You’ve indeed been crazy busy and I think we all greatly appreciate your efforts.
The UV wrap basically acts like the stamp node. I was hoping for something that takes the plate and a UV and the result is like that of your unwrapped UV result in the face tracker. Does that make sense?
I’m on set right now but Chris Noellert posted a batch that does uv unwrap/rewrap here a few years ago that should be findable with the search tool. I’ve done several shots with it. It’s great!
The OFX has been built so you can do it all in one node.
Oh yeah, I remember that. I seem to recall that there were limitations though, no? Either way, it’s quite nice to have it in one node so thank you again @space_monkey.
The revisions to FaceUV are great btw. The ability to adjust and nudge the mesh is ace.
On my github I just created an updater and installer for the node so you can go there pull that it will also check what the latest versions are , caveat first run it doesnt read the older versions so just update everything then from then on it will work or hopefully, I cant try it on my end.
Hey @space_monkey I’ve been automating Matte LTX Native to run shot-by-shot through a whole sequence (analyze → wait for done → next shot), watching the engine from outside Flame via its socket. The monitoring side all works great — progress, status, finished_jobs.json, re-queue. One thing is missing, though, and I wanted to ask if you might consider helping with it:
Would it be possible to trigger “Generate” somehow without the UI?
Right now only the Generate button in the OFX panel creates the job folder (input.f16 + job.json), and Flame’s Python API can’t set OFX params, so my script can’t start an analysis on its own. If you ever get a chance, two ideas that could work (whichever is easier for you, if either is feasible at all):
It will be in the next release I have to update the overall engine so I will doit when I get a few other things fixed as well
the nodes have been update to include that feature. Let me know how it goes
I will say I am disappointed that I don’t have more votes that I asked you all to do, a total of 8 people have shown any effort to help out , if I don’t see that number go up I will just pull them off , We and I need this and if you could literaly take 10 seconds that would be appreciated of the 100s + people downloading these OFX’s
Just voted sorry it’s been a busy stint for me, thank you for all of your contributions
So curious about cloud rendering and what people would want me to add, so far Ive done Runpod which is not as secure as AWS or Azure but cheap so if security isn’t an issue and you just want fast renders this seems like a decent choice. I have it cue up and delete once the render is done so each 200 frame render is a few cents. Azure I was getting implemented but they seem to not have any nodes available in most of the areas I am looking. I will also add LAN rendering once I get through this. I will push the update how it stands now and more features are coming
Runpod seems pretty attractive, first I’m hearing of them. Out of curiosity, why is it not as secure as AWS?
On the technical basics, they’re closer than the label suggests. Both encrypt traffic, isolate your workload, and hold SOC 2 Type II, and Runpod also holds ISO 27001. The real differences are about who controls the hardware and what’s been independently audited for film content. AWS has been assessed under the MPA’s Trusted Partner Network, the security programme studios actually check vendors against for pre-release material. Runpod hasn’t. If a show’s security addendum says “TPN-assessed infrastructure only”, AWS satisfies it on paper and Runpod doesn’t.
That being said Runpod you pre load a certain amount of money to it and each render is cents while AWS is like $3 so its a real cost difference. So far I’ve been testing run pod for the last 2 weeks and its pretty nice , now I have a script that shuts the whole thing down in one script and spools it back up in like 45 min so I dont get billed when I am not using it so thats been pretty great. I was going to get AWS or Azure on as well but its hard to get some nodes up since they are booked probably with the hyperscalers.
Thanks for the explination.
So you’re having to spin up the machine (presumably w/ the models already cached) and then send work to it as opposed to making an end-point API and just sending tasks to that. Am I understanding that correctly? 45 min to spin up an instance seems super long though, no?
That was going to be one of my questions. Is it a scenario where each user has to sign-up, clone a pod you’ve pre-built and keep those gated models cached, and then we send tasks to it all the while having to manage the uptime OR is it one of the serverless workflows where the user just provides their API key & they’re billed accordingly.
If it’s route 1, I assume they’re charging for the persistent storage of the models & container(s)…?
yes you spin up the drives copy the models directly from hugging face through a key, then render. I was copying all the models needed so something like 450 GB , ideally i could just select the model per node and that would save a lot of copying time but for testing I just did them all which isn’t totally necessary. Then when I am done with using them I spin them down with a script so I am not getting charged for storage. They all work basically the same or similar way. They need to have the models on a drive to render and the GPU are a separate cost.
I’ve got my ‘ai’ workstation (windows with big bad rtx pro running mostly comfy) next to my Mac flame.. would it be technically possible to leverage that one for rendering? In stead of a cloud solution? Thanks!
yes LAN rendering is next up on the list of things to do
Thank you all for all the hardwork on this. The tools are really great and already extremely useful. Is there an outpainting tool in the works? again thank you
LAN rendering is up, I cant test it on windows since I dont have one, but I got it working on other workstations. So if you can set up sharing so that the windows can see the drive with the models
