\src\process.py", line 432, in process
return await self.__process(node)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Hosnt\AppData\Local\chaiNNer\app-0.24.1\resources\src\process.py", line 602, in __process
for fn in context.node_cleanup_fns:
RuntimeError: Set changed size during iteration

Would someone help?

[vagrant moon]

sadly i have no idea what would be doing that

[craggy sparrow]

Reinstalled to nightly version, same:
Traceback (most recent call last):
File "C:\Users\Hosnt\AppData\Local\chaiNNer\app-0.24.2-nightly2024-10-29\resources\src\process.py", line 466, in process
return await self.__process(node, perform_cache)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Hosnt\AppData\Local\chaiNNer\app-0.24.2-nightly2024-10-29\resources\src\process.py", line 640, in __process
for fn in context.node_cleanup_fns:
RuntimeError: Set changed size during iteration

[pastel temple]

you unzipped it before opening it ?

[craggy sparrow]

xD

[sweet dagger]

oh my god

I realized I didn't have hardware acceleration on in chainner

I turned it on and look how fast it's going

[vagrant moon]

😛 that only affects the GUI

the actual models should run the same

[sweet dagger]

Ohhhh haha, it's so incredibly fast Lightspeed

[vagrant moon]

and chaiNNer is the bottleneck here still. if i ever have the courage i'll make a custom script to do all of these things and it'll probably be 10x faster

[sweet dagger]

Instantanious upscales/pbr. It'd probably be the fastest one out there right now, they're surprisingly rare

[vagrant moon]

okay unless you have 64 GB of RAM, i'd recommend against selecting everything in a scene and upscaling it with the comfyui workflow 😛

chaiNNer workflow is definitely better for this

it's taking about 30 minutes and it ate 50 GB RAM

the ingestion process is the slowest part

python is from comfy, kit is the toolkit

[steep coyote]

oof

pbrify is the easiest and fastest, i select so much stuff before clicking start 😄

do have 64g tho

[flint slate]

Hey Kim, and anyone else who could help
Could you please take a look at this 17 seconds vid and explain what is happening here? Am I doing something wrong? The question is, shouldn't I notice any difference in the toolkit right away?
It's the first time I'm using ComfyUI so please excuse me for the skill issue
https://youtu.be/FVqCzokuLvg

I'm running PBRify 1.7.1 and NVIDIA App release of RTX Remix

[steep coyote]

yea it should update the texture in the toolkit

thats less than 1 second for texture upscale? takes at least 10 seconds for me if not a minute

looks like its not doing the upscaling at all perhaps

upscaled version should show up here

do any of the nodes have a red border?

ahh i do see one has a red border

this one

@flint slate change it from dat2 to rkplsd or whatever

[flint slate]

Yes! It's working now

Thank you!

[steep coyote]

awesome! 😄 maybe we should collaborate on texturing PK ?

[flint slate]

I'd be glad to help if you told me what to do 🙂
I have a little time though

[vagrant moon]

yeah i need to push an update to Nvidia's repository for the workflow

just noticed that issue yesterday

[vagrant moon]

someone wanna sanity check please?

https://github.com/Kim2091/ComfyUI-RTX-Remix/raw/refs/heads/main/workflows/rtx_remix_pbrify_remix_workflow.png

this workflow should fix the issue

well that's not right

[peak dagger]

Height is a 1 channel image so it's probably just populating R in RGB. You could create a ticket for that. I thought we fixed that a long time ago but apparently not

[vagrant moon]

yeah you're right. it's proper grayscale in other tools

will do 🙂

[peak dagger]

Thank you!

[vagrant moon]

now for the second mystery

why isn't displacement working at all in the renderer?

[peak dagger]

In the toolkit ?

[vagrant moon]

yeah

it worked in another project, but not this one

[peak dagger]

Could be differing MDL versions ?

[vagrant moon]

this is the newer project, so i'd hope that isn't the issue

just made it yesterday to test

[peak dagger]

Fresh captures as well ?

[vagrant moon]

yep

[peak dagger]

it's possible the public toolkit & public runtime are not in proper sync for the MDL versions... That may happen when using ToT

If you create a ticket we can take a look at that too

[vagrant moon]

that's interesting. i was using the latest github actions builds for dxvk-remix and bridge-remix btw, latest builds from yesterday

is there any way i could check MDL versions on my end?

[peak dagger]

yeah, that's probably it. We still have a separate branch with MDL work for the toolkit

[peak dagger]

If you open the toolkit's install dir (or the build dir if using the repo), and go to omni_core_materials\Base, you should see the MDL files

You can compare those to the ones in your captures

[vagrant moon]

hm

they're both mdl 1.4

[peak dagger]

Yeah no but the MDL contents

Also the bundled runtime (deps\remix_runtime\runtime) may be older

[vagrant moon]

can i just send it here? i don't know what to look for

[vagrant moon]

i can try to upgrade that

[peak dagger]

Just do a text diff of the 2 MDL files

if they're the same then there's no change

[vagrant moon]

ah i had the wrong mdl file open, oops

yeah they're identical

i'll try updating the runtime now

no luck 😦

[peak dagger]

Unsure then! You can create a ticket, we'll take a look

[vagrant moon]

might just be something buggy with this capture or game. it's another one of those games that renders everything through shaders

[vagrant moon]

thank ya for the help

[peak dagger]

You could always try with a "known to work" game like Portal RTX

[vagrant moon]

yeah i'm about to 🙂

[vagrant moon]

yup it's working fine in other games

would you still like me to open the issue? or does it seem more like a runtime issue as a whole

[peak dagger]

If it's game-specific it's probably a runtime issue. You could always open a ticket but it'll probably be low-prio, especially if the game is shader-heavy

[steep coyote]

For creating albedos, can we infer a direct and ambient light from the diffuse and kind of divide by that

Aka infer an image which looks like a whiteish surface with heights and normals, lit by the inferred lighting, and divide the original by this

[vagrant moon]

mm sadly that wouldn't really work from my understanding

it'd just end up having the effect of making it look washed out

which we could just do with normal filters anyway

@rugged wigeon what was that workflow you mentioned in photoshop to try to remove baked shadows?

[steep coyote]

Why's that

[rugged wigeon]

You duplicate the layer, apply a highpass filter and then set the blend mode to uhh.... Overlay?

[vagrant moon]

hmm

[vagrant moon]

just my experience with it while trying to bake textures that i generated with PBRify

maybe i'm just misunderstanding

[vagrant moon]

didn't quite work

i may have just done it wrong though

[acoustic cape]

I am thinking of trying this, does it offer AI metallic and roughness maps?

[vagrant moon]

PBRify does normal, roughness, and height

metallic is too hard to do, at least for the type of models i train

if you're looking for highest quality results you'll want PBRfusion instead. PBRify prioritizes speed

[acoustic cape]

One more question, does yours have batch conversion? Low-key tired of replacing textures one by one 💀

[vagrant moon]

give me like 10 mins and i can respond properly

it does though, just not directly

[steep coyote]

in theory this is the most mathematically perfect way to remove the lighting

i am doing a test in blender

[acoustic cape]

I have looked into delighting solutions, found one but it works in a 3dscan tool called agisoft delighter

https://www.agisoft.com/forum/index.php?topic=11086.0

[steep coyote]

i have a good feeling about this

[vagrant moon]

so PBRify has two methods of operating. one is through chaiNNEr, where it'll upscale and generate PBR for all of your dumped textures. it needs a separate script to replace the assets

the second way is through ComfyUI + the toolkit, where you jusy load up comfyui and the toolkit, select the mesh with the texture in the toolkit, then press Queue in comfyui and it'll upscale and replace

either way, it's replaced for you instead of having to do it all manually

[acoustic cape]

Gonna try the ComfyUI method

Sounds better

[vagrant moon]

it's a lot slower unfortunately

but it is more selective

[acoustic cape]

I really hope it isn't cpu intensive

[vagrant moon]

the only part that should be is the ingestion process in the toolkit

@sweet dagger you happen to know where that replacement script is?

#general-remix message

there it is. but the hashing method in Remix changed which means it no longer works

that sucks

i haven't found a way to update the hash generation yet. mostly because i can't find the code in the dxvk-repo for it

@acoustic cape for one of the comfyui steps you need to clone this repo instead of Nvidia's: https://github.com/Kim2091/ComfyUI-RTX-Remix

here are the steps: https://github.com/Kim2091/PBRify_Remix?tab=readme-ov-file#comfyui

[steep coyote]

it works extremely well

i put this metal texture on a lumpy surface with some light and rendered it in blender

then used this method ive been harping on about to remove the shadin

[sweet dagger]

#1231062867697209354 message

[vagrant moon]

could you show what you're doing in blender? an image would help a lot lol

[sweet dagger]

Found it! There’s also a little video tutorial a few messages below it, I don’t think you need to adjust the slashes in the address bar thingy anymore though 🙂

[vagrant moon]

yeah it's broken unfortunately

the hash method in Remix changed

[steep coyote]

this is the jpg version of the image i generated to remove the lighting (divide over the original texture to remove the lighting)
the actual one is an HDRI though

in theory you could generate millions of these automatically or procedurally for a training set

[vagrant moon]

we already have datasets for this task

we need a way to remove baked lighting without relying on AI models, because it just isn't working

[vagrant moon]

this method requires accurate normal and height maps to function though, right?

[steep coyote]

not necessarily

what i was thinking is you get a normal diffuse texture with baked in highlights and shadows

generate a normal and heightmap

use that to generate this division texture?

or maybe just got straight from diffuse to delit by training a model on a massive dataset generated with this method in blender

i guess you already have a dataset though

[vagrant moon]

that's what i was thinking too, but the generated normals and height just aren't accurate enough. at least not with my models

this might work for @limpid flame though

[vagrant moon]

yeah our dataset isn't too dissimilar. we gathered high quality CC0 textures, applied the PBR maps, then baked them with lighting. it's basically the same thing

nothing we train can handle it though

[vagrant moon]

actually i misread the script. no hashing needed. i'll test it

well

i advise you guys not to try to ingest with processor set to 32 in the toolkit

thankfully can confirm the script still works

[limpid flame]

yeh i was gonna say i didnt know we figured out hashing the first time lol

[vagrant moon]

blueamulet did, i integrated it into my MagicUSDA script before the toolkit was a thing. apparently a few months ago though the hashing method changed to be based on the application name or something

https://github.com/Kim2091/RTXRemixTools/blob/1fe9162ad0c4bbc05447f3bd45ab8afe362d6848/MagicUSDA/MagicUSDA.py#L9

[vagrant moon]

re-reading it all, it seems like what actually happened is that there's an even older hashing system that was being used in some games

so it should actually still work fine

[rugged wigeon]

Okay it had a few more steps I might've forgotten about at 4 AM 😅 https://tolas.wordpress.com/2009/05/26/tutorial-how-to-equalize-textures-in-photoshop/

[acoustic cape]

Saved in my bookmark, very useful 👀

[vagrant moon]

i tried it

the average blur thing, at least in affinity photo, just makes my texture totally gray for some reason

[rugged wigeon]

Oh it probably calculates the average color of the entire image

You have to use a blur the same size as the high pass.

[vagrant moon]

@steep coyote did you post an image? i thought i saw one

[steep coyote]

ya it was wrong though

This is how you can do this in blender, of course it's easy when you have the original displacement and lighting information.

if you use the result of the mix shader as a divide layer it cancels the lighting

i posted example in https://discord.com/channels/1028444667789967381/1196836285545971712

[rugged wigeon]

Reminds me of this method I saw a while ago https://youtu.be/w6YPbidAAzQ?si=FE6cxlLbMLwcv3Uj

[steep coyote]

ha mine's better

[rugged wigeon]

possibly ¯_(ツ)_/¯

[steep coyote]

objectively 😉

he is using a fullbright emissive surface for bounce lighting, im using the actual scene lit for bounce

more accurate

theres still room for improvement

[rugged wigeon]

I don't know much about kitchen appliances, I use C4D so anything in Blender is foreign to me.

[sweet dagger]

Oh it did? I swore I just used it the other day. When I get home I’ll send my local version, maybe chatgpt auto fixed it for me without me noticing lol

[vagrant moon]

i wsa wrong

[rugged wigeon]

PS & Affinity Photo 2

here's one with different colors using a gaussian blur instead of average

the average method works for surfaces with roughly equal color across the entire surface while gaussian blur will help with maintaining multiple colors but it requires a bit further tweaking with dodge and burn brushes to equalize dark spots

[vagrant moon]

pretty cool

thank ya

these are the kinds of textures i'm really struggling with

it really just needs a lot of dodge brush work i guess

[steep coyote]

a challenge!

[steep coyote]

some gimp

very little information in the dark areas to work with

[silent yarrow]

Getting this error when trying to upscale with the captures from the latest github runtime.

Can't get texture type using the USD attribute /RootNode/Looks/mat_F08851FC9EC2F174/Shader.inputs:diffuse_texture and texture type ROUGHNESS

[vagrant moon]

yeahhh that's a Remix issue most likely, not this tool

[peak dagger]

I don't see where remix would throw this error. Maybe it's in the ComfyUI node.

The texture type name in Remix itself for roughness is "Roughness" so if you set that anywhere you may just need to fix the caps, otherwise I don't know why you'd be gettin that error

[flint slate]

I have no idea if it's relevant, but this roughness stuff reminds of this crazy bug
https://github.com/NVIDIAGameWorks/rtx-remix/issues/703

I couldn't believe my own eyes when I saw "noughness" in the attached logs

[vagrant moon]

at what point does this happen?

[silent yarrow]

Any idea where would that be set? I didn't change anything like that at all that I can recall. I have also reinstalled,The toolkit,Comfyui, and PRBify several times now since this started happening. Even wiped all My kit settings at least twice.

[peak dagger]

There's a ComfyUI node to get the input path for newly created textures. Probably there

[silent yarrow]

hmm yeah it is all capitals there in the texture node.. I'll report back in a minute.

[silent yarrow]

yeah I'm at a loss for what I'm doing wrong here. I can't seem change the capitalization at all even. So I don't even know how this got messed up in the first place if I can't change it. I just reinstalled everything the toolkit,Comfyui,the depndances ,the emmedded thing... Python... and My web browser so Can't even fathom how this is carrying over.

[peak dagger]

Wonder what's happening and why other users aren't seeing this error

[silent yarrow]

I'm gonna say user error but I just don't know where lol. like I had this working fine.

here's the full log

bleh

uploaded the wrong log

[silent yarrow]

Well wiped My drives,reinstalled windows completely.everything else. fresh.not from back ups... same result.

[peak dagger]

You should create a GitHub issue. We have a few high prio issues atm so I may not be able to take a look tomorrow

[silent yarrow]

Well at this point I don't even think anyone else is affected besides Me I guess. I'll try to find something else to use rather then eat up dev time.

[peak dagger]

Up to you but it’s fine to create a ticket and if it’s deemed low prio we’ll just get to it later!

[peak dagger]

#1289913588852592662 message

[silent yarrow]

nice ha ha still pulling my hair out over this one. thanks. hope it's a easy fix.

[peak dagger]

#1289913588852592662 message

fix should be merged tonight!

[silent yarrow]

Oh nice,thanks!

[silent yarrow]

@peak dagger Tried the fix .Getting a different error now Unprocessable Entity for url stagecraft/textures. so progress I guess.

[peak dagger]

That means whatever object is sent in the request is invalid

So it’s not a remix issue per-se

[silent yarrow]

hmm. maybe the workflow needs to be updated. I tried this with this tool and PBRfusion3. That's outside My wheelhouse though so it's back to grabbing everything manually I guess.

[vagrant moon]

😦 i dunno what to change

it still works on my end

[silent yarrow]

well FWIW here's my error report. this is on a fresh install of ComfyUI and set up for remix, latest Remix, fresh captures,and clean install of the latest github toolkit.

[proper vale]

I'm getting the same issue, it definitely seems to be an outdated workflow, but I'm not sure at which node it happens specifically

[vagrant moon]

🤕

[vagrant moon]

the node should be highlighted with a red border

[proper vale]

It is in roughness section specifically

[vagrant moon]

mm

well that makes sense but i don't know what the issue is either, lol

😦

what triggers this error, using latest comfyui + latest toolkit?

[proper vale]

I'm actually not quite sure how to use toolkit, I have only used the nodes from the project above

[vagrant moon]

ahhh

you need the toolkit running to use this workflow

you click on what you want to upscale in the toolkit, then queue it in comfy

[proper vale]

Oh that's what you mean by toolkit. Yes, I do run RTX Remix app

[vagrant moon]

yeah

i REALLY wish the naming was better

[proper vale]

[vagrant moon]

sigh

well

so what build of the toolkit are you on?

[proper vale]

1.0.0

The fix is not released yet, I see now

Just merged

https://tenor.com/view/hachiko-dog-akita-waiting-gif-5319949

[vagrant moon]

actions builds seem broken too

but yeah that seems like the problem?

[proud robin]

Oh is the ingest error that pete mentioned still happening? #1289913588852592662 message

[vagrant moon]

yeah, it's not in any current build of the toolkit

the fix i mean

[proud robin]

Damn, at least that keeps me from getting everything set up and not letting it eat the rest of my day while upscaling so thats good lol

[proper vale]

This only seem to be happening for roughness tho, I have ran another workflow that just switches albedo texture and it works fine

[silent yarrow]

https://tenor.com/view/cry-pain-why-why-me-fake-laugh-gif-21556867

Yeah IDK, I've tried just about everything I can think of, save make a new workflow(tempting but ha ha no).
Hoping Pete can figure it out Monday.

Also sorry Kim for clogging up this thread. I know it's not You're fault! Luck of the draw. It was either this thread or Nightravens.

[vagrant moon]

lol it's okay. it is an issue directly related to it, so i'm glad you posted it

[limpid flame]

its affecting me too? 😦

[proud robin]

Uuuh not sure but its affecting both BlenderToRemix and PBRify

I assume it would

[silent yarrow]

yeah any tool that used the rest api I'd say. the blender tool is affected as well

[limpid flame]

thats aight only the 5090 can run pbrfusion and toolkit at same time

[silent yarrow]

lol yeah

For the most part unless it's a 8k texture I can still keep the toolkit open. 4 k still gets a bit laggy though

Primarily I use both Yours and Kims tool. PBRify is great for those small odd dimension ones PBRfusion doesn't like.

[proud robin]

Anyone still having issues with PBRify or other workflows after the recent commit?

Just built the toolkit and get this, not sure if its the same error as before because I never tried it

Seems to be failing here, so no fix yet or did I do something stupid? At least I can manually ingest them

Aaah but since it failed at the diffuse ingestion thats all I end up with in the assets/ingested folder, fuuuuuugggg

[peak dagger]

Haven’t taken a look at that yet but I’d be surprised if that endpoint was broken… I should get through everything by EoW

[proud robin]

No worries man, thanks again for looking into all this

[limpid flame]

try this i wasnt able to test it, it will only be able to do one at a time since ittl overwrite file names, output will be in ComfyUI/output

well no ittl be able to more since comfy appends numbers i think

[proud robin]

Can confirm it works and allows for multiple textures at once without overwriting, thank you sir 💚

Also this just shows how slow the ingestion process is, upscaling takes only a few seconds while ingestion makes the whole process take minutes per texture

[rugged wigeon]

Is ingestion set to process and 1 by any chance?

[vagrant moon]

the ingestion process is automatic with these workflows, and it is going one by one afaik 😦

i don't think that can be changed?

[rugged wigeon]

Oh this isn't via the toolkit?

[vagrant moon]

it uses the REST API. the toolkit sends textures to comfyui to be processed, which then sends them back to the toolkit to be ingested & replaced

[rugged wigeon]

The toolkit can spawn multiple processes to ingest in parallel

[vagrant moon]

i'm aware, but i'm not sure if the REST API allows that to be configured

[peak dagger]

Yeah, whatever config the toolkit uses to ingest you have full control of in the REST API. The node may not have all the options implemented but the API supports it

[peak dagger]

So everything works? Or is there something still broken ?

[limpid flame]

The edited adhoc non toolkit ingestion one I made for temporary uses was working they're referring to. Ingest nodes are still not working in comfyui remix nodes

[peak dagger]

I see! Thank you 🙂

[flint slate]

Hey guys, now I'm getting these too. Diffuse maps are the only ones that are not affected (they get upscaled and ingested just fine)
I kinda updated my whole environment in one go and it's broken now, so here's all the changes in case it's good to know:
GPU driver: 572.70 - 572.83
bridge-remix: 126th - 129th
dxvk-remix: 738th - 750th
Toolkit: 2024.5 (nvidia app) - 1.0.0 (nvidia app)

Curiously working diffuse

[peak dagger]

FYI I'm currently in this part of the code, "executor": 1 is how you would use the "external" processor (multiprocessing), and normally you should be able to call that endpoint many times and all the assets should be processed in parallel.

If your calls to the API are Async, which they should be, it should be as simple as sending many requests one after the other and gathering the results of the futures (which basically equates to launching a bunch of tasks in parallel and waiting for all of them to complete).

That said, you may need to update the max number of processors in the Toolkit's combobox for that to work as expected. I don't think that part can be controlled via the API

[peak dagger]

Got the fix with tests for basically all endpoints (except ingestion atm) so the API should be more stable now 🙂

Will be merged somewhat soon!

[peak dagger]

Gotta look at this one now

Hmm, seems to work on my side... Can you try to ingest using the API directly?
http://localhost:8011/docs#/Ingestion/add_item_to_queue_ingestcraft_mass_validator_queue_material_post
That's all you need to put in the body. You can replace everything with:

{
  "context_plugin": {
    "data": {
      "input_files": [["D:\\Tests\\Assets\\Textures\\invalid.png", "DIFFUSE"]],
      "output_directory": "D:\\Tests\\Assets\\AI Tools"
    }
  }
}

And of course change the file paths and texture type to whatever you want to ingest

@proud robin

[silent yarrow]

https://tenor.com/view/noted-gif-20569892

had I known I was the canary in the coal mine...

[proud robin]

That seems to work, took a second to figure out its double slashes lol but yeah that looks like it worked hmmm

Let me double check everything is up to date with my copy of comfyui and all that, the toolkit is "current"

[proud robin]

okay some progress, it completed the upscale in ComfyUI and assigned the diffuse texture but nothing else, the roughness and everything else is present in the ingested folder though

Gonna triple check that my toolkit is up to date real quick

[silent yarrow]

yeah that's the current behavior I think most of us are getting.

[proud robin]

And latest toolkit build crashes on launch uuuuuuhhhhh

The one I was running before was the first commit for the endpoint fixes

[peak dagger]

Yeah that’s what I fixed today. Waiting for it to get merged

[peak dagger]

You can submit the crash and send the logs but I’m guessing it’s the same issue other have been seeing with Glyphs

[proud robin]

Aaah okay makes sense now, for a moment I was getting confused wondering if I was still on a previously broken build. Thanks again for working on this and keeping us updated 💚

[peak dagger]

My pleasure 😁

#github-updates message

[peak dagger]

Should be good to retry using that commit

[proud robin]

Lets fuckin gooooo everything appears to be working now with the latest build, you're the fuckin man Pete, thanks for the speedy fix 🫡

[peak dagger]

Awesome! Great news!

[vagrant moon]

tyvm for all the hard work Pete

[silent yarrow]

just got home and installed. looks like everything is fine for Me as well! thanks Pete!

[flint slate]

Yes, it's completely fine now! Thanks a lot

[proper vale]

Is API being released separately from toolkit?

[peak dagger]

No the REST API is part of the Toolkit

[proper vale]

https://tenor.com/view/hachiko-dog-akita-waiting-gif-5319949

[vagrant moon]

@peak dagger this PR has been open for a while now, it includes a pretty important fix for PBRify (loading the correct model): https://github.com/NVIDIAGameWorks/ComfyUI-RTX-Remix/pull/8

is there any way that this could be merged soon? ty

[peak dagger]

Woop, I don’t think we’ve been monitoring that repo actively sorry! I’ll check it out today !

[peak dagger]

Approved! You should be able to merge whenever. Let me know if you can't

[peak dagger]

Actually looks like our process requires me to merge on our interal repo and the changes will be pushed to GitHub 🙂

[vagrant moon]

thank you

[steep coyote]

Hmm any ideas? fresh clean install / build of everything, comfyui, remix nodes, workflow, pbrify, toolkit

[steep coyote]

huh.. i try again and now I get this one:

[vagrant moon]

😦 not again

[steep coyote]

Kim's fault

😄

[vagrant moon]

naaaah i'm pretty sure this is on the toolkit side again 😛

[steep coyote]

did i do the right stuff? i git clone toolkit and built

[vagrant moon]

yeah

[steep coyote]

is there meant to be a .png on that filename maybe

[vagrant moon]

that included a fix for this issue before

i don't know what's happening tbh and i don't have much time to look into it 😦

[steep coyote]

https://tenor.com/view/ahoklollmao-noooo-nooo-noo-no-gif-17139315586294057944

no worries. ill have a look at the thread a bit more above maybe i messed up

[steep coyote]

2025-03-29T23:36:32.863375 - Requested URL: http://127.0.0.1:8011/ingestcraft/mass-validator/queue/material
Raw Response:

{'detail': 'The validation did not complete successfully. See the logs for '
'more information.'}

Anyone know where I can find these logs?

[peak dagger]

The regular toolkit logs

[steep coyote]

Cheers Pete

These ones right:
toolkit-remix_build\windows-x86_64\release\logs

I'll have a look through this

[peak dagger]

If you open the toolkit there’s a “show logs” button on the home page. The logs should be called kit_*

[steep coyote]

i think it might have been either my firewall or antivirus or port in use

[steep coyote]

very strange. every time I try to upscale, one of three things happens seemingly randomly
500 server error
Works normally
Can't get the ingested texture with name AutoUpScale_.... from the folder ....

[glacial monolith]

Sometimes I'll get those errors but if I try again on the same queue it works without issue, not sure what causes the intermittent error

Happens maybe 1 out of every 10 upscales

[crimson kayak]

all textures are darker than their original versions. What i should do to keep original colors?

[vagrant moon]

can you post an example of the issue

[crimson kayak]

Sorry I've didn't did it before.
I'l post ss's tomorrow

[crimson kayak]

the only thing i've changed in default chain is exported format from png to dds

[crimson kayak]

that's happening to every texture

[vagrant moon]

that might be causing the issue. you don't need to do that

are you using it through chaiNNer? or ComfyUI

[crimson kayak]

chainNNer

[vagrant moon]

okay. definitely don't use DDS

[crimson kayak]

PNGs should work with remixed Gmod?

[vagrant moon]

you need to export as PNG, then ingest with the toolkit. then it'll work

[crimson kayak]

the problem is, upscaled textures doesn't work in-game if i use this script w/png's. Only with .dds

[vagrant moon]

yeah

you need to ingest the PNGs in the toolkit first, then run that script on the output DDS files

[crimson kayak]

in convention, i should select source normals dx or octahedral?

[vagrant moon]

dx

[vagrant moon]

is it working now?

[crimson kayak]

i've faced a strange issue and im trying to figure out for myself why that's happened

Okay, nvm, it's works

but i need to test 1 thing

some certain textures if they were upscaled via pbrify - caused game to crash

but that was when i made them .dds in chain

so i need to test how they would work now

but atm everything works nice. Thanks

[steep coyote]

Hey @vagrant moon, it seems that PBRify reduces the alpha somewhat on all textures.
First image is original texture, second is after PBRify

ie. the original texture had no alpha transparency, but after pbrify it is reduced to 99 ish percent opaque

could cause some weird issues in remix rendering

here is the alpha channel with contrast boosted.
Before pbrify it was completely 100% white (totally opaque)

Same thing happens with PBRFusion @limpid flame

[flint slate]

The Great Upscaler conspiracy

they want to make my games less opaque😱

[steep coyote]

BUSTED

now y'all have to make me a script that fixes this on a few thousand textures 🤣

(not really)

[limpid flame]

We both split the alpha and upscale it with her Span model, which I never thought about till you brought it up,

On images with 100% opaque, ig any upscaler would upscale into noise even most of the algorithmic ones

Which means we both need to add logic to detect if the alpha is opaque then to upscale it with point sampling or nearest neighbor instead of span. Or make a full white alpha at new res

[steep coyote]

something like if original image has no transparent pixels just drop alpha channel and export as BC1 maybe

half the file size, only 4bpp though

might be imperceptible?

[limpid flame]

Probs keep at PNG since remix import will compress again to dds

Don't wanna compress more then once

[steep coyote]

true

[steep coyote]

it may not actually matter at all, i tried using a 75% alpha channel on an albedo texture and it looked the same. as soon as I went below 50% the model with the texture became completely invisible

maybe remix is just using a binary visibility threshold or something

[rugged wigeon]

the default mode is alpha test which would explain that behavior, if you toggle the blend option then it would act like a translucent object

[steep coyote]

so crisis averted! (probably)

not sure if there could be some other weird ramifications of 99% alpha on all albedo textures

[steep coyote]

ok well this is relevant:
"BC7 varies on a per block basis as needed, allowing it to use the unused alpha bits for additional color definition"
https://www.nexusmods.com/fallout4/mods/25799?tab=posts
so having the alpha in there unneccessarily is probably reducing quality somewhat

[vagrant moon]

unfortunately it's not really possible to strip alpha in specific scenarios without greatly harming performance in the chaiNNer workflow, and i don't know enough about ComfyUI to attempt it there. at the very least you'd need additional node packs for that

[rugged wigeon]

wasn't there a split channel node?

[vagrant moon]

there is, i already utilize that. but from my understanding binq is suggesting removing the alpha layer under specific conditions. the math for that is extremely costly, i know because i tried it before for something else

it took 5x longer per image

[rugged wigeon]

huh interesting

should be as simple as taking the original image and doing a if texture.a < 1 then process, else discard

[vagrant moon]

i'll see if i can do that

that specific setup shoooould maybe work in a way that would only be like a 15% additional cost

the problem is i have to split the alpha layer in a specific way ahead of time, so it's doing a lot of calculations twice

[rugged wigeon]

could always do a branch for people that want to use it

[vagrant moon]

very doable

[steep coyote]

i wouldnt stress about it for my sake at least

[flint slate]

never say that

say "yes do it yes do it I need it yes I do"

[wispy sparrow]

fixify

[analog cairn]

when I try to Run comfyui for RTX Remix I get this error:

any ideas?

[vagrant moon]

it looks like you downloaded an old version of the pbrify models

or didn't download them at all

https://github.com/Kim2091/PBRify_Remix/releases/tag/v1.7.1_ComfyOnly

you need these

follow these instructions: https://github.com/Kim2091/PBRify_Remix?tab=readme-ov-file#comfyui

[analog cairn]

ah, let me try this. thanks

[analog cairn]

I downloaded the new release and placed it where it suppose to go.

I select the objects with texture than open ComfyUI and open the workflow .png in there but after that nothing really happens. If I click run in ComfyUI I now get only two errors:

[vagrant moon]

right, so that's the other common error. you need a new build of the toolkit from github i think

https://drive.google.com/drive/folders/1iaSz_R-OhMpbZfvR2r9Nc2p99UsjUXKi

this is a prebuilt copy of it someone made a few days ago

download that and extract, and it should work

[analog cairn]

ok thanks, let me try that right now

still same rror

could be it be due to rtx 5080?

do I need to install creator drivers?

I also see this error:

[vagrant moon]

@peak dagger any idea what's going on here?

it's happening even on the toolkit build from a few days ago

[peak dagger]

Thos errors are not related and also not a big deal

[peak dagger]

would need more info than that

[analog cairn]

I am trying to execute the workflow through compfyui on a texture in rtx remix and keep getting these errors when I click run on ComfyUI:

[peak dagger]

yeah I mean it's trying to get the layers in your project and doesn't find any. Not too sure what that's about. The endpoint works find on my end.

The only thing I can think of is if no project is opened but otherwise not too sure

You can always try to debug by running the function directly:
http://localhost:8011/docs#/Modding/get_layers

[peak dagger]

Actually looks like the layer_count argument of that endpoint is not a number ?

[analog cairn]

I am new to rtx remix, not sure where to find that?

[peak dagger]

that's in the ComfyUI node code, it's not in the Toolkit UI

@vagrant moon may be able to assist

[vagrant moon]

tbh i'm not really sure what to do here

downloading the latest toolkit and testing on my end though

[peak dagger]

After the latest toolkit is running, just go here and execute the endpoint. You should see it work fine. From there you can set layer_count to -1, 1, 2, etc.

If all that works then the ComfyUI node is the culprit. Idk if there was a ComfyUI update recently that may have broken the node

[vagrant moon]

was able to reproduce it

looking at that now

[peak dagger]

Using the endpoint directly ?

[vagrant moon]

latest toolkit + comfy has the issue

[peak dagger]

yeah I was more asking about the toolkit specifically

[vagrant moon]

trying to use the link you sent to debug, but i'm not very familiar with these types of things unfortunately

[peak dagger]

Click Try and Execute

[peak dagger]

yeah worked fine

now where there's a -1 you can change that value

[peak dagger]

you can also set layer_type to "autoupscale" since it seems to be what comfyUI uses

[analog cairn]

it works as long as it is number

[analog cairn]

it is already set to that in the workflow:

[peak dagger]

yeah worked fine too

so the issue is with ComfyUI cause you tested the entire endpoint

[peak dagger]

yeah I meant in the docs website

[analog cairn]

should I go to an older built of the rtx remix

if that is less of a trouble 😄

[peak dagger]

my guess is ComfyUI broke something

because Remix works as intended, as you just proved haha

[analog cairn]

well finger-crossed. Anyway to get older build of comfyui that was working with rtx remix

[vagrant moon]

can confirm my older version of comfy works fine

[peak dagger]

So someone will have to debug the ComfyUI nodes and push a fix MR 😛

[analog cairn]

Anyway you can share the installer for older version

[vagrant moon]

the toolkit keeps crashing though

hmm

froze my entire PC too

[vagrant moon]

looking for it

https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.2.2

this is the version i have

[analog cairn]

Thanks

[vagrant moon]

@limpid flame heads up for you. latest versions of comfy seem to be causing issues with the remix comfy extension

[analog cairn]

I am using this, I uploaded the upscale models and copyed the files from old custom_nodes over to this portable version. I guess no luck or it just doesn't want to work for me:

I loaded older upscaler models v1.7.0 and that seem to have gotten rid of the error but here is what I see when I Queue Prompt:

Nothing really changes on the RTX side related to textures but there is a new layer now under mod.usda called aiupscale_textures.usda

New error, lol:

[vagrant moon]

yeah. you need to download and add the rplksrd model

[vagrant moon]

this can happen if you have shadowplay on, recording the desktop or something else

[analog cairn]

thanks

so i will go back to 1.7.1 and add this one you provided

I need to reboot

[analog cairn]

I literally just have comfyui and rtx remix open :P. I will play around with it a little more to see if I can figure this out

i think i need to use my rtx toolkit build

the one you linked me might not have support yet for 5080

nope still same error

[vagrant moon]

😦

[analog cairn]

i think it has to do with portable build of comfyui

[limpid flame]

I haven't kept up with 50series support on comfy but for a time you had to build it with a specific pytorch version.

You can try the prebuilt old comfy version with 50 series environment from pbrfusion I built awhile ago

https://huggingface.co/NightRaven109/PBRFusion/blob/main/PBRFusion3.0.2_portable_50seriesonly.zip

Rtx remix node is already installed

[analog cairn]

thanks, let me give this a try

Mission Impossible: RTX Remix

😄

[limpid flame]

If I find time this weekend ill look into the new version problem

[analog cairn]

that'd be awesome

[analog cairn]

so I it generating textures now in comfyui but I get this:

[vagrant moon]

man. why are there so many issues 😦

idek how to go about fixing that one

i can't reproduce it on my end

[analog cairn]

i guess time to give up and pick up in couple week in hopes that some how magic happens on my machines 😄

[vagrant moon]

😦 sorry

[analog cairn]

its all good

it was worth a try

[limpid flame]

I think if you update rtx remix node it will work now

On top right you can hit manager, custom node search and click update on it

[analog cairn]

ok let me try

same error

i am just out of luck 😄

[limpid flame]

well damn

[vagrant moon]

https://cdn.discordapp.com/emojis/757304232717058088.webp?size=512&animated=true

[analog cairn]

I am still trying to figure out how to first make it all work 😛

[vagrant moon]

PBRify 1.7.2 Update

Download
ComfyUI Only Version

This update brings another new upscaling model, 4x-PBRify_UpscalerV4. This model does a much better job at refining existing detail, and more accurately removes artifacts.

Download use it in chaiNNer, download this release: https://github.com/chaiNNer-org/chaiNNer-nightly/releases
To use it in ComfyUI, follow these steps: https://github.com/Kim2091/PBRify_Remix?tab=readme-ov-file#comfyui

🚀 If you like my work, please support me on Ko-fi! 🚀

[wise iris]

just tried it, new upscaler looks great so far

[peak dagger]

FYI the comfyUI node will be updated soon as they will not work on the latest latest commit of the Toolkit.
See #1289913588852592662 message
I'll take care of updating the nodes but you'll need to update them in your provided workflows

[vagrant moon]

thanks for the heads up 🙂

[peak dagger]

ComfyUI Nodes (& Toolkit) have been updated to work with the latest version of ComfyUI & Toolkit
https://github.com/NVIDIAGameWorks/ComfyUI-RTX-Remix/commit/711583119ff6a788ceca8b77689837c4d783f712

[vagrant moon]

it'll be a while before i'm able to update the workflow

[vagrant moon]

well, i regret to inform you all of this, but there's no time for me to maintain PBRify

if someone else is willing & able to update the comfyui workflow, i'd greatly appreciate it. i'm dealing with too much right now to be able to dedicate time to this

[peak dagger]

FYI the example workflow in the remix comfyUI node repo uses pbrify so it should be usable as-is

[vagrant moon]

oh, i didn't realize it had been updated

thank you 💚

[peak dagger]

Pleasure 🙂

[neat crypt]

it keeps throwing an error on this node

what could be the issue?

[weary cape]

i tried giving it a go after swapping the diffuse upscaler (likely not relevant)
and i ended up getting these errors in the get layers part of the workflow

[amber badger]

This is super useful! Thank you

[peak dagger]

You can try getting the latest workflow from the GitHub repository (update the nodes)

[weary cape]

thanks, i hadn't noticed it was different :D

[young plaza]

@peak dagger

Hey, i am using comfyui with the rtx remix template workflow/rest api, using pbrify. and i keep getting this error:

422 Client Error: Unprocessable Entity for url: http://127.0.0.1:8011/stagecraft/layers/target/G%3A%2FRTX+Doom+3%2FDOOM+3+RTX+MOD%2Faiupscale_textures.usda

, it created the .usda in my poject, i am not sure what to do here.

[peak dagger]

Are you able to execute this correctly outside ComfyUI?
http://localhost:8011/docs#/Modding/set_edit_target_layer

You can use the GetLayers endpoint to get the list of LayerIDs easily

You can test the set edit target this way

And then in the toolkit you should see the layer you set as the edit target in the layers panel

If you have issues during any of these steps it should give you a better idea of where the issue comes from.

[young plaza]

cheers for helping.

yeah that all worked fine, it changed the edit layer target to the new .usda

hmmm

[peak dagger]

Seems like it's a ComfyUI flow issue then. Maybe @limpid flame would know more since he's been working on improving it 😛

[young plaza]

Ok thanks..

First time I used.

I installed comfyui.

Git cloned comfyui remix. And installed.

Grabbed the pbrify models.

Didn't seem like anything else needed doing

[peak dagger]

Should still be like that just with new updates 😛 Maybe something broke in the mean time on the ComfyUI side though. They've been updating lots of things

[young plaza]

Yeah i noticed the model name they used for the upscale node didnt match any from the pbrify model folder. So I changed that, that solved earlier issue, but then hit this. Does sound like needs update.

@limpid flame do you have any idea about this error?
using comfyui with the rtx remix template workflow/rest api, using pbrify. and i keep getting this error:

422 Client Error: Unprocessable Entity for url: http://127.0.0.1:8011/stagecraft/layers/target/G%3A%2FRTX+Doom+3%2FDOOM+3+RTX+MOD%2Faiupscale_textures.usda

[limpid flame]

hmm not seeing this with latest comfyui and latest remix comfy node. can you please send whole comfyui console readout?

[young plaza]

hi, yeah sure:

[limpid flame]

@peak dagger seems odd shouldnt it make layer

[young plaza]

The layer was made, it seems to not recognise it

And rest api by itself allows to select layer

I've never even used comfyui before today, so I haven't been messing with stuff either

[limpid flame]

can you send the usda file

[young plaza]

btw i am am aware chatgpt could be spewing bs here, but jsut in case, this is what it thought:

➡ Explanation:

The RTX Remix REST API is saying: “The layer does not exist”.

The file path appears incorrectly parsed:

vbnet
Copy
Edit
G:\RTX+Doom+3\DOOM+3+RTX+MOD\aiupscale_textures.usda
Note that the path separator + is being used instead of a space — this happens when:

Improper URL encoding is done

Or the system is double-encoding or not decoding %20 back into spaces.

🎯 Root Cause Breakdown
🔹 1. Improper Encoding
The file path was originally:

vbnet
Copy
Edit
G:/RTX Doom 3/DOOM 3 RTX MOD/aiupscale_textures.usda
But the REST call tried:

perl
Copy
Edit
G%3A%2FRTX+Doom+3%2FDOOM+3+RTX+MOD%2Faiupscale_textures.usda
%3A = :

%2F = /

• = improperly used for space

✅ Correct encoding should use %20 for spaces, not +, especially in URL path components.

[limpid flame]

no its getting and parsing the request fine

[peak dagger]

I wonder if you're the first that tried to use the ComfyUI node with a path that has spaces

[limpid flame]

mmm

[young plaza]

ha

well maybe?

i dunno, anyway that's really all the info i can give

[peak dagger]

you can try using a path with no spaces and see if you get the error

[young plaza]

so just to be clear. if i had RTX DOOM 3, i should try RTXDOOM3?

[peak dagger]

a quick and easy way to try that without breaking all your paths would be to create a symlink

and yeah correct

[young plaza]

oh ok

[peak dagger]

you can run something like mklink /J "G:/TEST/DOOM_3_RTX_MOD" "G:/RTX Doom 3/DOOM 3 RTX MOD"

it'll create a "folder" "G:/TEST/DOOM_3_RTX_MOD" but it'll just be a symlink to your existing project

then you can open "G:/TEST/DOOM_3_RTX_MOD/PROJECT.USDA" in the toolkit and try in comfyUI

[young plaza]

well it seem like a symlink day, as i have been creating a bunch today to free up space on my c drive lol

ok lets try that

it worked

@limpid flame

@peak dagger thanks

[limpid flame]

🙂

[peak dagger]

So we have an issue with spaces in paths in the ComfyUI node

good news but we gotta fix that haha

[limpid flame]

Pete the toolkit allows for spaces now right?

[peak dagger]

yeah!

[young plaza]

well it felt good to help

[limpid flame]

ok lets try and fix it in this new pull then lol

[young plaza]

how on earth am i the only one using spaces?

[peak dagger]

I think ChatGPT is right and the issue is with how spaces are encoded

[young plaza]

wellatleastwegotitsortedrightguys

[limpid flame]

ChadGPT

[peak dagger]

Good question lol

[young plaza]

ok well thansk for your help both of you, i'm excited to use comfyui. I couldn't keep doing my workflow, was killing me

@peak dagger you reckon i am ok to keep using symlink or shouldi change my path name?

[peak dagger]

Up to you! Whatever you prefer. There should be no problem using Symlinks but if you just want 1 instance of the dir then you can rename your folders

[young plaza]

oh my i feel the power

[limpid flame]

@young plaza can you try replacing layers.py with this file in ComfyUI\custom_nodes\ComfyUI-RTX-Remix\nodes

and trying again on the project with spaces? , make sure to restart comfy

[young plaza]

yes, but gimme a few mins

[limpid flame]

all good, preciate' it

[young plaza]

ignore last message, i didnt restart comfyui beforehand

@limpid flame Well done, it's fixed

[limpid flame]

thanks for testing!

P simple fix Pete, had to use quote instead of quote_plus

[young plaza]

i am going to use this so much, so thank so much for making it

[limpid flame]

Well its an effort on everyones part! enjoy!

[peak dagger]

probably wanna make sure we also mark no characters as safe but yeah that's an acceptable solution 😛 Do you wanna add it to your PR ?

[limpid flame]

yes

just adding in safe param here is correct?

[peak dagger]

Awesome! Thank you !!

[peak dagger]

yeah! otherwise "/" is safe by default and that'll break the URL parsing

[distant lance]

Im confused when it comes to the get layer parts. How do I set this?

[distant lance]

@peak dagger anyway you can help me out with the getlayers functions in the workflow? Im not sure how to specify a replacement layer and an aiupscale layer.

[flint slate]

You don't need to make any changes to the comfyui workflow (you can skip specific parts like Height map generation if needed)

You just select materials in the Toolkit viewport and run the workflow

It will upscale, generate PBRs, ingest, everything

[distant lance]

If I wanted to change it to where instead of upscaling the texture, I have an ai models completely generate a different one, how could I do that?

[peak dagger]

You just give a layer type, which you can get from the “get layer type” node

[peak dagger]

Depends on the implementation of the model. You’ll have to look at ComfyUI documentation and simply replace the upscale node with your own model inference

[distant lance]

@vagrant moon is there a metallic map pth anywhere?

[vagrant moon]

no, it wasn't possible to train a model for that

[fickle pebble]

decided to try out the low vram workflow and I'm getting this error on the 4th node
I am using the toolkit build from 7 days ago

[crimson kayak]

yup, i've got the same issue

and it's only related to the low vram one

[peak dagger]

layer_types=1 is not valid, it would need to be layer_types=AUTOUPSCALE. Not sure if that's a workflow issue or node issue.

FYI @limpid flame

[limpid flame]

Using Latest Toolkit, Latest Comfy, Latest nodes, i dont see this problem

[flint phoenix]

how do you use this

[limpid flame]

Context ^ this was from using portable version of PBRFusion to run this, but nodes have been updated since that

Open a new workflow with that + button

[flint phoenix]

alr

[limpid flame]

click in the empty space and search for upscale

I'm not at my computer rn so if there's a lot of options u can sc and ill tell u which

[flint phoenix]

?

[limpid flame]

Uh try double clicking

[limpid flame]

Lets do load upscale model top one

And then select Kim's upscale model in list for node

[limpid flame]

Yeh! Now choose upscaler

[flint phoenix]

safetensors or span

[limpid flame]

Safe tensor is full fat latest upscaler, span is faster but lower qual

[flint phoenix]

safe tensor it is

now what?

[limpid flame]

Ok now click and drag out the bubble on the right next to UPSCALE_MODEL

And release it in space

It should come up with options

[flint phoenix]

just like that!

[limpid flame]

Yeh choose ImageUpscaleWithModel

[flint phoenix]

ImageUpscaleWithModel!

[limpid flame]

Now do same to left blue image bubble and select load image

[limpid flame]

Nice now same with right image bubble and select save image

[limpid flame]

Nice now you can drag in images into left load image node, and hit queue to run

Note when you drag in a DDS the image name will change but it won't preview it like it is that green particle board and particle board img preview may stay

[flint phoenix]

lets see

[limpid flame]

Then it will be in ComfyUI/output or you can right click the done img and hit save

[flint phoenix]

and like magic!

ta- da!

[limpid flame]

Ok now to save, hit file on top right and click export

To Save the workflow

All of pbrifys models go through that upscale node, so if ya want you can make the output of the upscale into 3 more of those nodes to do the pbr

And that's essentially what ComfyUI is, each node is a Python script, with input and outputs, with it you can easily combine alot of cool things to make workflows

[sweet dagger]

PBRify my sweet - its so incredibly fast with good quality

[cinder bloom]

?

[sullen hatch]

Hey guys looking for some assistance getting RESTAPI_PBRify working with Remix and cant seem to get it to link up. am i doing something wrong?

[silent yarrow]

@sullen hatch are You making the workflow Yourself? or Is this with the workflow provided on Kim's Github? If it's the latter, afaik The toolkit changed a lot of stuff many months ago that broke that workflow.

[vagrant moon]

🤔

https://github.com/NVIDIAGameWorks/ComfyUI-RTX-Remix/tree/main/workflows

the workflow here should work

[silent yarrow]

https://tenor.com/view/idk-shrug-tarryn-gif-22143231

[sullen hatch]

ah thats prob why its not working

[sullen hatch]

how do i replace my existing Workflows?

[vagrant moon]

i just updated the readme to point to the correct workflow

[vagrant moon]

😓 going to be honest. i don't use comfy at all and haven't in over a year now

[silent yarrow]

Yeah I don't know either. I'd like to know though... tried getting AI to make Me a PBRfusion4 one the other day with no success.

[sullen hatch]

so Chainer or what ever is better?

im new to this so im trying to learn alot at once