![[ADVANCED WORKFLOW] Create CONTROLLABLE AI CHARACTERS for your MOVIES](https://img5.xaiju.com/storage/3/br/yk/d38796-019e8569-da44-79d7-91b2-4d4464ad80e7.jpg)
[ADVANCED WORKFLOW] Create CONTROLLABLE AI CHARACTERS for your MOVIES
Added 2025-07-17 13:14:19 +0000 UTCThank you so much for your support and making these videos possible!
Here is the advanced AI VFX workflow for adding controllable creatures & consistent characters to any footage using just one reference image and Wan 2.1. VACE in ComfyUI.
Note that this is just the standard configuration using the base Wan 14B model + 14B Vace module, but you can also use other models like the 3b versions. Use it as a foundation and do whatever you want with it, make it your own!
We will release a version for lower VRAM systems too. You can also check out the version by Rekes in the comments below, thank you so much for providing it. The current one will use 14GB of vram. This can be reduced by increasing the block swap. You can find more information in the Guide.
If you have issues, questions, or ideas please let me know on our discord channel!
π Check out the ADVANCED WORKFLOW GUIDE: https://mickmumpitz.ai/guides/create-controllable-characters-for-your-ai-movies-advanced
π Download the ADVANCED VERSION, EXAMPLE FILES & SCRIPTS for AE and blender below!
Have fun generating!
Comments
Hi, I've got a few questions! I've also been reverse engineering this workflow. Got a few questions. - First issue I've faced was slow motion generation - I tried to generate a stork that shakes it's head. Couldn't achieve that. Any suggestions? Tried 16fps and 25fps (that's what's needed), 720p, 480p - nothing. - Also, your workflow used causevid rank32 v1 - which with strength of 1 additionally adds sort of a white glow around a subject and could make the character too bright and too much contrast. On kijai github I was able to find v2 version which seemed to work better, but required more steps. - This might be helpfull information for those on youtube or elsewhere. - Also why you WanVideoWrappers and not default ones? how did you find this "extra model" selection workflow? I don't see it quite as a standard to bundle two models together in wanvideo model loader (VACE-module and Wan2_1-T2V-14B). Could you explain your reasoning behind creating such a workflow? because I see VAE models different from default ones as well - would be interesting why you diverged from the standard ones. - why use flowmatch_causvid scheduler? it's limited to 9 steps only? P.S. I can share a cleaned json workflow, which I've used to reverse engineer, if that helps.
Tadas Pranulis
2025-10-28 12:11:19 +0000 UTCAfter pressing run, it goes to green. How can I run the iteration (purple) parts? (250717_MICKMUMPITZ_CREATURE_PATH_ADV_v01)
TD
2025-08-25 21:08:38 +0000 UTC
