I have the same problem

Is that the whole message? Sounds like that the face is too zoomed in or being cropped.

What’s the question?

@@clayrainco When i follow the tutorial and I move the leg, it does not follow properly. Its a mechanical leg and the skin on top is organic but it does not move correctly.

Glad you like. What are you creating?

Is that Donald Trumps face on Shakira's body? Dude you're disgusting!!

Yes, I code.

@@creatorbrew ahh okay.... So ummmm ... So I'm building an app right now... in fact I'm a bit of a ways in to MVP at the moment. So you know the way Comfy uses JSON to actually create the workflows....? well this is the method I'm planning on using to create the heavy AI lifting on my app.... basically a CrewAI or Autogen agent setup calls a GPU instance on say runpod or another vendor, creates the workflows based on a brief, ( if they are not already in a library)... sets up the flow, prompts the models and changes the settings, reviews the output, stiches it all together... downloads it and then stops the instance... (cos video man needs an A100 and those are not cheap online) Most of the actual face to face with users is done via HeyGen's streaming avatar service (endpoint) but I'm also looking to eventually replace that... but you see users on this app get to download a full music video of their performance or video output of their comedy routine.... Right now for example I'm cutting up video snippets to make controlnet splines etc... If I put a link in here then this comment will vanish into cyberspace... see bandy buddy to have a look at the general idea

Hey, you can go with a photogrammetry route or try Trellis as a solution. Click the "multiple image tab" huggingface.co/spaces/JeffreyXiang/TRELLIS

@@creatorbrew thanks

Cool! spread the word! :)

@creatorbrew seriously I couldn't find any working tutorial except for this one for AE 2024

I think the best path would be to train a Lora (on 22 images or so) then do this turn around. Follow it up by using these images as the basis to create a 3D model. Whether that is using photogrammetry approach, AI, or by hand. I've seen recent posting of a new "one image to 3D turnaround" that looks great when it is released. I haven't experienced a good image-to-model that works with cartoon or hand drawn art, yet! Yet because maybe something new will drop any minute from now. I see the trouble I have with the artwork I want to create, I want to hand draw art and then convert it to 3D good enough that I can then clean up the model. So far, I can kind of take my own art, have it rotated into a turn around, or a low res 3D model and then use that as a template to build from in 3D.

Like this is an old tutorial and with a little effort by you, you will learn comfy and discover that you can update the video nodes. Comfyui changes week by week, there are newer methods to achieve this effect, at the time, months ago this was a great solution. Still is a good solution if you are willing to do a little learning.

(the shocked face with hands on cheeks… lol)

great to read! good luck on your project

@@creatorbrew Actually i'm having an issue, when my character walk, there is no gravity so he's walking in the air, is there any fix to that ? (rb and scripts don't work, it is override by timeline).

I don't know how much you use scripting. Basically you can create a script that uses a raycast to have an object follow the normals of the surface that is below it, this way an object can glide along a hilly surface. If you aren't into coding, you can use ChatGPT to create this script. Tested it out with a box and hilly surface mesh, and more or less GPT will give a similar answer as long as you ask the right questions and let it know you aren't using rigid bodies.. I've done this a few times during demos with ChatGPT. Make sure to have a collider on both the object being moved and the surface. For a character, make the character the child of the box. If your characters motion is driving the character forward, vs motions in place, then use the box as a "follower" of the character, then have the "rotation to match the normals of the surface" applied to the character's root or a parent of the root location.

​@@creatorbrew I've tried this, but the problem is that timeline really override on everything, it kind of disable every component on character. The only script that I could make that would resist to this override ended up messing with timeline animation and making everything glitchy. I'm a pro developer and I never wanted to do too much animation/3d stuff etc, but I may dig into it know as it seems to be the only way (editing animations in blender or something)

Without having to buy Elements3D :) free and no installs

An easy one, everyone forgets especially when creating late at night. Happy modeling.

You can try small segments when an object crosses vs longer. Otherwise the goal would be to create a clip with the face clear using and hacky technique then run a face swap and add back the missing object on another layer. Advances happen monthly and there could always be a better solution, keep looking and researching.

What GPU you're using, I'm using RTX4070 Ti Super, but still occured: Warning: Ran out of memory when regular VAE encoding, retrying with tiled VAE encoding. Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.

Check face index - the template default is set to 1, but if there is only 1 face it should be 0