akatz (@akatz

akatz

208 posts

akatz

@akatz_ai

AI Art Appreciator, Vibe Lifer and Claude Abuser. Building ComfyGit (among many other things).

Joined March 2012

akatz
@akatz_ai
Oct 18, 2024
I ported DepthCrafter to ComfyUI! 🔥 Now you can generate super stable depthmap videos from any input video. The VRAM requirement is pretty high (>16GB) if you want to render long videos in high res (768p and up) It pairs well with Depthflow! Repo link in comments below👇
00:00
1.9M
akatz
@akatz_ai
Nov 14, 2024
I ported X-Portrait to ComfyUI! 🔥 Generate performances for any character. It performs well with certain styles and input videos but comes up short in a lot of ways (X-Portrait 2 soon? 👀) Currently requires (>20GB VRAM) to run, I've been testing on a 4090. Repo link 👇
00:00
46K
akatz
@akatz_ai
Nov 27, 2024
I just updated my DepthCrafter ComfyUI nodes to support v1.0.1 🚀 The original config running a 200 frame video at 896p with 25 steps takes around 433s. The new v1.0.1 config with the same resolution at 5 steps takes 93s. (4x speedup) You can see the differences below:
00:00
15K
akatz
@akatz_ai
Oct 18, 2024
Replying to @akatz_ai
Original dancer: sterlingtorress (tiktok) Original DepthCrafter repo (not mine): github.com/Tencent/DepthC… DepthCrafter Custom Nodes: github.com/akatz-ai/Comfy… (Example workflow available in the repo)
GitHub - Tencent/DepthCrafter: [CVPR 2025 Highlight] DepthCrafter: Generating Consistent Long Depth...
From github.com
41K
akatz
@akatz_ai
Oct 19, 2024
Replying to @EditzbyMalik
Good question! I have not tried any large landscape footage or drone shots yet. If you have a chance to try out the node pack I would love to see what kind of results you’re able to get!
29K
akatz
@akatz_ai
Aug 17, 2024
A short supercut of videos I made recently superimposed with the original dancers. Some don't realize that these videos are "transformations" of existing dance videos, as opposed to original generations. I wanted to give the dancers used in my videos the credit they deserve!
00:00
2.3K
akatz
@akatz_ai
Oct 20, 2024
Replying to @shmkane
Better temporal consistency, but lower output resolution and higher VRAM requirements. Also not realtime capable, the model needs the full video to be loaded. It’s a good tradeoff if you need the stability, and you can upscale the resulting depthmap as well
6.4K
akatz
@akatz_ai
Oct 20, 2024
WOW I did not expect to get 1.6M+ views on my first DepthCrafter post! 🤯 Thanks everyone! I put together a few more sample videos so you can see how the model performs under different circumstances. Here's footage from Minecraft running through the model:
00:00
4.3K
akatz
@akatz_ai
Nov 14, 2024
Replying to @akatz_ai
Original X-Portrait Repo (not mine): github.com/bytedance/X-Po… X-Portrait Custom Nodes: github.com/akatz-ai/Comfy… The repo uses an Apache-2.0 License and the model uses a CreativeML Open RAIL++-M License.
GitHub - bytedance/X-Portrait: Source code for the SIGGRAPH 2024 paper "X-Portrait: Expressive...
From github.com
1.8K
akatz
@akatz_ai
Oct 18, 2024
Replying to @CFSStudiosAI
I would say that for temporal consistency nothing beats DepthCrafter (that I'm aware of). However due to the VRAM limitations I've only been able to render videos up to ~896x896 in resolution with a 4090. I still love depthanything v2 for high-resolution images and some vid2vid.
21K
akatz
@akatz_ai
Nov 26, 2024
Looks like DepthCrafter just got a huge speed (and quality!) boost! I’ll work to update my custom nodes over the next few days to support v1.0.1
HU Wenbo
@wbhu_cuhk
Nov 26, 2024
#DepthCrafter v1.0.1 is available now, with improved quality and speed! Issue of "over-saturated" depth estimation is greatly alleviated. The only modification is the inference configuration. Have fun 🥳🥳
00:00
1.4K
akatz
@akatz_ai
Oct 19, 2024
Replying to @Franker85654856
Not real time unfortunately, the whole video must be processed at once to get the temporal consistency. We can dream for the future though 😊
6.8K
akatz
@akatz_ai
Oct 20, 2024
Replying to @notdylaan
Good question, I’d love to put together more sample videos like that to test out the model and see where it chokes
4K
akatz
@akatz_ai
Oct 12, 2024
Last week I released 🌊 Depthflow Custom Nodes for ComfyUI! This is a wrapper around Tremeschin's original Depthflow code (link to repo's below). With Depthflow you can create awesome 3D animations like the one below from a SINGLE input image! It even works with video!
00:00
1.9K